OPTIMIZING NEAR-INFRARED SPECTRAL TOMOGRAPHY FOR DIAGNOSTIC

IMAGING AND MONITORING OF BREAST CANCER TREATMENT

A Thesis

Submitted to the Faculty

in partial fulfillment of the requirements for the

degree of

Doctor of Philosophy

by

YAN ZHAO

Thayer School of Engineering

Dartmouth College

Hanover, New Hampshire

November 2017

Examining Committee:

Chairman_______________________

Shudong Jiang, PhD

Member________________________

Brian W. Pogue, PhD

Member________________________

Keith D. Paulsen, PhD

Member________________________

Bruce J. Tromberg, PhD

___________________

F. Jon Kull, Ph.D.

Dean of Graduate and Advanced Studies

ii

ii

Abstract

Near-infrared spectral tomography (NIRST) has been intensively investigated for clinical

application in breast imaging, by providing functional information about physiologically

related biomarkers such as oxy- and deoxy-hemoglobin, water, lipid and scatter

component. In this thesis, a series of studies on system development and reconstruction

algorithm were completed to improve the imaging quality of MR-guide NIRST and to

predicate breast cancer response to neo-adjuvant chemotherapy. To optimize image

recovery which maximizes difference between malignant and benign lesions, non-linear

iterative reconstructions of MR-Guided NISRT images were recovered using an L-curve

based algorithm, and applied to clinical trial data. The statistical analyses have shown

that the new approach dramatically improved the statistical significance for

differentiating malignant from benign lesions. While MRI guide NIRST has been utilized

to detect breast cancer, NIRST is also used to predicate and monitor breast tumor

responses in patients with locally advanced breast cancer undergoing neoadjuvant

treatment.

Based on an existing hybrid NIRST system developed at Dartmouth, a compact

and portable NIRST system has been developed for imaging patients in the infusion unit

while patients are awaiting or undergoing infusion. This system can acquire frequency-

domain and continuous-wave data simultaneously at 12 wavelengths in the wavelength

range of 660nm to 1064nm. Novel soft gel based homogenous and heterogamous tissue-

mimicking phantoms with sphere-shape inclusions have been developed, to mimic human

breasts. The phantom experiments indicate that the reconstructed optical images highly

depend on the position of imaging plane, especially in the case of small inclusions.

iii

Tomographic images of breast collagen content have been recovered for the first time,

and image reconstruction approaches with and without collagen content included have

been validated in simulation studies, which indicate that including collagen content into

the reconstruction procedure can significantly reduce the overestimation in total

hemoglobin, water and lipid, and underestimates in oxygen saturation. A group of 10

normal subjects were imaged, and significantly higher (p<0.05) total hemoglobin and

water were estimated in the high-density relative to low-density groups. The performance

of the NIRST system was validated in an ongoing clinical trial, and the recovered optical

biomarkers were correlated with pathologic response to neoadjuvant chemotherapy.

iv

Acknowledgements

I would like to thank all of the people who have made my time at Dartmouth such

a great and amazing journey. Without the individuals acknowledged here, I would not

have been able to finish the work presented in this thesis.

First of all, I would like to thank my advisor Prof. Shudong Jiang. She gave me

great support and encouragement over the last five years and played a key role in my

research. Shudong has been a dedicated mentor, taking time out of her busy schedule to

work with me in the lab, having numerous discussions on issues related to the imaging

system, and giving me step-by-step instructions. I am impressed by her enthusiasm in

managing clinical trials and abundance of knowledge. Her patience and valuable advice

helped me stay on the right track and become a qualified researcher. I could not have

found a better research advisor and mentor for my PhD.

My special thanks go to Prof. Brian Pogue, who is the leader of the Optics in

Medicine group. I was brought to the world of biomedical optics by Brian, and inspired

by his tireless work ethic, expertise and kindness. He was always willing to meet with

me, read my work and give me insightful advice.

I am grateful to Prof. Keith Paulsen for his constant support through my PhD

work. Keith has been leading several large research projects simultaneously, and he still

found time to give thorough edits of my papers. I also learned a lot from his constructive

suggestions on my research ideas.

I want to thank Prof. Bruce Tromberg for serving on my committee, and his

thoughtful suggestions on my thesis proposal.

v

Though not on this committee, Dr. Scott Davis and Dr. Steven Chad Kanick

deserve my sincere thanks. I had a great time working with them on the dosimetry project

in my first year. I would like to thank Dr. Junqing Xu for her help during my stay in

Xi’an. I learned a lot from the intensive discussions with Dr. Limin Zhang and Dr.

Jinchao Feng on the DRI project. I would like to thank Dr. Michael Mastanduno and Dr.

Fadi El-Ghussein for their guidance and help on the reconstruction program and imaging

system.

I appreciate the help of everyone from the entire Optics in Medicine lab. My

sincere thanks go to the following people: Dr. Kristian Sexton, Dr. Kelly Michaelsen, Dr.

Venkataramanan Krishnaswamy, Dr. Alisha Dsouza, Dr. Rongxiao Zhang, Dr. Robert

Holt, Mingwei Zhou and William Burger.

Finally, I would like to thank my friends and family for providing encouragement

during my PhD study. The solid support and endless love from my parents and girlfriend

Xuan, gives me all the motivation to move forward.

This work has been founded by NIH grant R01 CA069544 and R01 CA176086.

vi

Table of Contents

Abstract .............................................................................................................................. ii

Acknowledgements .......................................................................................................... iv

Table of Contents ............................................................................................................. vi

List of Tables .................................................................................................................... ix

List of Figures ................................................................................................................... xi

List of Acronyms .......................................................................................................... xxiii

Chapter 1: Introduction ....................................................................................................1

1.1. Project overview ...............................................................................................1 1.2. The current state of clinical breast cancer imaging .........................................1 1.3. NIRS/NIRST imaging of breast cancer ...........................................................4 1.4. NIRST imaging of breast cancer at Dartmouth ...............................................8 1.4.1. Breast cancer diagnosis ............................................................................9 1.4.2. Monitoring treatment response to NAC .................................................10 1.5. Organization of this thesis .............................................................................12

Chapter 2: Theory and Image Reconstruction Methods..............................................14

2.1. Introduction .....................................................................................................14 2.2. Modeling of photon propagation in highly scattering medium .....................17

2.2.1. Optical characteristics of biological tissues ..........................................17 2.2.2. Photon diffusion equation .....................................................................17 2.3. Numerical modeling of the forward problem .................................................19 2.4. Inverse problem solver ....................................................................................20 2.5. Spectral prior reconstruction ...........................................................................23 2.6. Spatial prior reconstruction .............................................................................25 2.6.1. Hard-prior reconstruction......................................................................25 2.6.2. Soft-prior reconstruction .......................................................................29 Chapter 3: Optimization of Image Reconstruction in MRI-guided NIRST for Breast

Cancer Diagnosis .............................................................................................................31

3.1. Introduction .....................................................................................................31 3.2. Optimization of regularization parameter in MRI-guided NIRST for breast cancer diagnosis .....................................................................................................33

3.2.1. Reconstruction and visualization of optical images in MRI-guided NIRST .............................................................................................................33 3.2.2. Fixed regularization parameter of 0.1 and 1 using only amplitude data ........................................................................................................................35 3.2.3. Optimal regularization using only amplitude data................................36 3.2.4. Fixed regularization parameter of 0.1 and 1 using both amplitude and phase data ........................................................................................................38 3.2.5. Optimal regularization parameter using both amplitude and phase data ........................................................................................................................39

vii

3.2.6. Optimal regularization leads to better separation between malignant and benign lesions ...........................................................................................41 3.2.7. Discussions ...........................................................................................43

3.3. Direct regularization from co-registered anatomical images for MRI guided NIRST image reconstruction .................................................................................47

3.4. Discussions ....................................................................................................51 Chapter 4: A Hybrid Frequency-Domain/Continuous-Wave NIRST System with

Simultaneous Measurements at Twelve Wavelengths .................................................53

4.1. Introduction .....................................................................................................53 4.2. Imaging system and patient exam settings ......................................................56

4.3. Laser source sub-systems ................................................................................58 4.3.1. 6-wavelength FD source module .........................................................58 4.3.2. 6-wavelength CW source module ........................................................60 4.4. Hybrid PMT-PD detector sub-system .............................................................61 4.4.1. Detector array of 15 PMT and PD detectors ........................................61 4.4.2. Calibration of PMT/PD detectors ........................................................62 4.5. Adjustable parallel breast interface .................................................................64 4.5.1. Classical fiber-breast interfaces ...........................................................64 4.5.2. Design of adjustable parallel breast interface ......................................66 4.5.3. Phantom imaging with the parallel breast interface .............................68 4.6. Simultaneous acquisition at twelve FD+CW wavelengths .............................70 4.6.1. Simultaneous acquisition at twelve wavelengths .................................70 4.6.2. Hybrid gain setting of PMT detector ...................................................72 4.6.3. Data acquisition GUI ...........................................................................75 4.7. Systematic characterization of the system ......................................................75 4.7.1. Comparison between sequential and simultaneous acquisitions .........75 4.7.2. Variation of phase and amplitude data.................................................77 Chapter 5: Tissue Simulating Phantoms for NIRST Imaging ....................................79

5.1. Introduction .....................................................................................................79 5.2. Comparison between major tissue mimicking phantoms ..............................80

5.3. Phantom Preparation .......................................................................................83 5.3.1. Preparation of homogenous silicone soft gel phantom .......................83 5.3.2. Preparation of heterogeneous silicone soft gel phantom ....................84 5.4. Characterization of homogenous silicone soft gel phantom ...........................86

5.5. Validating the performance of the NIRST system using heterogeneous breast mimicking phantoms ..............................................................................................91

5.5.1. NIRST imaging of heterogeneous phantoms at different depths .......91 5.5.2. NIRST imaging using partial transmission/reflectance data ..............94

Chapter 6: In vivo Collagen Quantification in Breast Tissue .....................................99

6.1. Introduction .....................................................................................................99 6.2. Simulation .....................................................................................................100

6.2.1. Homogeneous phantom simulation..................................................100 6.2.2. Heterogeneous phantom simulation .................................................102 6.3. In vivo collagen quantification in breast tissue .............................................105 6.3.1. In vivo collagen quantification in normal subjects ..........................105

viii

6.3.2. In vivo collagen quantification in cancer patients ..........................107 6.4. Discussions ...................................................................................................108 Chapter 7: Imaging Normal Subjects ..........................................................................109

7.1. Introduction ...................................................................................................109 7.2. Imaging setup ................................................................................................110

7.3. Imaging normal subjects with various breast sizes .......................................111 7.4. Continuous imaging of normal subjects .......................................................112 7.5. Intra-subject and inter-subject variations ......................................................113 7.6. Discussions ...................................................................................................116 Chapter 8: Towards monitoring breast cancer response to neoadjuvant

chemotherapy ................................................................................................................118

8.1. Introduction ...................................................................................................118 8.2. Optimal workflow of NIRST breast imaging ..............................................120

8.3. Case studies of imaging breast cancer patients .............................................122 8.3.1. Imaging breast cancer patients .........................................................122 8.3.2. Monitoring breast response to NAC ................................................126 8.4. Discussions ...................................................................................................129 Chapter 9: Conclusions and Future Directions...........................................................132

9.1. Completed work ...........................................................................................132 9.1.1. Optimization of MRI-guided NIRST image reconstruction ............132 9.1.2. A hybrid FD/CW system with simultaneous measurements at twelve wavelengths ................................................................................................133 9.1.3. Silicone soft-gel based tissue mimicking phantoms for NIRST imaging .......................................................................................................133 9.1.4. Collagen quantification using the NIRST system ............................134 9.1.5. Imaging normal subjects ..................................................................134 9.1.6. Imaging breast cancer patients .........................................................135

9.2. Future Directions ..........................................................................................135 9.2.1. Optimization of NIRST system for monitoring patient response to NAC ..........................................................................................................135

9.2.2. Optimization of the sampling geometry in MRI-guided NIRST .....136 9.2.3. Imaging small tumors using MRI-guided NIRST ...........................143 Appendices ......................................................................................................................147

Appendix A: LabVIEW Acquisition Program ............................................................147 Appendix B: Matlab Code ..........................................................................................149 Appendix C: Itemized Components List .....................................................................150

References .......................................................................................................................151

ix

List of Tables

Table 3.1: Comparison of statistics using fixed regularization of 0.1 and 1, and optimal

regularization. Only amplitude data (AMPL) was used for image reconstruction. ...........38

Table 3.2: Comparison of statistics using fixed regularization of 0.1 and 1, optimal

regularization, and fixed regularization of 0.1 for amplitude and 2 for phase. Both

amplitude and phase data (AMPL/PH) were used for image reconstruction.....................40

Table 3.3: Comparison of the three regularization approaches in all clinical exams,

relative to selected exams when the optimal regularization parameter was found. The

group of all exams included 16 malignant and 9 benign pathology-confirmed diagnoses

whereas the selected exams included 15 malignant and 7 benign cases (3 exams from the

former did not have optimal regularization at the 1st iteration). Both amplitude and phase

data were used during image reconstruction. .....................................................................45

Table 5.1: Comparison of major tissue-mimicking optical phantoms. ..............................80

Table 5.2: Maximum recovered (10-3/mm) at different depths for phantoms with sphere

shape inclusions with diameter of 12mm, 18mm and 24mm. ...........................................93

Table 5.3: Comparison of the reconstructed inclusion/background contrast using different

subsets of measurement data acquired at different depths. Four subsets of measurement

data are compared: (A) full dataset, i.e., both transmission and reflectance; (B) two sides

of reflectance data; (C) upper side of reflectance data and (D) only transmission data. ...97

Table 5.4: Comparison of the sensitivity of the inclusion (%) using different subsets of

measurement data acquired at different depths. Four subsets of measurement data are

compared: (A) full dataset, i.e., both transmission and reflectance; (B) two sides of

reflectance data; (C) upper side of reflectance data and (D) only transmission data. .......98

x

Table 7.1: Mean and standard deviation of optical parameters of both sides of the breast.

..........................................................................................................................................114

Table 7.2: Mean, standard deviation and total range of physiological and optical

parameters of 10 normal subjects. ..................................................................................115

Table 8.1: Reconstructed optical contrast in terms of HbT, StO2, water, lipid, SA and SP

for three visits of before treatment, on day 9 of cycle 1, and after treatment, respectively.

..........................................................................................................................................127

Table 8.2: Reconstructed optical contrast in terms of HbT, StO2, water, lipid, SA and SP

for three visits of before treatment, on day 19 of cycle 1, and after therapy, respectively.

..........................................................................................................................................129

Table 9.1: Initial guess of SA and SP for four categories grouped by MRI-identified

breast density. ..................................................................................................................140

xi

List of Figures

Figure 1.1: Three common types of sources used in diffuse optical imaging. The far-left

figures show the “banana patterns” of light sampling path for transmission and

reflectance geometries. The detected light intensity over time is illustrated for continuous

wave (CW), frequency-domain (FD), and time-domain (TD) measurement in (a), (b) and

(c), respectively [1]. ............................................................................................................5

Figure 1.2: Six NIRS/NIRST systems for breast imaging. (a) The DSP based CW NIRST

system developed at Columbia University. (b) The CW NIRST system developed by

Philips Healthcare. (c) The stand-alone clinical NIRST system developed at University of

Pennsylvania, utilizing a CCD-based heterodyne detection scheme. (d) The DOSI system

developed at University of California, Irvine. (e) Combined US and NIRS systems and a

handheld probe developed at University of Connecticut. (f) The MRI-guided NIRST

system developed at Dartmouth College. ...........................................................................8

Figure 1.3: Images from a patient with a 11x21x14mm biopsy-confirmed Invasive Ductal

Carcinoma (IDC) in her right breast. (a) non-contrast T1 MRI with tumor location

indicated (arrow); (b) Reconstructed images for HbT, (c) StO2, (d) water, (e) lipid, (f)

scattering amplitude, and (g) scattering power is overlaid on the MR scan. The value of

each parameter in the adipose region is suppressed for clarity of visualization. ..............10

Figure 1.4: (a) Reconstructed optical images of a pCR case prior, during and post NAC.

(b) Axial post contrast subtraction MRI before NAC shows and enhancing mass indicated

by arrow [2]. ......................................................................................................................11

Figure 2.1: FEM meshes with nodes and elements created in NIRFAST for three

geometries: (a) 2D circle, (b) 2D football-shape and (c) 3D actual breast. ......................20

xii

Figure 2.2: Image segmentation in NIRFAST-Slicer. ......................................................25

Figure 2.3: (a) Flowchart outlining the sequence for the optimization algorithm. In (b),

the L2 norm of the prior property error vs. L2 norm of the model error creates the L-

curve of values for each regularization parameter value from 0.001 to 100. The

optimization algorithm is implemented at each iteration, and optimal regularization

parameter from previous iteration is used once the L metric falls behind the threshold. .28

Figure 3.1: Images from a 33-year-old patient with a 11x21x14mm biopsy-confirmed

Invasive Ductal Carcinoma (IDC) in her right breast. (a) non-contrast T1 MRI with tumor

location is indicated (arrow); Reconstructed images of (b) HbT, (c) StO2, (d) water, (e)

lipid, (f) scattering amplitude, and (g) scattering power are overlaid on the MR scan. The

value of each parameter in the adipose region is suppressed for clarity of visualization. 34

Figure 3.2: Tumor-to-adipose contrast in HbT vs. regularization for (a) benign, and (b)

malignant cases. Circles have a regularization of 0.1, and asterisks have a regularization

of 1. Box plots of contrast for the two pathologies are shown with fixed regularizations of

(c) 0.1 and (d) 1 for all patients. ........................................................................................35

Figure 3.3: L-curves are shown for the first 3 iterations of a single case with a malignant

tumor. The regularization was varied from 0.001 to 100 at each iteration, and the optimal

regularization was (a) 0.18, (b) 0.22, and (c) undefined for the three iterations,

respectively. Histograms of optimal regularization parameter for the 1st and 2nd iteration

are inserted to Figs. 3-3(a) and (b), respectively. .............................................................36

Figure 3.4: (a) Log scale of projection error vs. number of iterations. (b) Tumor/adipose

contrast in HbT vs. number of iterations. .........................................................................37

xiii

Figure 3.5: Tumor-to-adipose contrast of HbT vs. regularization for (a) benign conditions

and (b) malignant tumors. Circles have a regularization of 0.1, and asterisks have a

regularization of 1. Box plots of contrast are shown for regularization of (c) 0.1, and (d) 1

for all patients. Both amplitude and phase data were used in the image reconstructions. 38

Figure 3.6: ROC curve for fixed regularization of 1 (black) and 0.1 (blue), and optimal

regularization (red), when HbT and TOI are combined. Both amplitude and phase data

were used for image reconstruction. .................................................................................41

Figure 3.7: Box plots of the contrast for (a) total hemoglobin (HbT), (b) oxygen

saturation (StO2), (c) Tissue optical index (TOI), (d) scattering power (SA), and (e)

scattering power (SP), as recovered using the optimal regularization and amplitude and

phase data. aSignificant difference. ...................................................................................42

Figure 3.8. MR images from a patient with a malignant lesion (20mm 27mm 33mm)

seen on DCE MRI. (a): Screenshot of the Nirview 3D surface rendering of the T1 MRI.

Fiducial markers and fiber bundle positions are shown; (b): Standard T1 image; and (c):

Dynamic contrast-enhanced MRI. ....................................................................................49

Figure 3.9: The reconstructed HbT images overlaid in three planes with x=-100.0, y=-

19.8 and z=-26.6 respectively. (a) Segmented images from corresponding T1 and DCE

images. Optical images reconstructed by hard-prior reconstruction using L-curve based

optimization of regularization parameter (b), and DRI with λ=1 and σg=0.001 (c),

respectively. ......................................................................................................................50

Figure 4.1: The 12-wavelength FD-CW NIRST system. (a) A photo of the NIRST

system. (b) and (c) Recline chair with two groups of fibers in the imaging suite. (d)

xiv

Adjustable fiber-breast interface. (e) Subject being imaged with the system. (f) Surface

image of breast-interface. ..................................................................................................56

Figure 4.2: 6-wavelength FD source module. (a) System diagram; (b) Photo of the sub-

system with major components labeled. (c) Photo of the customized 6-wavlength source

module and multi-channel synthesizer. Raw data acquired from one PMT detector after

heterodyned with reference signal, for f1=100.0004MHz (d), f2=100.0007MHz (e),

f3=100.0011MHz, and mixed signal including frequency components at all three

frequencies (g). .................................................................................................................58

Figure 4.3: 6-wavelength CW source module. (a) System diagram. (b) Photo of the sub-

system with major components labeled. Raw signal acquired from a PD detector with

source light modulated at 30Hz (c), 50Hz (d), 80Hz (e), and mixed signal including all

three modulated laser sources (f). .....................................................................................60

Figure 4.4: Photos of the hybrid detector sub-system. The actual assembly of bottom

plate and hybrid detector array are shown in (a) and (b), respectively [3]. ......................61

Figure 4.5: PMT calibration. Input power (log10) (a) and phase (degree) (b) versus PMT

AC amplitude for different gain settings from 0.5 to 1.1. .................................................63

Figure 4.6: Uncalibrated and calibrated amplitude/phase data. Uncalibrated amplitude

(a1) and phase (a2) versus source-detector distance. Calibrated amplitude (b1) and phase

(b2) versus source-distance distance. ................................................................................64

Figure 4.7: Three typical fiber-breast interfaces: (a) circular interface, (b) dual-plate

interface and (c) triangular interface. ................................................................................65

xv

Figure 4.8: Adjustable parallel fiber-breast interface. (a) Solidworks file showing

dimensions of the interface. (b) A soft gelatin breast phantom being imaged with the

interface. (c) Corresponding FEM mesh with fibers marked in red circles. .....................68

Figure 4.9: Experimental setup and reconstructed optical images for two heterogeneous

phantoms with 1-inch diameter inclusions. The corresponding interface had deep

curvature (a) and flat curvature (b). For both phantoms, the blood concentrations inside

and outside the inclusion were 1.5% and 1%, respectively. .............................................69

Figure 4.10: System diagram for simultaneous acquisition. FD source module, CW

source module, and data acquisition/processing module are highlighted in blue, green,

and violet blocks, respectively. The flow of low frequency electrical signal, high

frequency electrical signal, and light is shown by the black, blue and red solid lines,

respectively. ......................................................................................................................70

Figure 4.11: Hybrid gain adjustment of PMT detectors. (a) Flow chart illustrating the

hybrid gain adjustment scheme. (b) A photo of the adjustable fiber-breast interface (c)

Corresponding football shape mesh created with 16 fibers assigned along the surface. (d)

Amplitude data acquired at source position #1 using automatic gain adjustment scheme.

(e) Amplitude predicted for the other source-detector pairs, based on the parameters fitted

from (d). The actual amplitude and phase data acquired using the gain from the lookup

table for the rest of source-detector pairs, shown in (f) and (g) respectively. ..................72

Figure 4.12: Standard deviation of amplitude (a) and phase (b) of 30 measurements for

two gain adjustment schemes. ...........................................................................................74

Figure 4.13: LabVIEW GUI for data acquisition, pre-processing and display. ...............75

xvi

Figure 4.14: Reconstructed optical images for the same heterogeneous phantom with a 1-

inch diameter inclusion. The optical images were reconstructed using boundary data

acquired from sequential measurement (a), and simultaneous measurement (b),

respectively. The blood concentrations inside and outside the inclusion were 2% and 1%,

respectively. ......................................................................................................................77

Figure 4.15: Standard deviation of phase (a) and AC amplitude (b) versus AC amplitude

for different gain settings from 0.7 to 1.1. ........................................................................77

Figure 5.1: Major tissue mimicking phantoms developed at Dartmouth. From left to right,

gelatin phantom with ink (a), gelatin phantom with blood (b), resin phantom (c), RTV

silicone phantom (d) and silicone soft gel phantom (e) are presented, respectively. .......81

Figure 5.2: The preparation of silicone soft gel phantom. (a) Base materials of A-341:

Silicone Soft Gel. (b) Silicone coloring materials which are used as absorber/scatter. (c)

The mixing of base and silicone coloring materials. ........................................................83

Figure 5.3: Detailed steps of making breast mimicking phantoms. (a) The base material

was mixed with coloring materials in a food mixer. (b) The mixed solution was poured

into 3D printed molds. (c) Three sphere-shape inclusions were taken from the molds after

curing, with radius of 6mm, 9mm and 12mm, respectively. (d) One sphere-shape

inclusion was fixed inside a large mold, which was filled with mixed solution later. The

optical properties of the inclusion are different from those of the background, in order to

create inclusion/background contrast. (e) A group of heterogeneous breast mimicking

phantoms, with either sphere shape inclusion inside, or cylindrical cavity. .....................86

Figure 5.4: DOSI measurement of a silicone soft gel phantom. (a) Measured (blue points)

and fitted (red line) amplitude and phase at four wavelengths, while the laser modulation

xvii

frequency was scanned from 50 to 400MHz. (b) Measured at four wavelengths (red

points) and fitted scattering spectrum (blue line) using Mie theory. (c) Fitted broadband

absorption spectrum. .........................................................................................................87

Figure 5.5: The fitted a (a) and s (b) at 661nm are plotted with error bars representing

the standard deviation among different measurements, for five phantoms. Five silicone

soft gel phantoms were made using the same recipe, and each phantom was measured 10

times at randomly selected positions on the phantom, using the DOSI system. ..............88

Figure 5.6: Measured a (a) and s (b) at 661nm of one phantom in 10 days. Each time

the same silicone soft gel phantom was measured 10 times at randomly selected positions

on the phantom, using the DOSI system. ..........................................................................88

Figure 5.7: Measured a (a)) and s (b) are plotted versus pink paint concentration for

the group of 0.3ml (blue points) and 0.8ml (black points) white paint, respectively. Two

groups of seven phantoms were made, using 50g of base A and 5g of catalyst B as base

material for each phantom. 0.3ml and 0.8ml of white paint were added into each of the

seven phantoms in the 1st and 2nd group, respectively. An increasing amount of pink

paint, from 0.1ml to 0.7ml with an increment of 0.1ml, was added into corresponding

phantom in each group. Each phantom was measured 5 times. ........................................90

Figure 5.8: The measured a (Fig. 5-8(a)) and s (Fig. 5-8(b)) are plotted versus white

paint concentration for the group of 0.3ml (blue points) and 0.5ml (black points) pink

paint, respectively. Two groups of seven phantoms were made, using 50g of base A and

5g of catalyst B as base material for each phantom. 0.3ml and 0.5ml of pink paint were

added into each of the seven phantoms in the 1st and 2nd group, respectively. An

increasing amount of white paint, from 0.5ml to 1.1 ml with an increment of 0.1ml, was

xviii

added into corresponding phantom in each group. Each phantom was measured 5 times.

............................................................................................................................................91

Figure 5.9: Phantom experiments using a silicone soft gel phantom with a sphere

inclusion. (a) Reconstructed absorption images from the data acquired at different depths

of 0mm, -3mm, -6mm, -9mm, and -12mm, respectively. The depth of 0mm corresponds

to the case where imaging plane was placed across the center of the sphere inclusion. (b)

Profiles of the reconstructed a along the X-axis, crossing the center of the inclusion

projected on the surface, at 830nm. (c) The sphere has a diameter of 24mm. The actual

inclusion/background contrast is 2. ...................................................................................92

Figure 5.10: Phantom experiments using a silicone soft gel phantom with a sphere

inclusion, which has a diameter of 24mm. Reconstructed absorption images from the data

acquired at different depths of -9mm, -6mm, -3mm, 0mm, 3mm, 6mm, 9mm, and 12mm

respectively, using (a) both reflection and transmission data; (b) both sides of reflectance

data; (c) one side (upper side) of reflectance data; and (d) two sides of transmission data.

The plane at 0mm corresponds to the case where imaging plane was placed across the

center of the sphere inclusion. The actual inclusion/background contrast is 2.5. .............95

Figure 6.1: Recovered HbT (a), StO2 (b), water (c) and lipid (d) in a simulated

homogeneous phantom, with collagen content increased from 0 to 10%. .......................101

Figure 6.2: Reconstructed images of a simulated heterogeneous phantom. (a) Images with

true values. The diameter of the circular inclusion is 20 mm. An inclusion/background

contrast of 2 is assigned to HbT, with homogeneous background value of 75%, 45%,

45%, 10%, 0.8 and 0.3 assigned for StO2, water, lipids, collagen, SA and SP,

xix

respectively. Reconstructed images without collagen (b) and with collagen included (c).

..........................................................................................................................................103

Figure 6.3: Image reconstruction in the presence of collagen, which was not included in

the reconstruction. A similar simulation setup was used as Figure 2, except that the

contrast was assigned in HbT (a), StO2 (b), water (c), and lipid (d), respectively. The

extracted inclusion/background contrast was plotted versus collagen concentration

accordingly. .....................................................................................................................104

Figure 6.4: Contents of breast tissue recovered for HbT (a), StO2 (b), Water (c) and

Lipids (d), with and without collagen included in reconstruction. The radiographic

density type of subject #1 and #2 is heterogeneously dense (HD) and scattered

fibroglandular dense (Scattered), respectively. ...............................................................105

Figure 6.5: MRI T2 images of a patient with invasive cancer in the left breast: (a) Axial

view, (b) sagittal view and (c) coronal view. Reconstructed optical images without (d)

and with (e) collagen included. Recovered optical images are displayed in the same

orientation in (d) and (e) as in (c). ..................................................................................107

Figure 7.1: The setup of human subject imaging. (a) The NIRST system placed outside

the exam/infusion room. (b) Exam room. (c) A female subject being imaged on the left

breast. ..............................................................................................................................110

Figure 7.2: Reconstructed optical images of three normal subjects. Maximum separation

between the two fiber holders in the interface was 63mm (a), 85mm (b), and 40mm (c),

respectively. ....................................................................................................................112

Figure 7.3: Continuous measurements of (a) HbT; (b) StO2; (c) water; (d) lipid; (e) SA

and (f) SP for two normal subjects. ................................................................................113

xx

Figure 7.4: Comparison between radiographic dense and non-dense groups in terms of

HbT (a), StO2 (b), Water (c), Lipid (d), SA (e) and SP (e). ...........................................116

Figure 8.1: Optimal workflow for NIRST patient imaging. ...........................................120

Figure 8.2: Case study #1. Dynamic Contrast Enhanced MR Images (DCE-MRI) of a

patient with invasive cancer in the right breast: (a) Axial view, (b) sagittal view and (c)

coronal view. (d) Recovered optical images of HbT, StO2, water, lipid, SA, and SP for

right breast. .....................................................................................................................122

Figure 8.3: Case study #2. DCE-MRI of a patient with invasive cancer in the right breast:

(a) Axial view, (b) sagittal view and (c) coronal view. (d) Recovered optical images of

HbT, StO2, water, lipid, SA, and SP for right breast. .....................................................123

Figure 8.4: Case study #3. DCE-MRI images of a patient with invasive cancer in the left

breast: (a) Axial view, (b) sagittal view and (c) coronal view. Recovered optical images

of HbT, StO2, water, lipid, SA, and SP for right breast (d). ...........................................124

Figure 8.5: Case study #4. DCE-MRI images of a patient with invasive cancer in the left

breast: (a) Axial view, (b) sagittal view and (c) coronal view. Recovered optical images

of HbT, StO2, water, lipid, SA, and SP for right breast (d). ...........................................125

Figure 8.6: Case study #5. Clinical images of a 63-year-old patient with pathological

confirmed pIR. Postcontrast T2-weighted MRI images prior to initiation: (a) axial view,

(b) sagittal view and (c) coronal view. Pathological findings showed pIR to neoadjuvant

chemotherapy. Reconstructed optical images of HbT (uM), StO2 (%), water (%), lipid

(%), scattering amplitude (SA) and scattering power (SP) before treatment (d), on day 9

of cycle 1, and after therapy (29 days prior to surgery) are shown for abnormal breast.

..........................................................................................................................................126

xxi

Figure 8.7: Case study #6. Clinical images of a 54-year-old patient with radiologic

findings and pIR. Postcontrast T2-weighted MRI images prior to initiation: (a) axial

view, (b) sagittal view and (c) coronal view. Pathological findings showed pIR to

neoadjuvant chemotherapy. Reconstructed optical images of HbT (uM), StO2 (%), water

(%), lipid (%), scattering amplitude (SA) and scattering power (SP) before treatment (d),

on day 19 of cycle 1, and after therapy (24 days prior to surgery) are shown for abnormal

breast. ..............................................................................................................................128

Figure 9.1: Flowchart outlining the sequence for two reconstruction methods. .............137

Figure 9.2: Triangular interface with different sampling geometries. Three strategies of

choosing phase measurements, (a) with full transmission across many sources and

detectors, (b) with just partial reflectance data, and (c) with just partial transmittance data.

Fiber locations are shown as blue dots. ..........................................................................138

Figure 9.3: (a) Illustration of triangular patient interface for optical fiber placement. (b)

Relative difference versus detector number. ...................................................................139

Figure 9.4: (a) Relative difference of optical contrast versus number of FD detectors used

for homogeneous fitting of initial guess. ROC curves with 4 detectors (b), 6 detectors (c)

and 15 detectors (d) for estimating initial guess. ............................................................140

Figure 9.5: ROC curves of AMPL reconstruction (a), Lookup table I (b), and lookup table

II (c) for HbT, and AMPL reconstruction (d), lookup table I (e), and lookup table II (f)

for TOI. ...........................................................................................................................142

Figure 9.6: Equivalent tumor diameter of 12.2mm (a), 37.7mm(b) and 46.8mm (c). Three

regions of tumor, fibroglandular and adipose are represented in white (region 3), yellow

(region 2) and red (region 1), respectively. .....................................................................143

xxii

Figure 9.7: Recovered HbT (black asteroid) of tumor and adipose (black circle) are

plotted versus equivalent tumor contrast, for patient #51 (a) and #30 (b), respectively.

The corresponding tumor sensitivity is represented by red square. ................................144

Figure 9.8: Recovered tumor/adipose contrast versus actual tumor/adipose contrast for

the tumor with equivalent tumor diameter of 8mm, 10mm, 12mm, 20mm and 40mm,

respectively. A fixed regularization parameter of 1 was used in the image reconstruction.

..........................................................................................................................................145

Figure 9.9: Comparison between FD/CW reconstruction (a) and CW reconstruction (b),

with increasing equivalent tumor diameter. The recovered HbT of tumor and adipose are

represented by blue asteroid and red square, respectively. .............................................146

xxiii

List of Acronyms

Continuous Wave CW

Dynamic Contrast Enhanced Magnetic Resonance Imaging DCE-MRI

Diffuse Optical Tomography DOT

Estrogen Receptor ER

Frequency Domain FD

Finite Element Mesh FEM

Deoxygenated hemoglobin Hb

Oxygenated hemoglobin HbO2

Total Hemoglobin HbT

Human Epidermal Growth Factor Receptor 2 HER-2

Invasive Ductal Carcinoma IDC

Locally Advanced Breast Cancer LABC

Magnetic Resonance Imaging MRI

Near Infrared Spectral Tomography NIRST

Near Infrared Spectroscopy NIRS

Neoadjuvant Chemotherapy NAC

Pathologic Complete Response pCR

Pathologic Incomplete Response pIR

Photomultiplier Tube PMT

Photodiode PD

Progesterone Receptor PR

Region of Interest ROI

xxiv

Radiative Transport Equation RTE

Scatter Amplitude SA

Scatter Power SP

Time domain TD

Thayer Formatting Guidelines TFG

Tissue Optical Index TOI

Triple-Negative Breast Cancer TNBC

1

Chapter 1: Introduction

1.1 Project overview

The primary goal of this thesis is to improve the performance of near infrared

spectral tomography (NIRST) for diagnostic imaging and treatment monitoring of breast

cancer, from the perspective of development of both the imaging system design and the

nonlinear image reconstruction methods. Breast cancer is the 2nd most commonly

diagnosed cancer among women in the United States, which presents a constant

challenge for both early diagnosis, management and treatment. In 2017, it is estimated

that 252,710 new cases of invasive breast cancer will be diagnosed in women in the U.S.,

together with 63,410 new cases of non-invasive (in situ) breast cancer. Studies clearly

show that the mortality rate can be reduced by early detection and appropriate treatment

prior to tumor metastasis [4]. Accurate imaging can play a critical role in both diagnosis

and clinical management of breast cancer. This chapter briefly introduces the current

state of clinical breast cancer imaging and the role of optical NIRST imaging.

1.2 The current state of clinical breast cancer imaging

The typical clinical progression of breast cancer includes screening

mammography as the first step in most western countries including the U.S. Patients with

a suspicious screening mammogram will be asked to have callback mammography,

ultrasound, and/or MRI as the second step. Next a biopsy sample would be sent to

pathology and the patients who have been diagnosed with cancer would either be enrolled

in a neoadjuvant chemotherapy or go straight to surgery for mastectomy or lumpectomy.

X-ray mammography has been used as the standard of care for annual breast

cancer imaging screening, which is important since some women with breast cancer do

2

not have symptoms. It is recommended by the American Cancer Society that all women

should begin having yearly mammograms by age 45. Mammography has an overall high

specificity of 97%, but with low sensitivity of 77%, as reported in a randomized multi-

center clinical trial [5]. The performance of X-ray mammography can vary on various

factors such as age, hormonal status, and breast density. Mammographic sensitivity for

breast cancer decreases significantly for women with higher breast density, which are

related to higher increased breast cancer risk [6]. In addition, the ionizing radiation from

X-ray mammography could be another concern for young age women.

Ultrasound (US) imaging is commonly used to evaluate specific abnormalities

discovered either at clinical examination or on mammography [7]. It has different

contrast mechanisms from mammography, which makes it possible to detect small, node-

negative lesions not seen on mammography, especially in the case of imaging dense

breast tissue. It has been used mainly for differentiating cystic from solid lesion, and

imaging young and pregnant women with palpable masses. An example of combining

different contrast mechanisms was shown by Berg et al, where the combination of

ultrasound and mammography had higher sensitivity than ultrasound alone [8]. The real-

time capability of US also makes it an ideal candidate for guiding breast biopsies and

other interventional procedures.

Magnetic Resonance Imaging (MRI) is a rapidly developing imaging technique in

breast imaging, with high spatial and temporal resolutions. The standard breast MRI

protocol includes T1 gradient-echo sequences, T2 sequences, and injected dynamic 3D

sequences. The high spatial resolution makes it possible for morphology-based analysis

of the lesion [9], while the high temporal resolution has been used to provide functional

3

information from the dynamic enhancement curve of the lesions [10]. Several studies

have shown that women at high risk of developing breast cancer will benefit from the

combination of MRI and mammography screening [11-13]. Dynamic contrast-enhanced

MRI (DCE-MRI) has been used in staging local tumor, where the enhancement comes

from various factors such as the amount and relaxivity of contrast agent, T1 contrast of

the pulse sequence, and baseline T1 relaxation time of different tissues [14]. DCE-MRI

has a reported sensitivity of nearly 100% in detecting invasive cancer [15]. Breast cancer

detection and differential diagnosis using MR imaging mainly comes from the angiogenic

activity of cancers. However, angiogenic activity is not only found in malignant tumors,

but also some other conditions. The enhancement in such conditions such as

inflammatory changes will result in a high false-positive rate with relatively low

specificity and lead to unnecessary biopsies.

Besides the indispensable role in detection and diagnosis of breast cancer,

imaging techniques have been widely used in management of clinical treatments. The

heterogeneity of breast cancer is subdivided into three subtypes of breast cancer, ER-

positive/HER2-negative (ER-positive), HER2-positive, and triple-negative. Different

subtypes correspond to various expected outcomes and treatment options [16]. Therefore,

it is of great interest to individualize patient treatment plans at an early stage of treatment

or even before treatment starts. Neoadjuvant chemotherapy (NAC) is used to treat

patients with locally advanced cancers, which account for 6-10% of new cases of breast

cancer [17, 18]. Pathologic response to NAC has been regarded to be the best predictor of

long-term outcome [19]. Conventional imaging methods including ultrasonography and

mammography were found only moderately useful for monitoring neoadjuvant

4

chemotherapy [20]. DCE-MRI and Fluorine 18 fluorodeoxyglucose positron emission

tomography (FDG-PET) have been successfully used by several groups to quantify

changes of breast tumors during treatment [21-23]. In a clinical trial involving 28 patients

reported by Ah-See [24], significant correlation was found between DCE-MRI kinetic

features and final clinical and pathologic response (p<0.01). Berriolo-Riedinger [25]

showed that the relative decrease in FDG uptake after the first course of NAC was a

significant indicator (p < 0.000066) in predicting patient response. However, both MRI

and PET imaging require the injection of contrast agents and the cost of the procedures

could be prohibitive.

1.3 NIRS/NIRST imaging of breast cancer

The earliest utilization of light in breast imaging can date back to 1929, when

Cutler first used continuous illumination of the breast to produce shadow images, which

was later called “diaphanography” [26] and briefly commercialized in the 1980s. In

recent decades, the development of breast optical imaging modalities has focused on the

investigation of spectroscopy/imaging using near infrared light. Near infrared

spectroscopy (NIRS) imaging is an emerging functional technique, which estimates the

intrinsic biophysical composition of tissue, in terms of the concentrations of total

hemoglobin and oxy-hemoglobin, water and lipids [27-30]. In addition, the ultra-

structural cellular density and size ensemble associated with the extracellular matrix and

subcellular constituents of breast tissue can be interrogated from the NIRS scattering

spectrum [31, 32]. NIRS shows potential advantages over other imaging candidates

because of its noninvasive nature and relatively low cost and portable size, which makes

possible repeatable imaging procedures under various patient conditions. Near infrared

5

spectral tomography (NIRST) is an important subtype of NIRS, which reconstructs

tomographic 2D/3D images of major absorbers and scattering coefficients. Three are

three major source modulation/acquisition techniques used by NIRS/NIRST in breast

imaging: continuous wave (CW), time domain (TD), and frequency domain (FD). Figure

1.1 illustrates the detected light intensity over time for CW (a), FD (b) and TD (c),

respectively.

Figure 1.1. Three common types of sources used in diffuse optical imaging. The far-left figures show the “banana patterns” of light sampling path for transmission and reflectance geometries. The detected light intensity over time is illustrated for continuous wave (CW), frequency-domain (FD), and time-domain (TD) measurement in (a), (b) and (c), respectively [1].

CW imaging systems measure only the changes in amplitude after light

transmitted through the tissue. The source light either has constant intensity or is

modulated at low frequency to increase signal to noise ratio. Detectors without radio

frequency response can also be used to acquire transmittance light, which significantly

simplifies the instrumentation design and overall cost. As shown in Fig. 1.2(a), a DSP

based CW NIRST system was developed at Columbia University, with a large number of

imaging data acquired through 32 sources and 64 detectors per breast with four

wavelengths, at a frame rate of 1.7 Hz [33]. As shown in Fig. 1.2(b), another CW NIRST

system was developed by Philips Healthcare, with a total of 507 optical fibers mounted

on the surface of a scanning cup, operating at four discrete wavelengths [34]. A total of

6

253 source fibers and 254 detector fibers were used to deliver source light to the breast

tissue and collect transmittance light, respectively. CW NIRST systems have been used

by a number of research groups in breast cancer diagnosis and treatment monitoring,

which shows promising results [35, 36]. However, the effect of absorption and scattering

in CW imaging cannot be well separated, and prior information/guess of scattering

coefficient was required to initialize the image reconstruction procedure [37].

TD imaging systems usually utilize picosecond pulsed diode lasers in the NIR

range and compact detector with extended spectral sensitivity. The photon distribution of

time-of-flight or temporal point spread function, TPSF, was measured after a light pulse

with temporal spread below a nanosecond was injected into the scattering medium. Both

the absorption and scattering coefficients can be obtained by fitting the acquired TPSF

into a diffusion model. Extensive clinical breast imaging data have been collected using

time domain imaging systems by the research groups in Milan [38] and Berlin [39]. Both

groups developed TD scanning optical mammograms using a transmittance geometry.

FD imaging systems using radio-frequency (RF) modulated light, measure the

changes in both amplitude and phase of the transmittance/reflectance light. Several

research groups have developed FD NIRS/NIRST imaging systems with different

features [40-42]. In particular, the group at the University of Pennsylvania developed a

stand-alone clinical NIRST system, utilizing a CCD-based heterodyne detection scheme

[42], shown in Fig. 1.2(c). A large number of source-detector pairs (106) were acquired

and breast boundary segmentation was realized by a fringe profilometry system. The

group led by Bruce Tromberg at the University of California, Irvine (UCI) developed a

Diffuse Optical Spectroscopic Imaging (DOSI) system, which used a hand-held probe to

7

scan over 50 to 100 discrete locations on the breast (Fig. 1.2(d)). The system was tested

in an independently-executed, prospective multicenter clinical trial of breast cancer NAC,

with promising results for early detection of changes in tumor response [43].

NIRS/NIRST imaging in general suffers from relatively low spatial resolution due

to the diffusive nature of photon transport in scattering medium, which also comes with

the challenge of solving ill-posed systems in the image reconstruction. One solution is

multi-modality imaging, where the spatial information from another imaging method

(MRI, ultrasound, CT or mammogram) was encoded into the optical image

reconstruction procedure. For instance, a hybrid imager (Fig. 1.2(e)) was developed by

Zhu et al for simultaneous ultrasound and NIRS imaging [44]. A customized hand-held

probe was constructed, where a commercial US transducer was located in the middle, and

optical source and detector fibers were placed at the periphery. Besides ultrasound

imaging, NIRS was also combined with MRI imaging. A hybrid MRI-guided multi-

wavelength FD-CW NIRST system was developed by El-Ghussein et al at Dartmouth

College (Fig. 1.2(f)), which acquired simultaneous optical scan with DCE-MRI imaging

[3]. The system was used in a clinical trial conducted in Xijing Hospital, XI’an, China,

involving 44 subjects in total, of whom 28 had malignant pathological diagnoses, and 16

had benign lesions. The results of this study suggest that MRI-guided NIRST can

distinguish malignant lesions from benign conditions in women with undiagnosed breast

abnormalities via optical imaging biomarkers of total hemoglobin concentration and a

tissue optical index [45].

8

Figure 1.2. Six NIRS/NIRST systems for breast imaging. (a) The DSP based CW NIRST system developed at Columbia University. (b) The CW NIRST system developed by Philips Healthcare. (c) The stand-alone clinical NIRST system developed at University of Pennsylvania, utilizing a CCD-based heterodyne detection scheme. (d) The DOSI system developed at University of California, Irvine. (e) Combined US and NIRS systems and a handheld probe developed at University of Connecticut. (f) The MRI-guided NIRST system developed at Dartmouth College.

1.4 NIRST imaging of breast cancer at Dartmouth

The Dartmouth group has worked on developing NIRST systems and

reconstruction algorithms during the past two decades. The instrumentation of NIRST has

been developed and upgraded by McBride and El-Ghussein et al [3, 46, 47]. Srinivasan,

Dehghani and Jermyn et al developed and iteratively improved the reconstruction

algorithm and package, NIRFAST [48-50]. Wang et al combined FD and CW to acquire

broadband NRIST imaging [51, 52]. Mastanduno and Carpenter et al developed the

9

methodology for three-dimensional MR-guided NIRST imaging [45, 53-56]. Significant

clinical progresses have been made in the following two areas of breast imaging using

NIRST: breast cancer diagnosis and treatment monitoring to NAC, as outlined below.

1.4.1Breast cancer diagnosis

The performance of a MRI-guided NIRST system has been validated through a

recent clinical trial, of which the results suggested that MR-guided NIRST can increase

diagnostic performance of breast MRI [45]. Figure 1.3 shows the reconstructed optical

images overlaid on T1 MRI image for a patient with IDC breast cancer. The whole breast

was segmented into three regions of tumor, fibroglandular and fat, from MRI images.

Each region was supposed to have the same optical properties, and the contrast was

defined as the ratio between tumor to the surrounding normal tissue in terms of

concentration of different optical biomarkers. The contrast was then used to separate

malignant tumors from benign lesions.

Since the image reconstruction in NIRST is an ill-posed problem, an L-2 norm

regularization technique has been added into the objective function in the inversion

procedure, which can be optimized from different perspectives, as will be discussed in

later chapters. One of the major goals in this thesis was to develop optimal regularization

strategies in this MRI-guided NIRST image reconstruction.

10

Figure 1.3 Images from a patient with a 11x21x14mm biopsy-confirmed Invasive Ductal Carcinoma (IDC) in her right breast. (a) non-contrast T1 MRI with tumor location indicated (arrow); (b) Reconstructed images for HbT, (c) StO2, (d) water, (e) lipid, (f) scattering amplitude, and (g) scattering power is overlaid on the MR scan. The value of each parameter in the adipose region is suppressed for clarity of visualization.

1.4.2 Monitoring treatment response to Neoadjuvant Chemotherapy

A series of clinical trials have been conducted by Jiang et al [2, 30, 57], using the

stand-alone system for monitoring treatment response to NAC. In one previous study [2],

a group of breast cancer patients undergoing NAC were imaged with NIRS before, during

and after the treatment. Significant differences were found between pathologic complete

response (pCR) versus pathologic incomplete response (pIR) group, based on the relative

change in tumor HbT within the first cycle of chemotherapy treatment. Figure 1.4 shows

the reconstructed NIRST images for a pCR case prior, during and post NAC. It is clearly

11

seen from the HbT images that the tumor contrast decreased after the first cycle of

treatment.

Moreover, pretreatment HbT relative to the contralateral breast showed potential

to separate pCR from pIR group as well. Since pCR patients were reported to have higher

disease-free survival rates [58, 59], early monitoring and prediction of pCR/pIR category

has the potential to individualize patient treatment plan even before treatment starts. One

major logistical problem in adoption of imaging is that if the imaging requires an

additional patient visit, then the cost to the system and to the patient time may inhibit its

use. Thus, one of the major goals of this thesis was to develop a system which could

work by imaging patients in the chemotherapy infusion suit, and image quickly, such that

the time and effort involved in this exam does not compromise its value.

Figure 1.4. (a) Reconstructed optical images of a pCR case prior, during and post NAC. (b) Axial post contrast subtraction MRI before NAC shows and enhancing mass indicated by arrow [2].

12

1.5 Organization of this thesis

The structure of this thesis is listed as follows:

Chapter 2 introduces the physics between light-tissue interactions in scattering

medium, and the derivation of the diffusion equation and its implementation using the

finite element method (FEM). The basic theory about model inversion is explained, with

spectral and spatial prior information encoded during the inversion procedure. An

optimization algorithm of regularization parameter based on L-curve analysis is

developed for MRI-guided NIRST image reconstruction. Another optimal regularization

technique is also introduced, which directly encodes the spatial information into the

inversion matrix, without manual segmentation of MRI images.

Chapter 3 discusses the design of a 12-wavelength FD-CW portable NIRST

system, with simultaneous acquisition of three FD and three CW wavelengths. The

system provides wavelength coverage between 661nm to 1064nm, with one complete

data acquisition involving 12 wavelengths acquired in less than 2 minutes. A customized

breast-fiber interface is introduced as well.

Chapter 4 details the optimization of patient image reconstruction for MRI-guided

NIRST, using a dataset obtained in the clinical trial of 50 surgical patients in Xi’an,

China. The L-curve based optimization algorithm of regularization parameter, as

discussed in Chapter 2, is applied in the patient dataset. The performance of the other

optimal regularization strategy is also discussed in a case study.

Chapter 5 presents extensive phantom studies using the NIRST system. Typical

tissue simulating phantoms are compared. The making and characterization of a novel

soft gel phantom are discussed in detail. The soft gel phantom enables the design of

13

heterogeneous phantom with sphere shape inclusion, which better mimics the natural

shape of breast tumor than other existing phantoms. The reconstructed images of HbT,

StO2, water, lipids, and scattering properties are shown for a gelatin phantom made with

porcine blood.

Chapter 6 discusses collagen quantification in breast tissue. Tomographic images

of breast collagen content are shown for the first time, and image reconstruction

approaches with and without collagen content included have been validated in simulation

studies and normal subject exams. The reconstructed collagen image of a breast cancer

patient is presented, and the recovered tumor/background contrast in total hemoglobin

increases from 1.5 to 1.7 when collagen is included in reconstruction.

Chapter 7 presents an imaging study on a group of healthy volunteers with

various breast sizes and radiographic densities. Statistical analyses are performed using

the reconstructed optical biomarkers. The adjustable breast-fiber interface is tested and

validated as well.

Chapter 8 presents case studies of monitoring patient response to NAC. The

optical images are reconstructed prior, during and post NAC for two pIR cases. HbT

shows to be the strongest biomarker in predicting breast tumor response to NAC.

Chapter 9 summarizes conclusions learned from the work presented in this thesis,

and discusses the future directions.

The Appendix lists the reconstruction programs and data acquisition programs

developed in this thesis.

14

Chapter 2: Theory and Image Reconstruction Methods

2.1 Introduction

The goal of near infrared spectral tomography (NIRST) or more generally, diffuse

optical tomography (DOT), is to reconstruct a spatial map of optical absorption

coefficients, or absorption related chromophore concentrations, and scattering

coefficients, from fluence measurement, using a forward model for mathematically

describing photon propagation [60]. The most common approach to this model is the

neutral radiative transport equation (RTE), which has been used to simulate light

transport in a wide range of media, where light is treated as composed of distinct photons,

propagating in a medium with characteristic absorption and scattering properties [61]. In

the case of scattering medium such as breast tissues, where the scattering probability is

much higher than absorption, the RTE can be approximated by the photon diffusion

equation [62, 63], the derivation of which was presented in section 2.2.

Since photon diffusion equation is a nonlinear equation, the recovery of unknown

optical properties usually requires a two-step procedure, forward and inverse models. In

the forward model, boundary fluence was calculated given existing optical properties.

The difference between measured fluence and computed fluence was minimized during

the inverse model, giving an update to the optical properties.

The are several approaches for solving the forward RTE problem. Analytical

solutions are only available for simple geometries such as circles and rectangles, but

difficult to get in real clinical settings [64]. Statistical methods such as Monte Carlo has

shown to be the most accurate approach in modeling light transport [65], which is

performed by tracing individual photon histories [66]. Photons are treated as distinct

15

particles with given probability of scattering and absorption, at every point in discrete

geometry [67]. A large number of individual photons, usually millions, are launched from

the source position, and statistical sample results are collected at the detector positions.

However, in order to get precise results, a very large number of photon histories are

required, which can be expensive in computational time and complexity. Another

approach is numerical approximation of Partial Differential Equations (PDEs) that

approximate the RTE, such as the diffusion approximation. This PDE can be efficiently

solved with numerical approaches such as the finite element method (FEM) or the

boundary element method (BEM) [48, 68]. Section 2.3 discusses the implementation of

solving forward model using FEM approaches, which was used throughout this thesis.

Due to the diffusion nature of light transport in scattering medium, which is a

non-linear second order PDE, the problem in NIRST image reconstruction is also

nonlinear between the coefficients and the fluence, and so the inverse problem is highly

ill-posed and ill-conditioned [69]. Regularization and optimization techniques have been

applied in solving the inverse problem, in order to stabilize inversion of forward models

[70-72]. In this thesis, the inversion was achieved by a modified-Tikhonov minimization,

discussed in section 2.4.

The absorption and scattering coefficient can be recovered using fluence

measurement data acquired at a single wavelength. Given a set of fluence measurements

acquired at multiple wavelengths, and the absorption spectrum of absorbers of interest in

soft tissues, the concentrations of major chromophores including oxy- and deoxy-

hemoglobin, water, and lipid, can be recovered through NIRST image reconstruction. A

direct spectral reconstruction method has been used, where the spectral priors are directly

16

encoded into the inversion matrix, which is more accurate and robust than traditional

indirect spectral reconstruction approach [73, 74], and the details were outlined in section

2.5. The implementation of building a spectral Jacobian matrix combining FD

measurement data including both amplitude and phase data, and CW measurement data

including only amplitude data, was discussed as well.

Multi-modality imaging such as positron emission tomography/computed

tomography (PET/CT) has matured into an important diagnostic tool [75]. Similarly, the

combination of MRI and NIRST is has been studied by several research groups [53, 76-

78]. The excellent soft tissue contrast provided by MRI can be used as prior spatial

information to guide the reconstruction of optical images. In addition, the functional

information provided by NIRST shows potential in increasing the specificity of MRI in

breast cancer diagnosis [45]. Two approaches of incorporating MRI information into

NIRST image reconstruction were discussed in section 2.6. In the first “hard prior”

method, computational meshes were created for the whole breast consisting of three

regions composed of adipose, fibroglandular and suspected tumor tissue, using the

anatomical information provided by the MR images. Each region was assumed to have

uniform optical properties, and the absorption and scattering parameters were estimated

for all three regions. The number of unknown parameter reduces to three, which

significantly simplifies the image reconstruction. The other method encodes MRI

structural information into the NIRST reconstruction in a soft way, which has the benefit

of eliminating user intervention such as image segmentation of distinct regions.

Specifically, the Dynamic Contrast Enhanced Magnetic Resonance (DCE-MR) image

17

intensity value differences within the anatomical image were used to implement an

exponentially-weighted regularization function between the image pixels [79, 80].

2.2 Modeling of photon propagation in highly scattering medium

2.2.1 Optical characteristics of biological tissues

The optical properties of soft tissue have been investigated intensively [81-83].

The wavelength-dependent optical properties are usually described in terms of absorption

coefficient (µa (cm−1)), scattering coefficient (µs (cm−1)), scattering function (p(θ,ψ)

(sr−1)) where θ is the deflection angle of scatter and ψ is the azimuthal angle of scatter,

and the real refractive index n'. In thicker tissues where multiple scattering occurs and the

ψ-dependence is averaged and then ignored, scattering is always described as reduced

scattering coefficient µ’s = (1-g) µs. The g value is an anisotropy factor defined as the

averaged cosine of the scattering angle θ. Absorption coefficients and reduced scattering

coefficients are recovered as spatially distributed maps through NIRST image

reconstruction. In the near infrared (NIR) wavelengths (650nm-1000nm), absorption

coefficient of major absorbers in soft tissue is significantly lower than that in the other

wavelength range, which allows the light to penetrate soft tissues up to 8-10cm and be

measured at a detectable level [84].

2.2.2 Photon diffusion equation

The behavior of interactions among a large number of photons in turbid media

can be described by the radiative transport or Boltzmann transport equation:

1

𝑣

∂𝐿(𝑟,�̂�,𝑡)

∂𝑡+ ∇ ⋅ 𝐿(𝑟, �̂�, 𝑡)�̂� + 𝜇𝑡𝐿(𝑟, �̂�, 𝑡) = 𝜇𝑠 ∫ 𝑓(�̂�, �̂�′)𝐿(𝑟, �̂�, 𝑡)

4𝜋𝑑�̂�′ + 𝑄(𝑟, �̂�, 𝑡), (2.1)

where v is the speed of photons in the medium, and 𝐿(𝑟, �̂�, 𝑡) is the radiance (power per

unit area and unit solid angle) as a function of position r, in the direction Ω̂ at time 𝑡.

18

𝜇𝑡 = 𝜇𝑎 + 𝜇′𝑠 is the optical transport coefficient. Here 𝑓(�̂�, �̂�′) is the scattering phase

function, and 𝑄(𝑟, �̂�, 𝑡) is the radiant source function. The left-hand side of equation

(2.1) accounts for photons leaving a small element in phase space, and the right-hand side

accounts for photons entering it. The radiative transport equation can be simplified based

on diffusion theory if the scattering probability is much larger than that of absorption, or

𝜇′𝑠 ≫ 𝜇𝑎, with an isotropic source. Calculation of the diffusion model frequency domain

data then follows from

( ) ( , ) ( ) / ( , ) ( , )a oi c Q D r r r r r , (2.2) where an isotropic source, 0Q , with source frequency at position r delivers light

through turbid media. Here, represents the fluence rate at position r observed at

frequency . Also, ( )a r is the optical absorption coefficient and ( )D r is the optical

diffusion coefficient which is defined as 𝐃(𝐫) =1

3[𝜇𝑎(𝐫)+𝜇′𝑠(𝐫)]

.

To solve the diffusion equation in frequency domain, certain boundary conditions

need to be assigned. The air-tissue boundary can be represented by an index of refraction-

mismatched mixed type-III boundary condition, in which the fluence at the edge of the

tissue exists but does not return [68]. The relationship is described in the following

equation:

ˆ, 2An , 0 (2.3)

where is a point on the boundary, and n̂ is a vector pointing outwards, normal to the

surface. A can be derived from Fresnel’s law:

3

02

2 / 1 1 cos1 cos

c

c

RA

, (2.4)

19

where 1arcsin( / )c airn n is the angle at which internal reflection occurs, and

2 2

0 1 1/ 1 / / 1 .air airR n n n n At the external boundaries, refractive index (RI) airn is

generally assumed to be equal to that of free space, so that airn =1.

The assumption is that 𝜇′𝑠 ≫ 𝜇𝑎 is satisfied in breast tissues with typical 𝜇𝑎

between 0.002 and 0.1 mm-1, and 𝜇′𝑠 between 0.5 and 2 mm-1. The other assumption of

isotropic source is valid when source-detector separation is higher than three to five

reduced scattering lengths. In practice, the source is treated as an isotropic point source

and located one scattering distance interior to the actual source position.

2.3 Numerical modeling of the forward problem

The finite element method (FEM) has been widely used as a general and flexible

method in solving forward problem in arbitrary geometries [68]. In the FEM framework,

the diffusion equation in (2.2) can be expressed as a system of linear algebraic equations,

where the source term is defined as distributed Gaussian source. The imaging domain is

discretized into a series of small regions called elements, each of which consists of two or

more local nodes associated with piecewise linear basis functions. The photon fluence is

calculated at each location in the region in the forward model. The difference between

calculated fluence and measured fluence at the boundary detectors is minimized during

the inversion.

A MATLAB-based FEM software package, NIRFAST (Near Infrared

Fluorescence, Absorption and Scatter Tomography), has been developed in the optics in

medicine group at Dartmouth College [49]. This package includes toolboxes for creating

meshes of simple and complex geometries, calibrating raw measurement dataset, solving

frequency-domain NIRST reconstruction problem at single-wavelength or multi-

20

wavelength with spectral priors, encoding spectral priors and fluorescence imaging

problems. Figure 2.1 shows FEM meshes created in NIRFAST for three geometries: (a)

2D circle, (b) 2D customized football-shape interface, and (c) 3D actual breast.

Figure 2.1. FEM meshes with nodes and elements created in NIRFAST for three geometries: (a) 2D circle, (b) 2D football-shape and (c) 3D actual breast.

2.4 Inverse problem solver

The goal of the inverse problem is to reconstruct a spatial map of optical

absorption coefficients, or absorption related chromophore concentrations, and scattering

coefficients, from using measurements of light fluence from the tissue surface. This is

realized by minimizing the difference between measured and calculated fluence. An

objective function is defined as:

2 2

1

( )NM

M Ci i

i

, (2.5)

where Mi and C

i is the measured and calculated fluence at detector i, respectively. NM

represents the number of measurements. The inversion can be achieved by using a

21

modified-Tikhonov minimization. A penalty term is added into the original objective

function:

2 20

1 1

( ) ( )NM NN

M Ci i j

i j

, (2.6)

where NN is the number of FEM nodes or unknown optical parameters, and j is the

optical parameter at node j. 0 symbolizes the initial estimates of NIRS properties in the

tissue and it can be obtained from the homogeneous fit in the calibration procedure. Here,

is the regularization parameter which balances the relative magnitudes of the two parts

of the objective function – the data-model mismatch is represented by the first term, and

the difference between the current estimates of optical properties and the initial starting

values is expressed by the 2nd term.

In practice will not be equal to zero, and we are interested in finding the value

of j such that

is close to zero. We first expand this term using Taylor series

expansion method based on for some nearby point 0 :

0 0 0 ....dd

, (2.7)

Then after setting

=0 and ignoring higher order terms, we can get:

1

1i i i id

d

, (2.8)

22

This is an update equation of , where i and 1i represents updated at ith

and (i+1)th iteration, respectively. Solving for i

and i

dd

from equation

(2.7), we will get:

02 2Tc

c Mi i

, (2.9)

and 2

22 2 2T Tc c c

c Mi

d d dd d d

. (2.10)

The second order derivative term 2

22Tc

c Mdd

is very small compared

with the first term, and then ignored here. Plugging equation (2.9) and (2.10) back into

equation (2.8), the update equation of becomes:

1

1 0

T Tc c cc M

i i iI

, (2.11)

where I is an identity matrix, and c

is usually called Jacobian Matrix J using standard

terminology. The update vector 1i i is replaced with . The last term in equation

(2.11) comes from the penalty term in the objective equation, and can be ignored if i is

close to 0 . The update equation of (2.11) can then be written as:

1T T C MJ J I J

, (2.12)

The matrix TJ J , also called Hessian Matrix, is highly ill-conditioned and ill-

posed. The inversion of TJ J is stabilized by the addition of J , which makes the matrix

23

more diagonally dominant. For FD measurement, the fluence consists of both

amplitude I and phase . Jacobian matrix J is built as:

1 1 1 1 1 1

1 2 1 2

1 1 1 1 1 1

1 2 1 2

2 2 2 2 2 2

1 2 1 2

2 2

1 2

ln ln ln ln ln ln... ; ...

ln ln ln ln ln ln... ; ...

ln ln ln ln ln ln... ; ...

ln ln ..

NN a a aNN

NN a a aNN

NN a a aNN

I I I I I ID D D

D D DI I I I I I

D D D

JD D

2 2 2 2

1 2

1 2 1 2

1 2 1 2

ln ln ln ln. ; ...

ln ln ln ln ln ln... ; ...

ln ln ln ln ln ln... ; ...

NN a a aNN

NM NM NM NM NM NM

NN a a aNN

NM NM NM NM NM NM

NN a a aNN

D

I I I I I ID D D

D D D

(2.13)

In NIRFAT, the Jacobian matrix is built at each iteration using the adjoint

method, which is computationally efficient because it takes advantage of reciprocity [85].

2.5 Spectral prior reconstruction

The single-wavelength reconstruction outlined in previous sections provides

estimate of a and s at single wavelength. Once multiple-wavelength data are obtained

from the NIRST measurement, 2D/3D image of chromophore/absorber concentration can

be reconstructed. One approach is using constrained linear square fit to the Beer’s law

relation:

a i ii

C , (2.14)

where i and iC represents molar absorption coefficient and concentration of

chromophore with index of i, respectively. Similarly, the s spectrum can fit into an

empirical power law approximation to Mie scattering theory:

24

bs a , (2.15)

where scattering amplitude (a) and scattering power (b) are estimated [86, 87]. Instead of

reconstructing spatial map of a and s first at multiple wavelengths, and then applying

equation (2.14) and (2.15) in post-processing fitting, these constraints can be directly

incorporated into the inversion matrix for direct reconstruction of chromophore

concentrations and scattering properties. The modified Jacobian matrix becomes:

1 1 2 1 1 1 1

1 2 2 2 2 2 2

1 2

;

;

;

c c cM A b

c c cM A b

c N c N cM N A N b N

J J J J J

J J J J JJ

J J J J J

, (2.16)

where N is the number of wavelengths, and M is number of chromophores. The update

equation is also modified accordingly:

1 1

1

1

N N

C M

T TM

C M

c

c J J I Jab

, (2.17)

The ill-posedness of the modified matrix is improved because multiple-

wavelength data are used to reconstruct all parameters simultaneously, which also

reduces the number of unknown parameters. Compared with the traditional method, the

direct spectral prior method is more accurate and more robust to noise, and also shows

better performance in both phantom and patient experiments [88].

In this thesis, the dataset was acquired from a hybrid multi-wavelength FD-CW

NIRST system consisting of FD data (amplitude and phase) at six wavelengths, and CW

data (amplitude) at six wavelengths. Since there is no phase in CW data, the Jacobian

25

matrix must be adjusted by padding certain rows with zeros where phase data would be,

in order to include both two data types into the reconstruction.

2.6 Spatial prior reconstruction

2.6.1 Hard-prior reconstruction

Reduction of the recovered parameter space, , into a smaller number of larger

regions segmented from MRI scans is known as encoding hard-prior information into the

inversion [89]. The NIRST solutions were obtained with a three-dimensional image

reconstruction algorithm [69] and prior information extracted from the co-registered

breast MR images. Here, the assumption was made that the segmented regions from MRI

– adipose, fibroglandular, and suspected tumor – had relatively homogeneous NIRS

properties, and the goal of MRI/NIRST was to recover the corresponding region-based

values. Figure 2.2 shows an example of segmentation of breast MRI images using

NIRFAST-Slicer.

Figure 2.2. Image segmentation in NIRFAST-Slicer.

Once segmentation of the breast is finished, the spatial information can be

incorporated into the inverse process, and a region mapping matrix K is constructed [90]:

26

1 1 11,1 1,2 1,

1 1 12,1 2,2 2,

1 1 1,1 ,2 ,

2 2 21,1 1,2 1,

2 2 22,1 2,2 2,

2 2 2,1 ,2 ,

1,1 1,2 1,

,1 ,2 ,

NC NC NC

NC NC NC

c c cNR

c c cNR

c c cNN NN NN NR

c c cNR

c c cNR

c c cNN NN NN NR

c c cNR

c c cNN NN NN NR

k k k

k k k

k k k

k k k

k k kK

k k k

k k k

k k k

, (2.18)

where NC, NN, NR represents number of concentrations, number of nodes, and number

of regions, respectively. The element in K is defined as:

,

10

ci j

if i region jk

if i region j

, (2.19)

Then the Jacobian matrix is mapped into region space:

J JK , (2.20)

The update equation of becomes:

1T T C MJ J I J

, (2.21)

Hard-prior reconstruction significantly reduces the number of unknown

parameters from number of nodes to number of regions, and the image reconstruction

becomes well-determined. Though the ill-posedness of the problem is improved, it’s still

ill-conditioned, which means a small permutation in the boundary measurement data or

initial guess can have much larger effects in the estimated optical properties, because of

the diffusive nature of photon transport in scattering medium. Therefore, the

27

regularization technique is still indispensable and proper choice of regularization

parameter is critical during the inversion.

A common approach to selecting the regularization parameter, , is to use a

fixed empirical number based on prior tissue-phantom studies, and to apply this value to

data acquired from patient exams. In this study, an L-curve approach was used to find the

optimal regularization parameter [91, 92]. The L-curve is a parametric plot of the L-2

norm of the data-model mismatch ( ) versus the difference between optical properties

of two iterations ( ) where:

( ) J (2.22)

2

01

( )NN

ji

(2.23)

In other words, ( ) and ( ) represent the model-data error and spatial prior

error, respectively, each of which is being minimized during the model inversion. With a

relatively small , ( ) dominates the objective function, and a lower model error is

expected at the cost of a larger prior error, and vice-versa for the case of large . Plotting

( ) vs. ( ) for a range of illustrates the trade-off between these two types of error,

which typically exhibits an L-shaped curve. The corner of the L-curve is commonly

regarded as the optimal regularization because it minimizes the two error terms. In this

study, the L-curve method was applied to determine the optimal regularization at each

iteration.

28

Figure 2.3. (a) Flowchart outlining the sequence for the optimization algorithm. In (b), the L2 norm of the prior property error vs. L2 norm of the model error creates the L-curve of values for each regularization parameter value from 0.001 to 100. The optimization algorithm is implemented at each iteration, and optimal regularization parameter from previous iteration is used once the L metric falls behind the threshold.

Figure 2.3(a) outlines the sequence for optimization of regularization based on L-

curve analysis at each iteration. To begin, prior error, which is the difference in optical

property solutions between two iterations, is plotted vs. model error over a range of

discrete regularization from 0.001 to 100. Fig. 2.3(b) shows a typical L-curve at the first

iteration of image reconstruction for a patient with a malignant breast abnormality. Here,

the prior error, ( ) , is defined as the difference of the sum of squared reconstructed

optical parameters between the first iteration and initial estimate. The model-data

mismatch, ( ) , depicts the model error in the first iteration. When regularization

29

increases, the model error increases and prior error decreases. Next, constrained regions

are selected. In the range of regularization values from 0.001 to 0.02, the behavior of the

prior error vs. model error does not show a linear relationship because a small

regularization results in unstable solutions in the ill-conditioned inverse problem. Once

the constrained regions are selected, the L-curve can be fitted through least squares to

slopes of vm and hm for vertical and horizontal regions, respectively, and a point on the

L-curve with maximal second derivative can be obtained. If this L-metric is larger than a

threshold, an optimal regularization is assigned as the point which is closest to the

intersection of the two fitting lines. Otherwise, the fitting at this iteration is discarded,

and the optimal regularization value from the previous iteration is retained. The algorithm

is repeated for each iteration until the stopping criterion for image reconstruction is

satisfied. The optimization algorithm has been applied in clinical data, and the detail will

be discussed in Chapter 3.

2.6.2 Soft prior reconstruction

Other than hard-prior reconstruction introduced in the last section, there exists

another strategy of encoding spatial information, called soft-prior reconstruction. In this

section, a novel regularization scheme is applied which directly encodes information

about the structural images, rather than enforcing uniformity within manually-segmented

regions. It is referred to as the DRI method and constrains FEM nodes according to their

corresponding grayscale value differences within the coregistered companion image

volume. In this case, the regularization matrix operator can be written as:

30

(2.24)

where γ is the anatomical image grayscale value which corresponds to a particular FEM

node (in this thesis, grayscale values were normalized to the maximum within the image),

σg is the characteristic grayscale difference over which to apply regularization, and Mi is

a normalization factor chosen for each row.

By performing a similar iterative Gauss-Newton reconstruction method through

Eq. (2.12), the general update equation for the underdetermined form can be expressed as

(2.25)

where is the update to the parameters; is the Jacobian matrix which is the

derivative of the measurements, f (x), with respect to the optical property parameters of

interest at the k-th iteration, and has the dimension of , M is the number of

measurements, and N is the number of parameters, x; the superscript T denotes the

transpose, is the forward solution using the estimated parameters form the

iteration. The hard-prior and DRI reconstruction method outlined in this chapter was

compared in chapter 3.

1

1 exp , 2i jij

i g

i j

Lotherwise

M

1

1T T T

k k k kx J J L L J d f x

kx kJ

M N

1kf x 1k

31

Chapter 3: Optimization of Image Reconstruction in MRI-guided

NIRST for Breast Cancer Diagnosis

3.1 Introduction

Magnetic resonance imaging (MRI) guided near infrared optical spectral

tomography (NIRST) has the potential to add molecular information to the spatial maps

of MR imaging of the breast, and thereby increase the specificity of contrast-enhanced

MRI exams [78, 93]. Niziachrisos et al. and then Brooksby et al. first developed

combined MRI-NIRS systems for concurrent MRI and optical imaging [46, 94, 95]. El-

Ghussein and Mastanduno et al. later developed a hybrid FD-CW MRI-guided NIRS

system, which was validated in a clinical trial in Xi’an, China [3, 45, 56, 96]. In this

chapter, studies on optimizing MRI-guided NIRS image reconstruction from multiple

perspectives were presented.

NIRST image recovery [97] is nonlinear and ill-posed and has been the subject of

many years of research [89, 98, 99]. The diffuse propagation of NIRS light in tissue

generates a poorly conditioned matrix that requires inversion. A widely used approach to

solve the inverse problem is the Newton-Raphson technique regularized by a modified

Levenberg-Marquardt algorithm [69, 100]. Implementation of a method to automate the

choice of regularization is critical for patient imaging [91, 92, 101, 102]. While methods

to choose this parameter automatically are well established in computational studies, little

investigation of how the selection influences the diagnostic performance of the imaging

method in actual practice has been reported. In section 3.2, optimization of the

regularization parameter based on L-curve analysis was pursued in clinical MRI-guided

NIRST imaging with the specific goal of retrospectively maximizing the discrimination

between known malignant and benign breast scan [103]. A classically defined L-curve

32

optimization algorithm was developed for regularization parameter selection during

image reconstruction of data from 25 patients collected in a clinical study of women with

breast abnormalities (BI-RADS category 4-5) of unknown diagnosis at the time of the

imaging exam. The methodology was also used to study NIRST reconstruction with

either amplitude data or both amplitude and phase data acquired with a multi-channel 100

MHz frequency domain NIR spectroscopy system. The performance of the optimal

regularization was compared to fixed regularization in this setting of differentiating

malignant from benign breast abnormalities. This study represents the first time the

effects of regularization have been investigated in MRI-guided NIRST based on a

relatively large data set of clinical exams with the goal of optimizing task-based

discrimination.

Other than the “hard-prior” image reconstruction discussed in section 3.2, there are

“soft-prior” reconstruction methods realizing spatially encoded regularization, where co-

registered image information is applied in a pre-defined way, such as through a Laplacian

filter or a depth dependent function [95, 104]. Both hard prior and soft prior

reconstruction methods require the segmentation of breast into different small regions. In

section 3.3, a new approach for incorporating image information directly into the

inversion matrix regularization was examined using Direct Regularization from Images

(DRI), which encodes the gray-scale data into the NIRST image reconstruction problem.

This process has the benefit of eliminating user intervention such as image segmentation

of distinct regions. Specifically, the DCE-MR image intensity value differences within

the anatomical image were used to implement an exponentially-weighted regularization

function between the image pixels.

33

3.2 Optimization of regularization parameter in MRI-guided NIRST for breast

cancer diagnosis

3.2.1 Reconstruction and visualization of optical images in MRI-guided NIRST

The combined FD-CW data acquisition provides amplitude and phase recordings

from FD measurements involving six wavelengths, and amplitude data from CW

measurements involving three wavelengths. Using the amplitude and phase data at these

six wavelengths, the absorption and scattering coefficients at each wavelength were

estimated from homogenous fitting to obtain the initial estimates of oxy-hemoglobin

(HbO), deoxy-hemoglobin (Hb), water, lipid, scattering amplitude (SA) and scattering

power (SP). Using the anatomical information provided by the MR images,

computational meshes were created for the whole breast consisting of three regions

composed of adipose, fibroglandular and suspected tumor tissue. Each region was

assumed to have uniform optical properties, and the absorption and scattering parameters

were estimated for all three regions. From the recovered chromophores concentrations,

physiologically relevant parameters were calculated, including total hemoglobin HbT =

HbO + Hb, oxygen saturation StO2=HbO/HbT, and tissue optical index TOI = HbT

Water/Lipid. Optical property contrast, defined as the ratio of the suspected tumor to

background (adipose) properties, was used to differentiate malignant from benign

abnormalities.

34

Figure 3.1. Images from a 33-year-old patient with a 11x21x14mm biopsy-confirmed Invasive Ductal Carcinoma (IDC) in her right breast. (a) non-contrast T1 MRI with tumor location is indicated (arrow); Reconstructed images of (b) HbT, (c) StO2, (d) water, (e) lipid, (f) scattering amplitude, and (g) scattering power are overlaid on the MR scan. The value of each parameter in the adipose region is suppressed for clarity of visualization.

Figure 3.1 shows recovered images of HbT, StO2, water, lipid, scattering amplitude (SA),

and scattering power (SP), of a patient with a biopsy-confirmed breast malignancy. The

optical images are overlaid on the MR scan, and the value of each parameter in the

adipose region is suppressed for clarity of visualization. The tumor-to-adipose contrasts

of HbT, StO2, water, lipid, SA and SP were found to be 1.2, 1.1, 1.3, 1.6, 1.4 and 1.02,

respectively. A fixed regularization parameter of 0.1 was used through the image

reconstruction procedure for this patient.

35

3.2.2 Fixed regularization parameter of 0.1 and 1 using only amplitude data

Figure 3.2. Tumor-to-adipose contrast in HbT vs. regularization for (a) benign, and (b) malignant cases. Circles have a regularization of 0.1, and asterisks have a regularization of 1. Box plots of contrast for the two pathologies are shown with fixed regularizations of (c) 0.1 and (d) 1 for all patients.

First, the tumor-to-adipose contrast in HbT vs. regularization was plotted (Fig. 3.2) for

benign and malignant patients when images were estimated from amplitude data only,

and box plots of HbT contrast in the two diagnostic groups for fixed regularization

parameters of either 0.1 or 1, respectively. Compared to the malignant group, the HbT

contrast was less sensitive to variation in regularization in the benign group, especially

over the range of values from 0.01 to 0.2. Larger separation or difference in mean HbT

contrast, between the malignant and benign groups resulted from the lower regularization

of 0.1, even though less variation occurred within the same groups with a higher

36

regularization of 1, although a statistically significant difference existed between the

means of the malignant and benign groups with p-values < 0.05 in either case.

3.2.3 Optimal regularization using only amplitude data

Figure 3.3. L-curves are shown for the first 3 iterations of a single case with a malignant tumor. The regularization was varied from 0.001 to 100 at each iteration, and the optimal regularization was (a) 0.18, (b) 0.22, and (c) undefined for the three iterations, respectively. Histograms of optimal regularization parameter for the 1st and 2nd iteration are inserted to Figs. 3-3(a) and (b), respectively.

Figure 3.3 shows L-curves for the first 3 iterations during optical image

reconstruction of a patient with a malignant breast abnormality. The same range of

regularization (0.001 to 100) was used to plot the L-curve at each iteration. Both the

range of model error and prior error decrease with increasing number of iterations.

Optimization of regularization was implemented using the algorithm outlined in Chapter

2, and the optimal point is marked by a red asterisk in the first 2 iterations, and had values

of 0.18 and 0.22, respectively. At the third iteration, the L-curve metric was below the

threshold, and an optimal regularization could not be found using the same process.

Instead, the regularization value for the third iteration was set to be the same as used in

the second iteration, namely 0.22. The L-curve of the optical data acquired from most

patient exams exhibited similar behavior. The number of patient data sets which had an

optimal regularization based on L-curve analysis was 22, 12 and 0 for the first three

iterations. The average optimal regularization for the 1st and 2nd iterations was 0.21, and

37

0.24, respectively. After three iterations, none of the exam data had an L-curve with a

metric higher than the threshold.

Figure 3.4. (a) Log scale of projection error vs. number of iterations. (b) Tumor/adipose contrast in HbT vs. number of iterations.

The number of iterations used in the reconstruction was also an important factor. In Fig.

3.4(a), the projection error, which represents the residual of the inversion equation, is

plotted on a log scale as a function of number of iterations and decreases with increasing

number of iterations. The largest decrease in the projection error occurred at the first

iteration, which is typical behavior of a Newton-type iterative method. The change in the

projection error by the 9th iteration was relatively small, resulting in convergence of the

tumor/adipose contrast as shown in Fig. 3.4(b). 9 iterations were chosen as the stopping

criteria for image reconstruction of all patients.

38

Table 3.1. Comparison of statistics using fixed regularization of 0.1 and 1, and optimal regularization. Only amplitude data (AMPL) was used for image reconstruction.

aSignificant difference defined by p-value<0.05.

Table 3.1 summarizes the statistical and diagnostic results in terms of HbT and

TOI contrast using fixed regularization of 0.1 and 1, and optimal regularization based on

L-curve analysis. Data from 22 patient exams which had an optimal regularization in the

first iteration were included in the analysis. A significant difference (p<0.05) was found

in HbT contrast for malignant vs. benign patients for all three regularization selections.

3.2.4 Fixed regularization parameter of 0.1 and 1 using both amplitude and phase data

Figure 3.5. Tumor-to-adipose contrast of HbT vs. regularization for (a) benign conditions and (b) malignant tumors. Circles have a regularization of 0.1, and asterisks have a regularization of 1. Box plots of contrast are shown for regularization of (c) 0.1, and (d) 1 for all patients. Both amplitude and phase data were used in the image reconstructions.

39

Both amplitude and phase data were used in the image reconstruction to see if the

involvement of phase data obtained at the six FD domain wavelengths further improved

the accuracy of the optical image reconstruction. Figure 3.5 is organized the same as

Figure 3.2, except that the reconstructions utilized both amplitude and phase data, and the

3 clinical exams that did not have an apparent optimal regularization value in the first

iteration were excluded. In Figs. 3.5(a) and 3.5(b), the tumor-to-adipose contrast in HbT

was plotted as a function of regularization for benign and malignant patients,

respectively. When regularization decreased, the contrast increased in most cases. As a

result, better separation occurred between the malignant and benign groups, although

with higher standard deviation at the lower regularization of 0.1 as shown in Figs. 3.5(c)

and 3.5(d). Here, the HbT contrast in the benign group was more sensitive to the change

in regularization for low values when compared with the results in Fig. 3.2(a).

Apparently, determining an appropriate regularization for image reconstruction using

both amplitude and phase data is even more important than when using only amplitude

measurements.

3.2.5 Optimal regularization parameter using both amplitude and phase data

We also applied the optimization algorithm during image reconstruction with both

amplitude and phase data; but, L-curve analysis failed to find an optimal regularization.

The compromise between model error and prior error was not significant on the L-curve

at any point; hence, no optimal regularization parameter could be found that satisfied the

criterion of the optimization algorithm. One approach to improve the L-metric might be

to use separate regularization parameters for amplitude and phase data. In practice, we

40

found that a fixed regularization of 2 for phase data, and an optimal regularization based

on L-curve analysis for amplitude data worked well.

Table 3.2. Comparison of statistics using fixed regularization of 0.1 and 1, optimal regularization, and fixed regularization of 0.1 for amplitude and 2 for phase. Both amplitude and phase data (AMPL/PH) were used for image reconstruction.

aSignificant difference as defined by p-value<0.05.

Table 3.2 summarizes the statistical results of 22 patient exams, in which the

optimal regularization (on amplitude only, phase fixed at 2) was compared with fixed

regularization of 0.1 and 1, when both amplitude and phase data were used in the image

reconstruction. Reconstruction using a fixed regularization of 0.1 for amplitude and 2 for

phase data was compared as well. Significant differences (p<0.05) exist between the

malignant and benign groups in terms of both HbT and TOI for either a fixed

regularization of 1 or optimal regularization. Optimal regularization provides highest

AUC for TOI (0.94) among all the regularization choices. Meanwhile, the optimization

method still gives a relatively high AUC for HbT, without significant difference

compared with the fourth approach. In sum, the optimization technique provides a

systematic and automated way to find the optimal regularization parameter in each

iteration, which gives the best separation between malignant and benign groups in terms

of recovered optical parameter TOI. The L-curve based optimization technique utilized in

this paper aims at finding the tradeoff between prior error and model error in terms of Hb

and deoxy-Hb, water, and lipids. As a result, the increase of AUC for TOI is more

obvious than HbT compared with fixed regularization, since TOI (HbT*Water/Lipid) is

41

defined such that it represents all the recovered chromophore concentrations.

Interestingly, the addition of phase information appears to degrade the statistical

performance of the image reconstructions in the fixed regularization cases.

3.2.6 Optimal regularization leads to better separation between malignant and benign lesions

Figure 3.6. ROC curve for fixed regularization of 1 (black) and 0.1 (blue), and optimal regularization (red), when HbT and TOI are combined. Both amplitude and phase data were used for image reconstruction.

Instead of using a single predictor, either HbT or TOI, we evaluated the

combination of HbT and TOI as an indicator of malignant versus benign contrast-

enhancing MRI regions of interest. Specifically, we applied both HbT and TOI as

predictors, and fit a logistic regression using pathology-confirmed malignancy to obtain

the outcome. The predicted score from the logistic regression model was used to

construct the ROC curve [105]. We repeated this procedure for 3 different regularization

parameters and compared their AUCs with a bootstrapping method [106]. As shown in

42

Fig. 3.6, under these conditions, the optimal regularization improved the AUC (94.4%),

relative to the fixed regularizations of 0.1 (88.2%) or 1 (84.4%).

Figure 3.7. Box plots of the contrast for (a) total hemoglobin (HbT), (b) oxygen saturation (StO2), (c) Tissue optical index (TOI), (d) scattering power (SA), and (e) scattering power (SP), as recovered using the optimal regularization and amplitude and phase data. aSignificant difference.

Figure 3.7 presents boxplots of tumor-to-adipose contrast for (a) HbT, (b) StO2,

(c) TOI, (d) scattering amplitude, and (e) scattering power by applying the optimization

algorithm during image reconstruction with amplitude and phase data. HbT was the most

significant indicator for differentiating the malignant and benign groups, and provided the

largest mean difference in tumor-to-adipose contrast in the malignant and benign groups

(1.55x vs. 0.89x). A significant difference in TOI and scattering power contrast was also

observed. The average contrast in both HbT and TOI was significantly higher in the

malignant group than in the benign group. No significant difference in the StO2 contrast

was found.

3.2.7 Discussions

43

The results presented in this section show that the contrast recovered in both HbT

and TOI was diagnostically significant, but also depended on the choice of regularization

parameter. Thus, an objective methodology to select/identify the regularization parameter

for reconstructing data from individual patient exams is critical for practical application

of the combined MRI/NIRST imaging approach in the clinical diagnostic setting. Instead

of applying an empirical value for every subject exam, the optimization algorithm sought

an exam-specific regularization parameter (or series of parameters) for image

reconstruction that was derived from the measured optical data. The optimization

algorithm developed and applied here was based on L-curve analysis, which is widely

used to stabilize ill-conditioned problems by balancing the model error with the prior

error as constrained by the prior information. However, previous approaches of finding

the “elbow” on the L-curve have been limited primarily to theoretical or simulation

studies with few efforts being based on actual patient data. To deal with an L-curve

generated from clinical exam data, we defined a metric describing the characteristics of

the resulting L-shape, and developed an optimization approach based on least square

fitting of the response. The L-metric determined whether the L-curve was sufficient for

selecting an optimal regularization parameter, and if so, the optimal value from the curve

was utilized. Otherwise, the value from the previous iteration remained. The optimization

algorithm appeared to be robust and effective in the clinical data set we applied. We

found that the average L-metric of all exams decreased with increasing number of

iterations (although, the number of iterations was set to 9 in all cases, i.e., we did not

apply the L-metric or L-curve as a stopping criterion). Within 4 iterations, the projection

error and measurement noise amplification were balanced, and the contrast converged.

44

Table 3.1 shown comparisons of the diagnostic statistics resulting from three choices of

regularization parameters. Moreover, although oxygen saturation (StO2) or scattering

properties alone could not be significant indicators, their combined effects enable that

TOI becomes the best indicator for differentiating the malignant from benign lesions.

When the regularization parameter decreased, less dependence in the recovered

optical solution occurred with the prior information. The prior estimate was obtained

under the assumption that the whole breast had the same optical properties. As a result,

higher regularization will lower the contrast between tumor and background leading to

smaller separation in the mean HbT and TOI contrast between the malignant and benign

diagnostic groups at a regularization of 1. On the other hand, a high regularization also

suppresses the high spatial frequency variation in the recovered optical property

distribution resulting from measurement noise, and produced lower variation in HbT and

TOI within the malignant and benign groupings. A compromise between separation of

the mean value and noise driven variability was obtained by applying the optimal

regularization approach during image reconstruction.

The effects of optimal regularization on the recovered optical image when both

amplitude and phase data were used (relative to only amplitude measurements), was also

investigated. In this case, the regularization algorithm failed to find an optimal value

based on L-curve analysis, apparently because the relative contribution of phase data

noise was higher than amplitude data noise, requiring much higher regularization of the

phase data. Here, separate regularization values for phase and amplitude data improved

the resulting image outcomes. Specifically, improvement occurred when regularization of

the amplitude data was obtained through L-curve analysis, and regularization of the phase

45

data was fixed at 2 (~1 order of magnitude higher than the value used in the amplitude

regularization). With this approach, improvement in diagnostic performance in terms of

higher AUC for both HbT and TOI occurred relative to using fixed regularization. When

the phase data was added to the Jacobian matrix for image reconstruction, the optimal

regularization (with separate but fixed phase regularization) achieved maximal separation

of the malignant from benign diagnoses (highest AUC). Finally, the optimal

regularization generated the best AUC value (0.94) relative to the other regularization

choices considered when HbT and TOI were combined as the diagnostic indicator.

Although the best diagnostic performance occurred using both amplitude and

phase data, the three regularization approaches, either fixed or optimal, were still able to

separate the malignant and benign groups in terms of HbT (p<0.05), when applying only

amplitude data for image reconstruction. These results suggest the possibility of

simplifying the MRI/NIRS system into one with only CW channels, which would

significantly reduce cost while not sacrificing much in terms of diagnostic performance.

Table 3.3. Comparison of the three regularization approaches in all clinical exams, relative to selected exams when the optimal regularization parameter was found. The group of all exams included 16 malignant and 9 benign pathology-confirmed diagnoses whereas the selected exams included 15 malignant and 7 benign cases (3 exams from the former did not have optimal regularization at the 1st iteration). Both amplitude and phase data were used during image reconstruction.

aSignificant difference.

In the results presented here, data acquired in 22 out of 25 patient exams had an

optimal regularization parameter identified by L-curve analysis in the 1st iteration and

were included in the statistical analyses. For completeness, we compared outcomes of the

46

optimization algorithm in all 25 patients in Table III in terms of the p-value and AUC for

HbT and TOI contrast when applying the three regularization approaches. Both amplitude

and phase data were used for image reconstruction in these results. For the subset of

selected patients (22), optimal regularization provided the best AUC for both TOI and

HbT, relative to fixed regularization, but only for HbT when all exam data were

evaluated. Thus, the L-curve approach may be a pragmatic way to identify which

clinically acquired data sets are reliable. NIRS data can be compromised by a number of

factors during its acquisition, such as fiber-tissue coupling or reflections; hence, the

ability to objectively and conclusively determine which data sets will not lead to accurate

spectroscopic parameter recovery could be important in clinical practice.

3.2.8 Conclusion

In this section, a robust optimization algorithm for selection of the regularization

to be applied during image reconstruction based on exam-specific data acquired during

MRI/NIRST examination of the breast was developed. The optical contrast values for

HbT, StO2, TOI, and scattering parameters were estimated by applying the optimization

algorithm when amplitude only and both amplitude and phase data. A statistical

difference (p<0.05) occurred between malignant and benign groups for both absorption-

derived contrasts (HbT and TOI) as well as reduced-scattering derived contrast (SP).

To the best of our knowledge, these results represent the first time an extensive

study of regularization has been conducted on a relatively large amount of clinical breast

exam data with the MRI/NIRST multi-modality imaging approach. The optimization

algorithm better differentiated malignant from benign cases compared to a fixed

regularization parameter. The best diagnostic performance occurred with optimal

47

regularization values selected from the individual’s exam data, and when combining HbT

and TOI estimated from both amplitude and phase data as the diagnostic indicator.

3.3 Direct regularization from co-registered anatomical images for MRI-guided

NIRST image reconstruction

The image reconstruction approach presented in section 3.2 is an indirect two-step

procedure [107]. First, high resolution anatomical images provided by MRI were

segmented into a small number of sub-domains with assumed homogeneous or constant

optical properties representing the major tissues types. For breast imaging, tissues can be

segmented into adipose, fibroglandular and suspicious regions. Next, the prior structural

information from MRI imposed a hard constraint on the image reconstruction process.

Since the optical properties within an identified region were forced to be uniform, the

constraint was often named as a “hard prior”. The notable advantage of using a hard-prior

scheme is the dramatic reduction in the total number of unknowns alleviating the ill-

posedness of the inversion by reducing the number of unknowns to the few identified

homogenous volumes. This process has the peripheral benefit of significantly enhancing

NIRST accuracy within the localized regions. However, its stability and accuracy are

critically dependent on the accuracy of the structural priors derived from the co-registered

image, and the performance is degraded when incomplete or distorted structural priors

are employed. The choice of regularization parameter is still critical in the case of hard

prior reconstruction due to the diffusive nature of photons in turbid media, though the ill-

posedness is significantly improved. As discussed in section 3.2, an L-curve based

optimization algorithm was developed for hard-prior NIRST image reconstruction, which

improved the accuracy of the image reconstruction. However, the hard-prior

reconstruction requires manual segmentation from MRI images, which is time consuming

48

and requires expertise in breast radiology. To address this challenge, a soft-prior image

reconstruction method without the requirement of image segmentation has been

developed.

Schemes based on “soft priors” do not require optical-property boundaries to

coincide with the MR-defined boundaries; therefore, they allow changes across

boundaries, and reduce the likelihood that spatial biases will be introduced during the

inversion process. Other methods that encode some uniformity into the inversion are also

possible such as total variation minimization or Laplacian smoothing. However, the

traditional hard and soft prior approaches that have been tested require user input to guide

the image segmentation involved [53, 56]. Unfortunately, segmentation can be time

consuming for the user, especially when the tissues of interest are large, and is prone to

errors, for example, when identifying tumor boundaries if the radiological or anatomical

training of the user is not sufficient. Thus, a direct reconstruction method, which

implicitly incorporates the anatomical information into the inversion problem without

user intervention, would dramatically reduce processing time and expand the potential of

multimodal imaging such as MRI-NIRST by fully automating the image reconstruction

process. In this section, a Direct Regularization Imaging (DRI) method for MRI-guided

NIRST was developed. The performance of DRI method was compared with that of hard-

prior reconstruction using L-curve based optimization of regularization parameter

through patient data.

Figure 3.8 shows MRI images of this subject. The left image (a) displays the

Nirview, 3D surface rendering from the T1 image volume where the fiber locations are

evident from the tissue depressions of the breast surface and the fiducial markers. The

49

middle image (b) shows a representative MR image slice from the standard T1 sequence.

The tumor was not localized in this view but the fibroglandular (center, dark part) and

adipose (bright) tissue compartments are readily visible. The right image (c) is a DCE-

MR image. The lesion displayed wash in/wash out contrast enhancement kinetics and

was bright in this image data. The contrast of the grey scale value of the tumor to

surrounding normal tissues was approximately 1.4.

Figure 3.8. MR images from a patient with a malignant lesion (20mm27mm33mm) seen on DCE MRI. (a): Screenshot of the Nirview 3D surface rendering of the T1 MRI. Fiducial markers and fiber bundle positions are shown; (b): Standard T1 image; and (c): Dynamic contrast-enhanced MRI.

Hard prior reconstruction with L-curve based optimization of regularization

parameter was applied and the reconstructed images are shown in Fig. 3.9(b). Figure

3.9(c) shows reconstructed HbT images using DRI method, with λ=1 and σg=0.001.

Though the segmentation is not required for our proposed approach, to validate the

accuracy of the reconstructed tumor site, we compared to segmented images as references

which are shown in Figure 3-9(a) where the tumor was segmented from the T1 and DCE

images in these planes [55]. Visualization thresholds were chosen to suppress the optical

backgrounds. The optical images reconstructed with DRI method (c) exhibited good

agreement on tumor location relative to the segmented images. The estimated HbT in the

tumor region was higher than the surrounding normal tissue, and suggested that the tumor

was malignant. The recovered HbT contrast of the tumor to the normal surrounding tissue

50

reconstructed by hard prior and DRI was 2.6 and 3.6, respectively. The average optical

image size differences relative to the segmented tumor from DCE-MRI was 6% for the

DRI method.

Figure 3.9. The reconstructed HbT images overlaid in three planes with x=-100.0, y=-19.8 and z=-26.6 respectively. (a) Segmented images from corresponding T1 and DCE images. Optical images reconstructed by hard-prior reconstruction using L-curve based optimization of regularization parameter (b), and DRI with λ=1 and σg=0.001 (c), respectively. This figure has been modified from [79].

To conclude, a direct inversion matrix regularization approach from coregistered

anatomical images has been proposed and studied for MRI-guided NIRST. In this new

methodology, the gray-scale image information from coregistered DCE-MRI is encoded

directly into the inversion matrix regularization during NIRST image reconstruction

without any requirements for user intervention such as image segmentation. The method

51

was also tested on in vivo breast data acquired by our combined NIRST/MRI imaging

system. Compared to the hard-prior reconstruction using L-curve based optimization of

regularization parameter, the contrast of the tumor to the normal surrounding tissue

increased from 2.6 to 3.6.

3.4 Discussions

Compared with standard “no prior” reconstruction, both hard prior and soft prior

reconstruction encode the high-resolution spatial information into the inversion

procedure. However, due to the increasing complexity from data fusion of multi imaging

modalities, more attention need to be paid to the inversion, especially regularization. In

this chapter, the image reconstruction of MRI-guided NIRST was optimized in two ways,

with optimal regularization strategy using “hard prior” reconstruction and DRI

reconstruction, respectively. The “hard-prior” information allows the inversion problem

to be reduced in size by lumping regions together into just a few super pixels or regions.

However, the segmentation required by the “hard prior” reconstruction adds further

complexity to the processing and reduces objectivity when combining the image

information. Besides, an L-curve based algorithm has been developed for the choice of

regularization parameter, which yields better differentiation between malignant and

benign lesions, than using fixed regularization parameter. It has the benefit of robust

recovered contrast and stabilized matrix inversion.

In the case where accurate segmentation of MRI images was difficult to achieve,

the DRI method can be an ideal alternative for regularizing NIRST image reconstruction

using spatial priors of another imaging modality. The implicit assumption in DRI is that

the gray scale anatomical image contains structural information which should influence

52

the NIRST parameters, which is plausible for blood-based contrast such as hemoglobin.

However, recovery of water, lipid and scattering values may not directly correlate with

the types of gray scale structures evident on DCE-MRI. Utilizing a range of MRI scans

selected to better map these parameters may be possible. For example, diffusion MRI

could potentially best match with water images from NIRST, and T1 MRI might best

match with lipids and scattering parameters. Thus, multiple DRI regularizations could be

associated with multiple NIRST parameters and is an approach that requires further study

to determine its advantages.

53

Chapter 4: A Hybrid Frequency-Domain/Continuous-Wave NIRST

System with Simultaneous Measurements at Twelve Wavelengths

4.1 Introduction

Near infra-red spectral tomography (NIRST) systems typically have source-

detector schemes that include either frequency-domain (FD) [108], continuous-wave

(CW) [109], or time-domain (TD) [110, 111] data acquisition. FD measurements using

intensity-modulated sources are extremely stable and cost effective, but have limited

working range of wavelength since the response of the available photomultiplier (PMT)

detectors drops dramatically above 825nm. As a result, the ability to achieve accurate

recovery of water and lipid content by FD system alone is greatly weakened, since these

chromophores have characteristic absorption peaks at 975nm and 930nm respectively.

CW systems usually cover a much broader range of wavelength, but lack the ability to

provide patient specific scatter information. Scatter components, including scattering

amplitude and scattering power, have proven to be critical in accurate recovery of other

absorption derived optical parameters, especially in the case of NIRST imaging without

guidance about tumor positions from other imaging modalities [112]. Additionally,

scatter components themselves could be potential biomarkers for differentiating different

breast groups and predicting tumor responses to treatment [113]. Hybrid FD-CW NIRST

systems [3, 51], which take advantage of both FD and CW modules, have proven to

provide spatial reconstruction of both chromophore concentrations of oxy- and deoxy-

hemoglobin, water and lipid, and scatter components of scattering amplitude and

scattering power. Sequential measurement involving multiple wavelengths is time

consuming, and so one approach to speed up the acquisition is simultaneous acquisition

of multiple wavelengths [114], as developed here in tomographic mode.

54

In this Chapter, a portable 12-wavelength FD+CW system was developed.

Section 4.2 gives a brief introduction of the NIRST imaging system and patient exam

settings. The system was developed based on an existing MR-guided NIRST system [3],

but several unique features have been added to the new system for the purpose of

dynamic monitoring of responses to neoadjuvant chemotherapy within the infusion suite.

Compared to our previous stand-alone NIRST approach for monitoring patient responses

to neoadjuvant chemotherapy [2], components in the hybrid system were integrated into a

portable cart. No imaging bed was required, allowing us to acquire data in the clinical

infusion suite. This system provides tomographically reconstructed images of the breast

that can be used to monitor tumor response to neoadjuvant therapy dynamically.

A six-wavelength FD laser source sub-system and six-wavelength CW laser

source sub-system have been developed, as discussed in section 4.3. The combination of

both FD and CW laser sources provides wavelength coverage between 661nm and

1064nm, allowing more accurate recovery of water and lipid. Section 4.4 shows the

hybrid detection module of PMT and PD detectors. Dynamic calibration of PMT and PD

detectors are presented as well.

A major improvement in the hybrid system was the breast interface, designed to

fit different breast sizes and shapes easily. We have shown in chapter 3 that measurement

sensitivity, or tumor coverage, plays a critical role in accurate, spatial reconstruction of

optical properties [56]. The fiber-breast interface has been investigated extensively for

different diffuse optical tomography systems [96]. A common disadvantage of these

breast interface geometries was their lack of mobility and the requirement for the patient

to be positioned prone on a specific imaging bed during data acquisition. For the purpose

55

of monitoring patient response during neoadjuvant chemotherapy, where the intention is

to examine an individual frequently at different time points during treatment, the added

convenience by portable breast interface is significant. In some cases, continuous

measurements are desired for dynamic monitoring of response during the infusion

procedure. A portable NIRST system with corresponding fiber-breast interface is

required to satisfy these conditions. To overcome these challenges, section 4.5 discusses

a supine adjustable optical interface designed to accommodate different breast shapes and

sizes with easy setup. Its performance is validated on phantom, (see Chapter 5) normal

subjects (see Chapter 7) and cancer patient (see Chapter 8).

While several studies [2, 115-117] have focused on hemodynamic changes

throughout the treatment period, dynamic changes during a single infusion procedure are

also of interest. Thus, we adapted the system design to acquire twelve FD and CW

wavelengths simultaneously, which significantly reduced imaging time, as discussed in

section 4.6. Finally, the performance of the hybrid NIRST system was systematically

characterized in section 4.7.

56

4.2 Imaging system and patient exam settings

Figure 4.1. The 12-wavelength FD-CW NIRST system. (a) A photo of the NIRST system. (b) and (c) Recline chair with two groups of fibers in the imaging suite. (d) Adjustable fiber-breast interface. (e) Subject being imaged with the system. (f) Surface image of breast-interface.

Figure 4.1(a) shows the system configuration with its main components housed in

a portable cart. The FD source module consists of six laser diodes (661nm, 730nm,

785nm, 808nm, 830nm and 852nm), modulated by high frequency (~100MHz) signals

generated from a multi-channel RF synthesizer (HS2004, Holzworth Instruments). The

CW source module consists of six laser diodes (850nm, 905nm, 915nm, 940nm, 975nm

and 1064nm), and it is modulated by low frequency sinusoidal signals generated directly

from the data acquisition board (USB 6255, National Instruments). A six-to-one fiber

combiner couples six light signals into a single source signal inside the FD/CW module.

57

As shown in Figs. 4.1(b) and 4.1(c), sixteen bifurcated fiber bundles are grouped

into two plastic tubes, which are attached to the recline chair. The chair was placed inside

the exam/infusion room where nurse helps on the setup of breast-fiber interface, and the

portable cart was placed outside, leaving enough privacy for the patient. The single end

of two groups of bifurcated fiber bundles are attached to the breast through an adjustable

interface designed to fit various breast shapes and sizes (Fig. 4.1(d)). During optical

measurement, the patient sits in a chair, with one side of the breast connected to the

imaging system through the fiber-breast interface (Fig. 4.1(e)). A black sheet covers the

patient to prevent room light from interfering with the data acquisition. The optical

measurements are easily completed in the infusion room given the positioning flexibility

of this portable NIRST system.

58

4.3 Laser source sub-systems

4.3.1 6-wavelength FD source module

Figure 4.2. 6-wavelength FD source module. (a) System diagram; (b) Photo of the sub-system with major components labeled. (c) Photo of the customized 6-wavlength source module and multi-channel synthesizer. Raw data acquired from one PMT detector after heterodyned with reference signal, for f1=100.0004MHz (d), f2=100.0007MHz (e), f3=100.0011MHz, and mixed signal including frequency components at all three frequencies (g).

Figure 4.2(a) shows the system diagram of the house built 6-wavelength FD

source module. The data flow of radio frequency (~100 MHz) electrical signal and light

are represented by blue and red lines, respectively. A multi-channel synthesizer (HS2004,

Holzworth Instruments, USA) provides three RF channels of signals with the same power

of 13dB but at slightly different frequencies of F1=100.0004MHz, F2=100.0007MHz,

59

and F3=100.0011MHz, respectively. It also provides another reference channel at the

frequency of 100MHz for heterodyne detection. The three RF signals are used as inputs

of three single-pole-double-throw (SPDT) RF 1x2 switches, and the six outputs are

combined with six DC current lines through bias-tees to drive six laser diodes. A

customized six-to-one fiber combiner combines the light from six laser diodes into a

single source fiber as FD source output. The six FD diodes are divided evenly into two

sets, one represented by solid blue lines (661, 730nm and 785nm) and the other one by

dash lines (808nm, 830nm and 852nm). At one time, only three laser diodes

(661nm/808nm, 730nm/830nm, and 808nm/852nm) are turned on and modulated by RF

signals through the SPDT RF switches. Figures 4.2(b) and 4.2(c) show the actual layout

of the FD subsystem and multichannel synthesizer, respectively. As shown in Figs. 4.2(d)

to 4.2(g), the detected optical signal acquired from one PMT detector after heterodyned

with reference signal is displayed on an oscilloscope, for F1=100.0004MHz (Fig. 4.2(d)),

F2=100.0007MHz (Fig. 4.2(e)), F3=100.0011MHz (Fig. 4.2(f)), and mixed signal

including frequency components at all three frequencies (Fig. 4.2(g)).

60

4.3.2 6-wavelength CW source module

Figure 4.3. 6-wavelength CW source module. (a) System diagram. (b) Photo of the sub-system with major components labeled. Raw signal acquired from a PD detector with source light modulated at 30Hz (c), 50Hz (d), 80Hz (e), and mixed signal including all three modulated laser sources (f).

A 6-wavelength CW source module is shown in Fig. 4.3. The data flow of low

frequency (<100Hz) electrical signal and light are represented by black and red lines,

respectively (Fig. 4.3(a)). A data acquisition board (USB 6255, National Instruments)

provides three AC voltage outputs at different frequencies of f1=30Hz, f2=50Hz, and

f3=80Hz, respectively. Six laser current drivers (Thorlabs, LD2000R) take the AC

voltage output from data acquisition board, and then provide combined AC+DC current

to drive six laser diodes. Each of the laser current drivers was individually controlled by

one channel from the relay board. The six CW diodes are divided evenly into two sets,

one represented by solid black lines (850, 915nm and 975nm) and the other one by dash

61

lines (905nm, 940nm and 1064nm). At one time, only three laser diodes (850nm/905nm,

915nm/940nm, and 975nm/1064nm) are turned on and modulated by the combined

AC+DC signals generated from the current driver. Figure 4.3(b) shows the layout of the

sub-system with major components labeled. As shown in Figs. 4-3(c) to (f), the detected

optical signal acquired from one PD detector is shown on an oscilloscope with source

light modulated at 30Hz (Fig. 4-3(c)), 50Hz (Fig. 4-3(d)), 80Hz (Fig. 4-3(e)), and mixed

signal including all three modulated source lasers (Fig. 4-3(f)).

4.4 Hybrid PMT-PD detector sub-system

4.4.1 Detector array of 15 PMT and PD detectors

Figure 4.4. Photos of the hybrid detector sub-system. The actual assembly of bottom plate and hybrid detector array are shown in (a) and (b), respectively [3].

A custom programmable mechanical rotary switch hosting PMT (H9305-3,

Hamamatsu, Japan) and PD detectors for FD and CW measurement, respectively, shown

in Fig. 4.4. Fifteen pairs of PMT and PD detectors and one pair of light source couplers

are mounted evenly on the top plate of the rotary stage. Two optical fibers with core

diameter of 800 µm deliver light from the FD and CW source modules into the pair of

light source coupler, respectively. The two ends of each of the sixteen bifurcated optical

fiber bundles are mounted on another plate, which is fixed separately on top of the

rotating circular plate. The circular plate housing the PMT/PD detectors is controlled by a

programmable motor to enable source-detector multiplexing. The two ends of one

62

bifurcated fiber bundles are connected to each of the pair of source couplers, while the

other 15 bifurcated fiber bundles are connected each of 15 pairs of PMT and PD

detectors. The single ends of 16 bifurcated fiber bundles are attached to a breast interface,

to deliver source light to, and collect the diffuse light from the breast (Fig. 4.1(d)). The

rotary switch is incremented 15 times to complete the measurements, yielding a total of

240 (16×15) source-detector combinations. The bifurcated fiber bundles allow the

simultaneous acquisition of both FD and CW data.

4.4.2 Calibration of PMT/PD detectors

The AC amplitude and phase after lock-in detection can’t be directly fitted into

the diffusion model for estimating optical properties. A systematic calibration of the

PMT/PD detectors was completed to standardize the inter-detector data. In addition to

differences in detector responses, other factor, such as fiber loses and rotary switch

coupling errors are corrected during the calibration procedure. Using a central source

location relative to all detector fibers, the amplitude/phase response of each PMT detector

was characterized for every source position [118]. The same procedure was repeated for

all possible combinations of wavelengths, PMT gain settings, and modulation

frequencies. Similarly, the amplitude response of each PD detector was characterized for

every source position, for six wavelengths at corresponding modulation frequencies.

After calibration, changes in amplitude and phase shift for a given source-detector pair

arise largely from photon absorption and scattering in the breast tissue, which allows

accurate reconstruction of desired optical properties.

63

Figure 4.5. PMT calibration. Input power (log10) (a) and phase (degree) (b) versus PMT AC amplitude for different gain settings from 0.5 to 1.1.

For different PMT detectors, an input light signal with the same power might

result in different AC amplitude after lock-in detection. Additionally, AC amplitude can

also vary for the same PMT detector at different gain settings. Figure 4.5 shows the input

power (a) and phase (b) versus PMT AC amplitude for different settings from 0.5 to 1.1.

The actual input power shows a linear relationship with measured AC amplitude, with

different slopes for corresponding gain settings. The phase shift offset was extracted for

each gain, by averaging all the phase data with AC amplitude (log10) in the range

between -2 and -0.5.

64

Figure 4.6. Uncalibrated and calibrated amplitude/phase data. Uncalibrated amplitude (a1) and phase (a2) versus source-detector distance. Calibrated amplitude (b1) and phase (b2) versus source-distance distance. The uncalibrated amplitude and phase data are shown in Figs. 4.6(a1) and 4.6(a2),

respectively. By contrast, the calibrated amplitude and phase are shown in Figs. 4.6(b1)

and 4.6(b2). It’s clearly seen that the amplitude and phase data after calibration show

linear relationship with source-detector distance, as predicted by diffusion theory.

4.5 Adjustable parallel breast interface

4.5.1 Classical fiber-breast interfaces

Most fiber-based diffuse optical imaging systems require optical fibers have good

contact with breasts so as to satisfy assumptions made in the diffusion approximation to

the radiative transfer equation. As a result, breast interface is a critical component which

can significantly affect the quality of reconstructed optical images. Multiple fiber-breast

interfaces have been developed at Dartmouth, corresponding to different sampling

geometries and setups [52, 119, 120]. Three typical fiber-breast interfaces are shown in

65

Fig. 4.7, including circular interface (a), dual-plate interface (b), and triangular interface

(c).

Figure 4.7. Three typical fiber-breast interfaces: (a) circular interface, (b) dual-plate interface and (c) triangular interface.

The circular interface is the most classical interface which is widely used by

various research groups [121]. During patient exam, the patient places her breast through

a hole in the exam bed. Three planes of fibers (16 3) array are controlled by a precision

positioning system underneath a custom-built patient bed in order to fit different breast

sizes. A circular mesh is made with given diameter of the breast, for image reconstruction

and display. The circular geometry makes it straightforward to compare the reconstructed

optical image with coronal view of breast MRI images. The circular interface proves to

work well through a series of clinical trials [2, 122, 123].

Besides circular interface which was designed for optical imaging alone, two

interfaces have been developed for MRI-guided diffuse optical tomography system [96,

119, 124]. The dual-plate interface holds one row of eight fibers on each side of the

breast. Two lift bags can be inflated remotely to raise and lower the imaging plane. Such

66

design allows the imaging plane to be adjusted according to the position of tumor during

a MRI exam, which improves the sensitivity across the complete breast volume without

manually replacing the breast interface. However, this interface is hard to assess small

breasts and tumors close to the chest wall or the nipple area.

To overcome this problem and further improve the breast coverage in MRI-

guided diffuse optical tomography, a triangular breast interface was developed by

Mastanduno et al. [120], and then validated in the clinical trial in Xi’an, China. The

triangular interface divides the 16 fiber bundles into one set of eight, and two sets of four

fiber bundles. Each set is attached to the MRI breast coil, which can slide in both the

medial-lateral direction and the anterior-posterior direction. Before the combined

MRI/optical measurement, patient-specific adjustments were completed by the nurse.

Compared with previous design, the triangular interface provides more freedom on breast

size and tumor position.

4.5.2 Design of adjustable parallel breast interface

Although the interfaces discussed above prove to work well in both phantom and

patient imaging, they are not the ideal candidate for free space breast imaging of breast

cancer response to neoadjuvant chemotherapy. They either are fixed on the imaging bed,

or need to be attached to the MRI breast coil. None of them provide a feasible solution in

the case where patient is imaged while sitting in the chair in the infusion room. To

address this challenge, a parallel optical interface (shown in Fig. 4.8(a)) was developed to

provide robust optical measurements with a flexible patient setup [125]. The interface

consisted of opposing plates with a slight curvature, designed using Solidworks and

fabricated with a three-dimensional printer (Stratasys, Inc., Eden Prairie, MN). The

67

interface was blackened on its exterior to dampen stray illumination from light

reflections. The sixteen optics fiber bundles were divided into two sets of eight, placed in

the same plane, and connected through two slim rods that allowed adjustments to fit

specific breast sizes. Each group of eight fibers was placed along an arc with radius of

101mm, and two adjacent fibers have an angular separation of 4.65 degrees. During a

breast exam, the interface was opened to its maximum extent and then closed, until all

(most) fibers achieved good contact with breast tissue by applying a modest amount of

pressure. The clinical exam attendant positioned the optical interface measurement plane

across the tumor based on prior information from mammography/MRI images. The

position and orientation of the breast interface was also adjusted to maximize intersection

with the tumor, by imaging in one of the mediolateral (ML), mediolateral oblique (MLO)

or craniocaudal (CC) geometries commonly used in mammography. Setup of the breast

interface required 2~3 mins and most participants did not indicate feelings of discomfort.

After the interface was setup, the fibers which did not have good contact with the breast

were noted. The corresponding boundary data was eliminated from the dataset for future

image reconstruction. Separation between the two breast interface sections was measured

and used to make patient specific 2D FEM models. Two breast interfaces with different

curvatures were designed to accommodate various breast sizes and tumor locations. Their

performances are compared in Chapter 6.

68

Figure 4.8. Adjustable parallel fiber-breast interface. (a) Solidworks file showing dimensions of the interface. (b) A soft gelatin breast phantom being imaged with the interface. (c) Corresponding FEM mesh with fibers marked in red circles.

Figures 4.8(b) and 4.8(c) show a soft gelatin breast-mimicking phantom imaged

with the interface and its corresponding 2D FEM mesh, respectively. The interfaced was

placed such that there was modest pressure between the fibers and phantom, and each

fiber was in good contact with the phantom.

4.5.3 Phantom imaging with the parallel breast interface

Figure 4.9 shows phantom experiment setups and reconstructed images of HbT,

StO2, water, lipid, scattering amplitude (SA) and scattering power (SP) for two breast

interfaces with different curvatures. The recovered inclusion/background HbT contrasts

were 1.40 and 1.38, 6.7% and 8.0% different from the actual contrast of 1.5X,

respectively. Both interfaces were able to yield expected background values for StO2

(>95%), water (>90%) and lipid (<5%), from the spectral coverage provided by the

longer wavelengths in the six CW channels.

Surface artifact, or the unexpected enhancement along the mesh boundary, is a

well-known problem in diffuse optical tomography. The second interface generated more

69

surface artifacts than the first one, which partly accounts for the fact that the second

interface recovered lower contrast. Additionally, the first interface produced less

heterogeneity in the recovered images of chromophore concentrations. In general, the

performance of the first interface with deeper curvature was superior in terms of both

reconstruction accuracy and noise level. However, the second interface was preferred

when imaging small breasts because it maintained better fiber-tissue contact under these

conditions.

Figure 4.9. Experimental setup and reconstructed optical images for two heterogeneous phantoms with 1-inch diameter inclusions. The corresponding interface had deep curvature (a) and flat curvature (b). For both phantoms, the blood concentrations inside and outside the inclusion were 1.5% and 1%, respectively.

70

4.6 Simultaneous acquisition at twelve FD+CW wavelengths

4.6.1 Simultaneous acquisition at twelve wavelengths

Figure 4.10. System diagram for simultaneous acquisition. FD source module, CW source module, and data acquisition/processing module are highlighted in blue, green, and violet blocks, respectively. The flow of low frequency electrical signal, high frequency electrical signal, and light is shown by the black, blue and red solid lines, respectively.

Figure 4.10 shows a system diagram for simultaneous acquisition of both FD and

CW measurements [125]. Custom fiber combiners connect the 6 FD and 6 CW lasers.

Although the hardware is capable of delivering and collecting all twelve channels of light

at the same time, best practice divides the twelve channels into two sets, mixing signals

which are maximally separated but can be measured simultaneously. The total light

source power of any six sources used at one time is less than 120mW, and the cross-talk

is less than 0.8% between different channels.

The combined FD and CW light is coupled into two ends of one bifurcated fiber

bundle, whose distal end delivers the illumination consisting of six wavelengths

modulated at six different frequencies to the breast surface. The transmittance light is

71

collected by the single end of the other fifteen fiber bundles. For each of the fifteen

bifurcated fiber bundles, the transmittance light is delivered to pairs of PMT and PD

detectors. To ensure the PDs do not saturate in the presence of high frequency modulated

signals at shorter wavelengths, thin film long pass filters (87C, Kodak) were installed on

the detector windows to block light shorter than 850nm. The RF output from the PMT

detectors is amplified by a 20dB low-noise preamplifier, which also filters out residual

DC components. The output of the preamplifier is heterodyned with a 100MHz reference

signal through a mixer, down-converting it to the lower frequencies 400Hz, 700Hz and

1100Hz. These low frequency signals are amplified (100X) and filtered again to reduce

high frequency noise. The resulting signal is read and processed by a DAQ board (USB

6255, National Instruments), where the phase shift and amplitude are extracted for the

three shorter wavelengths. Since the phase shift data also depends on the initial phase of

each RF output signal from the synthesizer, these signals are passed through RF splitters,

and heterodyned with the 100MHz reference signal, to get the initial phase shifts of three

components, respectively, which are subtracted in the 3 FD channels. Unlike the FD

module, the output of each PD detector is directly connected to the DAQ board. Only the

change in amplitude of light propagating through the scattering medium is extracted at

the three modulation frequencies. The complete measurement dataset consists of

amplitude and phase at six shorter wavelengths (661nm, 730nm, 785nm, 808nm, 830nm

and 852nm), and amplitude at six longer wavelengths (850nm, 905nm, 915nm, 940nm,

975nm and 1064nm), for 240 source-detector combinations.

72

4.6.2 Hybrid gain setting of PMT detector

Figure 4.11. Hybrid gain adjustment of PMT detectors. (a) Flow chart illustrating the hybrid gain adjustment scheme. (b) A photo of the adjustable fiber-breast interface (c) Corresponding football shape mesh created with 16 fibers assigned along the surface. (d) Amplitude data acquired at source position #1 using automatic gain adjustment scheme. (e) Amplitude predicted for the other source-detector pairs, based on the parameters fitted from (d). The actual amplitude and phase data acquired using the gain from the lookup table for the rest of source-detector pairs, shown in (f) and (g) respectively.

In FD measurement, PMT detectors at different positions, relative to the source,

receive levels of light which could be different by orders of magnitude. To account for

the variability, various voltage gains need to be assigned for 15 individual PMT

detectors. Two gain acquisition schemes were used in previous studies [3, 51]. The fixed

gain scheme has been traditionally used for circular geometry, where source-detector

distance does not change for a given detector during the acquisition regardless of the

position of source fiber. However, it does not work for irregular geometries where

source-detector distance varies depending on the position of source fiber. In this case,

73

another automatic gain adjustment scheme was introduced to make sure that the

corresponding gain is found and applied at each of 240 source-detector pairs, which

ensures that the detected light intensity falls in the optimal linear response range. Though

the latter adjustment scheme works for more complicated geometries, it suffers longer

acquisition time and larger signal variation because of the rapid change of gain. To

accelerate the data acquisition and thus reduce the signal noise due to patient movement,

a novel hybrid gain adjustment scheme has been developed, as shown in Fig. 4.11.

Figure 4.11(a) outlines the chart illustrating the hybrid gain adjustment scheme.

First, the separation between two half-moon plates housing fiber bundles was measured

before breast/phantom experiments (Fig. 4.11(b)), loaded into the data acquisition

program, and then a patient/phantom specific mesh was generated automatically (Fig.

4.11(c)). An automatic dynamic gain searching algorithm was applied at source position

#1, and the gain (control adjustable 0 – 1.1 V) of 15 PMTs increased from 0.4 V (in

increments of 0.1V) until the AC component of the amplified output signal reached at

least 0.1V, or the highest possible gain setting was reached. The acquired amplitude times

source-detector distance in log scale was plotted versus source-detector distance for 15

PMT detectors. A linear relationship was obtained (Fig. 4.11(d)), which can be predicted

by diffusion theory. Linear regression was applied for transmittance (black) and

reflectance (red) data, respectively, and regression coefficients were extracted

accordingly. Next the source-detector distance was calculated for the other source-

detector pairs for the given mesh, and amplitude was predicted using the fitted regression

coefficients for transmittance and reflectance data, respectively (Fig. 4.11(e)). Given the

predicted amplitude of each source-detector pair, the optimal PMT gain can be obtained,

74

and a lookup table of optimal gain for all source-detector pairs was generated. Lastly, the

lookup table was applied during the data acquisition at source position #2 to #16, and

both amplitude (Fig. 4.11(g)) and phase data (Fig. 4.11(h)) were acquired.

Figure 4.12. Standard deviation of amplitude (a) and phase (b) of 30 measurements for two gain adjustment schemes.

Compared with dynamic gain adjustment scheme with automatic gain searching for

each source-detector pair, the hybrid gain adjustment scheme described in this paper

takes significantly shorter time, 55s vs. 90s, to complete acquisition for all 240 source-

detector pairs. A homogeneous gelatin phantom was measured 30 times using two gain

adjustment methods, and the variation (standard deviation) in amplitude and phase data

was shown in Figs. 4.12(a) and 4.12(b), respectively. It’s clearly seen that the hybrid

acquisition scheme has superior performance to the other one, in terms of signal stability

for both amplitude and phase, at all PMT gains from 0.5 to 1.1. The FD data acquired

using the hybrid gain adjustment scheme has an average standard deviation of 1.1% and

0.3 degree in amplitude and phase, respectively, compared with 2.1% and 0.7% using the

previous gain adjustment method.

75

4.6.3 Data acquisition GUI

Figure 4.13. LabVIEW GUI for data acquisition, pre-processing and display.

All the data acquisition/calibration was automated through a LabVIEW program.

As shown in Figure 4.13, the GUI presents acquired real-time amplitude and phase data

at three FD wavelengths, and only amplitude data at three CW wavelengths. The

acquisition was repeated twice to get a complete dataset involving 12 wavelengths.

Besides the regular acquisition routine, the GUI also allows users to select acquisition

mode of sequential or simultaneous, to disable specific laser/detector, to select different

adjustment schemes of PMT’s and so on.

4.7 Systematic characterization of the system

4.7.1 Comparison between sequential and simultaneous acquisitions

PMT detectors at different positions, relative to the source, receive levels of light

which could be different by orders of magnitude. In order to account for the variability,

corresponding gains are set for 240 source-detector pairs via a dynamic automated gain

adjustment algorithm where the PMT gain (control adjustable 0 – 1.1 V) of each source-

detector pair increased from 0.4 V (in increments of 0.1V) until the AC component of the

76

amplified output signal reached at least 0.01V, or the highest possible gain setting was

reached. The dynamic adjustment ensures that the input light intensity falls in the optimal

linear response range. During simultaneous measurements involving 6 frequency

(wavelength) signals, the dynamic gain adjustment algorithm applied gains based on the

AC signal of the 785nm (700Hz) component.

Compared to the sequential measurements recorded by previous system [3], the

new unit is much faster. Previously, a complete set of sequential measurements involving

six wavelengths required 12min, whereas the new simultaneous measurement scheme

only requires about 45 seconds, which is sufficient for monitoring of patient response

during neoadjuvant chemotherapy with adequate temporal resolution, since a typical

infusion procedure takes 2-3 hours. Moreover, reduced acquisition time encourages more

patients to participate in the clinical study.

A silicone phantom was used to compare amplitude/phase data obtained with

simultaneous versus sequential acquisition. The average relative differences between the

two measurement methods in amplitude and phase was 0.8% for intensity and 0.6 degrees

in phase, for the 661nm channel, 1.0% and 0.8 degree for the 785nm channel, and 1.1%

and 0.9 degree for 826nm channel, respectively. Relative differences were found to be

0.6%, 0.5%, 0.7%, 0.6%, 0.8%, and 1.0% for the CW wavelength channels of 850nm,

905nm, 915nm, 940nm, 975nm, and 1064nm, respectively. These results demonstrate the

data quality of simultaneous recording is essentially equivalent to sequential acquisition.

77

Figure 4.14. Reconstructed optical images for the same heterogeneous phantom with a 1-inch diameter inclusion. The optical images were reconstructed using boundary data acquired from sequential measurement (a), and simultaneous measurement (b), respectively. The blood concentrations inside and outside the inclusion were 2% and 1%, respectively.

Figure 4.14 shows the reconstructed optical images using boundary data acquired

from sequential measurement (a) and simultaneous measurement (b). The average

difference between two measurements is less than 2.5% for all chromophores. There is no

significant difference between the reconstructed images using two groups of

measurements, as shown in Fig. 4.14.

4.7.2 Variation of phase and amplitude data

Figure 4.15. Standard deviation of phase (a) and AC amplitude (b) versus AC amplitude for different gain settings from 0.7 to 1.1.

We also studied in detail the stability of the data acquisition in terms of standard

deviation of phase and amplitude. As shown in Fig. 4.15(a), standard deviation of phase

is plotted versus measured AC amplitude with different gain settings from 0.7 to 1.1. At

78

the same gain, standard deviation of phase increases as AC amplitude decreases, due to

the decrease of SNR. While at the same AC amplitude, standard deviation of phase data

increases as gain increases, since noise is amplified more while signal does not vary. A

typical plot of measured AC amplitude versus PMT number is shown in the middle upper

of Fig. 4.15(a). PMT #6 to #10 have high gain of 1.0 to 1.1, and the other PMT detectors

have relative gain from 0.7 to 0.9. High gain and low gain groups of detectors are

classified, showing that detectors with high gains yield in higher standard deviation in

phase data than the other group with low gains. The AC amplitude has similar noise

pattern (Fig. 4.15(b)), with standard deviation of phase as high as 3% for high gain

settings.

79

Chapter 5: Tissue Simulating Phantoms for NIRST Imaging

5.1 Introduction

Tissue mimicking phantoms, which have similar spectral properties to soft tissues,

have been intensively investigated in the past decades and many useful models have been

created [126-131]. An excellent overview of tissue simulating phantoms for optical

spectroscopy, imaging and dosimetry can be referred to Pogue and Patterson [132].

Characteristic tissue phantoms have proven indispensable in various optical imaging

modalities such as near infrared spectral tomography, photodynamic therapy [133],

luminescence imaging [134], fluorescence molecular imaging [135] and optical

coherence tomography [131, 136]. In general, these tissue phantoms are used in several

ways. First, they can be used for the purpose of routine quality control, where an imaging

system is tested with the phantom on a regular basis. Second, phantom can be used to

validate the system performance, when the output data/image of a phantom is compared

with the true value associated with the given phantom. Besides, they are also used for

calibrating raw measurement data as the first step of data/image processing. In NIRST

image reconstruction, a homogeneous phantom is usually imaged before patient imaging,

to calibrate any unexpected measurement errors such as thermal shift in lasers and

coupling error between optical fiber and detectors.

Tissue phantoms are usually made of matrix/base, scattering and absorption

materials, the choices of which correspond to different benefits and weaknesses. Matrix

materials provide an overall base, which decides physical/chemistry properties of the

phantom. Common matrix materials include water, gelatin/agarose, room-temperature

vulcanizing (RTV) silicone, resin, and polyvinyl alcohol gels. To mimic the diffusive

80

nature of biological tissues, a large scattering coefficient is desired for the tissue

phantom. There are three major types of scattering materials: lipid emulsion, polymer

micro particles, and white oxide powders. The choice of absorbers varies from ink,

molecular dyes and blood to provide relatively stable absorption at single wavelength, to

oxy- and deoxy-hemoglobin and cells to provide spectral features at multiple

wavelengths in the NIR range. Note for given matrix/base matrix materials, there is only

limited number of choices for both scattering and absorption materials. Over the past

decade, there have been several types of tissue phantoms developed and tested in the

optics lab at Dartmouth. The three most popular ones are gelatin phantom, resin phantom

and RTV silicone phantom, outlined in Table 5.1.

5.2 Comparison between major tissue mimicking phantoms

Table 5.1. Comparison of major tissue-mimicking optical phantoms.

Gelatin phantom uses water and agarose/gelatin powder as base materials. As one

of the main water based phantoms, it can use any of the three major scatters mentioned

above. Intralipid has become the most widely used lipid emulsion, which is well

calibrated and commercially available. The optical properties of Intralipid has been

systematically characterized by Staveren et al, who also proposed a simple power law for

the wavelength dependence of reduced scattering coefficient [86]. The absorption of such

phantom mainly comes from water in the NIR range, and is extremely low (<0.002mm-1).

81

Different absorbers can be added to add spectral features and tailor the absorption

spectrum. In addition, fluorophores can be added into the gelatin phantom as well in the

case of fluorescence imaging.

A unique feature of gelatin phantom is its capability of including whole blood as

absorbers to mimic the spectrum of soft tissues in NIR range, where the dominant

absorbers are hemoglobin and water [137]. The addition of blood as absorbers in other

matrix materials is challenging since blood does not mix well with either resin or

silicone. Notice that it’s critical to use saline instead of distilled water to preserve the

oxygen-binding function of hemoglobin. Gelatin/Agarose powder does not dissolve in

water solution under room temperature, and therefore the water needs to be heated to

boil. The gelatin phantom will then take shape after the solution cools down. Despite the

advantages of gelatin phantom, it does not last for a long time, and should be imaged

within the same day that the phantom is made. Another drawback of the gelatin phantom

is the likely contamination when the fiber/detector is in contact with the phantom. The

leakage of water from the gelatin phantom can cause serious problems when the phantom

is directly in contact with various photon detectors.

Figure 5.1. Major tissue mimicking phantoms developed at Dartmouth. From left to right, gelatin phantom with ink (a), gelatin phantom with blood (b), resin phantom (c), RTV silicone phantom (d) and silicone soft gel phantom (e) are presented, respectively.

82

Polyester resin phantoms were first introduced by Firbank, Delpy, and Oda, who

used both TiO2 [138] and polystyrene particle scatters [139]; Room-temperature-

volcanizing (RTV) silicone-based soft phantoms were introduced by Bayes et al [140]

and Beck et al [141]. The most common materials of constructing resin phantom are

resin, hardener, ink and TiO2. The addition of hardener into the resin helps create a

transparent solid resin, which usually takes at least 24 hours to cure. The mixed materials

need to be degassed in vacuum for 3-4 times to prevent any bubbles inside the eventual

phantom. A detailed outline of this procedure can be referred to [142]. The procedure of

making RTV silicone based phantom is similar to that of resin phantom, except that

RTV-based compounds are required instead of resin. The curing procedure typically

takes 5-6 days for RTV silicone phantom, depending on the volume of hardener, which

takes much longer than the 24-hour curing time of the resin based phantom. Both resin

and RTV-based silicone phantoms may last many years with relatively stable optical and

mechanical properties [143], and therefore they are the ideal candidates for repetitive

studies.

Compared with resin based phantoms, the major advantage of RTV-based

phantoms, is that the stiffness can be well controlled by lowering the hardener

concentration, which gives the RTV-based silicone phantom more flexibility than resin

based phantom. In practice, the stiffness of RTV-based silicone phantom was close to

that of soft tissue when the hardener concentration was around 3.4% [132]. Unlike resin

based phantoms where the interior cavities and exterior shape can be easily machined,

RTV-based phantom can’t be machined with complicated geometries.

83

Gelatin phantom, resin phantom, and RTV-based silicone phantom have been

extensively studied and tested at Dartmouth, with corresponding strengths and

weaknesses. Based on previous experience, a new method of making tissue mimicking

phantom is proposed in this chapter. The characterization of the proposed phantom is

discussed in section 5.3.

5.3 Phantom Preparation

5.3.1 Preparation of homogenous silicone soft gel phantom

Figure 5.2. The preparation of silicone soft gel phantom. (a) Base materials of A-341: Silicone Soft Gel. (b) Silicone coloring materials which are used as absorber/scatter. (c) The mixing of base and silicone coloring materials.

This soft gel tissue-mimicking phantom has been inspired by the masks used in the

movie industry, which have similar optical/appearances to human soft tissue. As shown

in Fig. 5.2(a), A-341: Silicone soft gel (Factor2, USA), translucent and low-viscosity

RTV, was used as the base material. The phantom was made by mixing base A and

catalyst B at a ratio of 10:1. Under room temperature, it only takes less than 1 hour to

cure. This is much shorter than that required by the RTV based silicone phantom

discussed in section 5.2. Since neither typical scatters (TiO2/Intralipid) nor absorbers

84

(ink/blood) dissolve in the mixed solution of base A and catalyst B, special silicone

coloring materials (FI-SK, functional intrinsic skin colors, Factor2, USA) were used as

scatters and absorbers. These coloring materials are blend of FD&C cosmetic pigments,

and can be crushed into a silicone crosslinking fluid (Fig. 5.2(b)). The white and pink

coloring materials were added into the mixed base materials, to get desired absorption

and scattering properties. Figure 5.2(c) shows the preparation of silicone soft gel

phantoms. The leftmost container shows the mixed solution after stirring, while the other

containers correspond to different steps of the preparation process. A recipe and mixing

procedure for making silicone soft gel phantom is listed below.

1. Add 50g base A of A-341 into the container.

2. Add white and pink coloring materials.

3. Stir for 3 minutes with an automated stir bar.

4. Add 5g catalyst B of A-341.

5. Stir for 3 minutes.

6. Pour the mixed solution into mold for setting.

7. Wait for approximately 45 minutes for the phantom to cure

The above procedure is significantly simplified compared with that of making

gelatin and traditional resin/RTV phantoms, since it does not require either vacuum or

microwave. Besides, the whole procedure can be completed within one hour.

5.3.2 Preparation of heterogeneous silicone soft gel phantom

The preparation of homogeneous silicone soft gel phantom was discussed in

section 5.3.1. In this section, heterogeneous phantoms with similar size, shape, optical

and mechanical properties to breast tissues with tumor-mimicking inclusions, were

85

developed using the recipe introduced in the earlier section. Figure 5-3 shows the detailed

steps of making breast mimicking phantoms with sphere shape inclusions. First, 1000ml

base materials were mixed with coloring materials in a stand mixer for 10 minutes to

provide background in the eventual heterogeneous phantom, shown in Fig. 5.3(a).

Meanwhile, 3D printed molds (Fig. 5.3(b)) were filled with mixed solutions which have

different optical properties from the background to provide tumor-mimicking inclusions

in the eventual breast mimicking phantom. Figure 5.3(c) shows sphere-shape inclusions

after curing. The sphere-shape inclusions were fixed inside a large mold (Fig. 5.3(d)),

which matches the 2D cross-section of the fiber-breast interface. The relative position of

the sphere, in terms of height and distance to the boundaries, was recorded as the ground

true. The mold was then filled with mixed solution prepared in the first step (Fig. 5.3(a)).

Figure 5.3(e) shows the heterogeneous breast mimicking phantoms with heterogenous

inclusions. Besides sphere-shape inclusions, cylindrical cavity can also be created inside

the phantom. A 20ml syringe was placed vertically inside the large mold, and the

background was filled with mixed solution. The syringe was taken out after the mixed

solution cured, resulting in a cylindrical cavity, which can be later filled with liquid

solutions with various optical properties.

86

Figure 5.3. Detailed steps of making breast mimicking phantoms. (a) The base material was mixed with coloring materials in a food mixer. (b) The mixed solution was poured into 3D printed molds. (c) Three sphere-shape inclusions were taken from the molds after curing, with radius of 6mm, 9mm and 12mm, respectively. (d) One sphere-shape inclusion was fixed inside a large mold, which was filled with mixed solution later. The optical properties of the inclusion are different from those of the background, in order to create inclusion/background contrast. (e) A group of heterogeneous breast mimicking phantoms, with either sphere shape inclusion inside, or cylindrical cavity.

5.4 Characterization of homogenous silicone soft gel phantom

A series of small homogenous phantoms (Fig. 5.2(c)) were made with varying

optical properties, for systematic characterization of the new silicone soft gel phantom.

The absorption and reduced scattering coefficients in the NIR range were measured using

a diffuse optical spectroscopic imaging (DOSI) system, developed at University of

California, Irvine [40]. The DOSI system performs high-resolution spectroscopy from

650 to 1000nm, which combines FD measurements at four wavelengths and broadband

spectroscopy. During measurement, a hand-held probe was placed on the upper surface of

the phantom with modest pressure, and there was a fixed separation of 22mm between

the source and detector fibers. FD measurements were performed sequentially at four

wavelengths. Both amplitude and phase responses were measured when the modulation

frequency of laser was scanned from 50 to 400MHz, as shown in Fig. 5.4(a). The reduced

scattering coefficients were fitted using Mie theory, as shown in Fig. 5.4(b). Meanwhile,

the high-resolution absorption spectrum was obtained with fitted scattering coefficients

87

and measured broadband spectroscopy (Fig. 5.4(c)). There are no obvious spectral

features in the absorption spectrum, except the absorption peak centered around 905nm,

which mainly comes from the absorption of water in the base materials.

Figure 5.4. DOSI measurement of a silicone soft gel phantom. (a) Measured (blue points) and fitted (red line) amplitude and phase at four wavelengths, while the laser modulation frequency was scanned from 50 to 400MHz. (b) Measured s at four wavelengths (red points) and fitted scattering spectrum (blue line) using Mie theory. (c) Fitted broadband absorption spectrum.

To validate the reproducibility of the silicone soft gel phantoms, five silicone soft

gel phantoms were made using the same recipe, and each phantom was measured 10

times at randomly selected positions on the phantom. The fitted a and s at 661nm are

plotted with error bars representing the standard deviation among different

measurements, for five phantoms, as shown in Figs. 5.5(a) and 5.5(b), respectively. The

average variation within the same phantom is 2.4% and 0.8% for a and s , respectively.

The variation among different phantoms, defined as standard deviation versus mean

value, is 1.0% and 1.4% for a and s , respectively.

88

Figure 5.5. The fitted a (a) and s (b) at 661nm are plotted with error bars representing the standard deviation among different measurements, for five phantoms. Five silicone soft gel phantoms were made using the same recipe, and each phantom was measured 10 times at randomly selected positions on the phantom, using the DOSI system.

Ideal tissue mimicking phantom should last for long time with relatively stable

optical properties. Figure 5.6 shows the a and s measured at 661nm of a silicone soft

gel phantom over 10 days. The phantom was measured at multiple time points during a

10-day period, and 10 measurements were performed at each time point. The fluctuation

(dev./average) is less than 1.5 % of both a and s , within the range of the imaging

accuracy of the DOSI system.

Figure 5.6. Measured a (a) and s (b) at 661nm of one phantom in 10 days. Each time the same silicone soft gel phantom was measured 10 times at randomly selected positions on the phantom, using the DOSI system.

89

Another important metric evaluating tissue mimicking phantom is the precise

control of optical properties by adjusting the amount of absorbers and scatters. The

optical properties of the silicone soft gel phantom can be controlled through a

combination of white and pink coloring paints. Two groups of seven phantoms were

made, using 50g of base A and 5g of catalyst B as the base material for each phantom.

0.3ml and 0.8ml of white paint were added into each of the seven phantoms in the 1st and

2nd group, respectively. An increasing amount of pink paint, from 0.1ml to 0.7ml with an

increment of 0.1ml, was added into corresponding phantom in each group. Figure 5.7

shows the measured a (Fig. 5.7(a)) and s (Fig. 5.7(b)) versus pink paint concentration

for the group of 0.3ml (blue points) and 0.8ml (black points) white paint, respectively. A

linear regression was fitted for each group of measurement data, with R2 higher than

0.98. It can be clearly seen that the addition of pink paint into the base material affects

both a and s linearly. One interesting fact is that the concentration of white paint

affects the slope of the linear regression. With a higher concentration (0.8ml) of white

paint added into the based material, the fitted slope is lower (0.0072 vs. 0.0087) than that

of lower concentration (0.3ml) for a . Meanwhile, the fitted slope is higher (1.36 vs.

0.96) of the group with more white paint for s . Such behavior suggests that the mixing

of two types of coloring paint may cause the aggregation of absorbers/scatters existing in

the paint.

90

Figure 5.7. Measured a (a)) and s (b) are plotted versus pink paint concentration for the group of 0.3ml (blue points) and 0.8ml (black points) white paint, respectively. Two groups of seven phantoms were made, using 50g of base A and 5g of catalyst B as base material for each phantom. 0.3ml and 0.8ml of white paint were added into each of the seven phantoms in the 1st and 2nd group, respectively. An increasing amount of pink paint, from 0.1ml to 0.7ml with an increment of 0.1ml, was added into corresponding phantom in each group. Each phantom was measured 5 times.

Similar procedure was repeated to investigate the effect of white paint

concentration on the optical properties of the silicone soft gel phantom. Two groups of

seven phantoms were made, using 50g of base A and 5g of catalyst B as base material for

each phantom. 0.3ml and 0.5ml of pink paint were added into each of the seven phantoms

in the 1st and 2nd group, respectively. An increasing amount of white paint, from 0.5ml to

1.1ml with an increment of 0.1ml, was added into corresponding phantom in each group.

Figure 5.8 shows the measured a (Fig. 5.8(a)) and s (Fig. 5.8(b)) versus white paint

concentration for the group of 0.3ml (blue points) and 0.5ml (black points) pink paint,

respectively. As we can see from Fig. 5.8(a), the variation of a among seven phantoms

with different concentrations of white paint is 1.1% and 1.4% for the group of 0.3ml

(blue points) and 0.5ml (black points) pink paint, respectively. This suggests that the

addition of white paint does not alter a of the silicone soft gel phantom. On the other

side, s presents linear dependence on the concentration of white paint, with R2 higher

91

than 0.99 for both groups of phantoms. Similarly, the slope of linear regression does

depend on the amount of pink paint in the phantom. With higher concentration (0.5ml) of

pink paint added in the based material, the fitted slope is lower (0.38 vs. 0.54) than that of

lower concentration (0.3ml).

Figure 5.8. The measured a (Fig. 5-8(a)) and s (Fig. 5-8(b)) are plotted versus white paint concentration for the group of 0.3ml (blue points) and 0.5ml (black points) pink paint, respectively. Two groups of seven phantoms were made, using 50g of base A and 5g of catalyst B as base material for each phantom. 0.3ml and 0.5ml of pink paint were added into each of the seven phantoms in the 1st and 2nd group, respectively. An increasing amount of white paint, from 0.5ml to 1.1 ml with an increment of 0.1ml, was added into corresponding phantom in each group. Each phantom was measured 5 times.

To conclude, the optical properties of soft gel phantoms have been characterized

in this section. The absorption and scattering properties can be controlled through

adjusting the amount of white and pink paints added inside the base materials. More

specifically, the addition of white paint affects a linearly, while the addition of pink

paint affects bot a and s linearly.

5. 5 Validating the performance of the NIRST system using heterogeneous breast

mimicking phantoms

5.5.1 NIRST imaging of heterogeneous phantoms at different depths

The effects of choice of imaging plane on the NIRST image reconstruction have

been intensively investigated by Wang and Mastanduno et al. The position and

92

orientation of fiber-breast interface is critical to get reliable measurement, since the 2D

fiber array has been used to sample the breast tissue in 3D space. Mastanduno [56]

showed that only the optical measurement with enough tumor sensitivity, can be used for

reliable reconstruction of optical images, based on the patient data acquired from the

Xi’an clinical trial. Wang showed that in NIRST the quantification of inclusion to

background contrast highly depends on the position of imaging plane, or fiber interface,

using experiments data of gelatin phantom with cylindrical inclusions.

Gelatin phantom with cylindrical inclusion provides a good solution to mimic

tumor/background contrast in breast. However, sphere-shape inclusion is naturally a

better estimate to the actual breast tumor than cylindrical inclusions. The new silicone

soft gel phantoms with sphere inclusion were imaged at different depths, to better

characterize the performance of the 12-wavelength NIRST system.

Figure 5.9. Phantom experiments using a silicone soft gel phantom with a sphere inclusion. (a) Reconstructed absorption images from the data acquired at different depths of 0mm, -3mm, -6mm, -9mm, and -12mm, respectively. The depth of 0mm corresponds to the case where imaging plane was placed across the center of the sphere inclusion. (b) Profiles of the

93

reconstructed a along the X-axis, crossing the center of the inclusion projected on the surface, at 830nm. (c) The sphere has a diameter of 24mm. The actual inclusion/background contrast is 2.

Figure 5.9(a) shows the reconstructed absorption images using the data acquired at

different depths of 0mm, -3mm, -6mm, -9mm and -12mm, respectively. The 0mm plane

corresponds to the case where the fiber-breast interface was placed across the center of

the sphere inclusion, which has a diameter of 24mm. We can see that the reconstructed

image at 0mm plane has the highest inclusion/background contrast, which corresponds to

the highest maximum a (blue squares) as shown in Fig. 5.9(b). When the imaging plane

was placed 6mm away from the 0mm plane, the reconstructed absorption image gives

much lower inclusion/background contrast, and the maximum a (red crosses) is 16.6%

lower than that of 0mm plane. The a images acquired at 3mm and -3mm plane provide

similar inclusion/background contrast, together with similar maximum recovered a .

Table 5.2. Maximum recovered a (10-3/mm) at different depths for phantoms with sphere shape inclusions with diameter of 12mm, 18mm and 24mm.

Similar depth varying measurements were performed for breast mimicking

phantoms with sphere inclusions of a diameter of 12mm and 18mm as well. The

maximum revered a (10-3/mm) extracted from reconstructed optical images at different

depths are compared in Table 5.2. When a small sphere (d=12mm) was included in the

breast phantom, the maximum reconstructed a is limited, even at the center plane with

94

depth of 0mm. The maximum recovered a , obtained from the reconstructed optical

image of the phantom with a sphere inclusion of a diameter of 24mm, is 16.7% higher

than that of a diameter of 18mm, at the same depth of 0mm. In addition, the group of

measurements (depth at 3mm, 6mm, and 9mm) with imaging plane placed higher than

the baseline (depth at 0mm), has reasonable symmetry with those with imaging plane

placed lower than the baseline (depth at -3mm, -6mm, and -9mm).

The phantoms experiments presented in this section show that the quantification

of inclusion/background contrast in NIRST highly depends on the position of imaging

plane/fiber interface, especially in the case of imaging small inclusions. Furthermore, the

proposed heterogeneous silicone soft gel phantom with sphere shape inclusion is an ideal

candidate for characterizing the performance of NIRST imaging systems.

5.5.2 NIRST imaging using partial transmission/reflectance data

Various sampling geometries have been used in different NIRS/NIRST imaging

systems [40, 144]. In general, there are two types of boundary data: reflectance and

transmission. When the sources were placed along the same side as detectors, reflectance

data were acquired. On the contrary, transmission data was acquired when sources were

placed on the opposite side of detectors. The first sampling geometry simplifies the

design of fiber interface and has been widely used in hand held devices [43, 44].

However, a large source-detector separation is desired when a deep tumor/inclusion

needs to be imaged.

The adjustable fiber-breast interface introduced in Chapter 3 acquires both

reflectance and transmission boundary data. However, in this section, NIRST image

reconstruction using partial reflectance/transmission dataset is investigated.

95

Figure 5.10. Phantom experiments using a silicone soft gel phantom with a sphere inclusion, which has a diameter of 24mm. Reconstructed absorption images from the data acquired at different depths of -9mm, -6mm, -3mm, 0mm, 3mm, 6mm, 9mm, and 12mm respectively, using (a) both reflection and transmission data; (b) both sides of reflectance data; (c) one side (upper side) of reflectance data; and (d) two sides of transmission data. The plane at 0mm corresponds to the case where imaging plane was placed across the center of the sphere inclusion. The actual inclusion/background contrast is 2.5.

96

Figure 5.10 shows the reconstructed a images from data acquired at depth of -

9mm, -6mm, -3mm, 0mm, 3mm, 6mm, 9mm, respectively using both reflection and

transmission data (a), both sides of reflectance data (b), one side (upper side) of

reflectance data (c), and two sides of transmission data (d). The plane at 0mm

corresponds to the case where imaging plane was placed across the center of the sphere

inclusion. In the case of using complete datasets, including both reflectance and

transmission data for NIRST image reconstruction, a circular-shape inclusion can be

found from the recovered a image (Fig. 5.10(a)). The inclusion/background contrast is

most obvious on the a image at depth of 0mm, and there is almost no contrast in the

image at depth of 12mm.

In contrast, when only two sides of reflectance data were used for NIRST image

reconstruction, the recovered inclusion on the a image presents a football shape rather

than circular shape, as shown in Fig. 5.10(b). Such difference suggests that the

reconstructed tumor/inclusion may have distortion along X-axis, the same as the

sampling direction, if only reflectance data was used for NIRST image reconstruction.

Similar distortion along the X-axis can be found in the a images when only the upper

side of reflectance data was used for NIRST image reconstruction, shown in Fig. 5.10(c).

Meanwhile, in the case of using only transmission data for NIRST image reconstruction,

distortion along the Y-axis can be found in the a images, shown in Fig. 5.10(d).

97

Table 5.3. Comparison of the reconstructed inclusion/background contrast using different subsets of measurement data acquired at different depths. Four subsets of measurement data are compared: (A) full dataset, i.e., both transmission and reflectance; (B) two sides of reflectance data; (C) upper side of reflectance data and (D) only transmission data.

Table 5.3 compares the reconstructed inclusion/background contrast using

different subsets of measurement data at various depths. Reconstruction using full dataset

provides the highest contrast at each depth. Comparing the reconstructed contrast using

different subsets, we can find that the reconstruction with only one side of reflectance

data provides slightly higher contrast than that using both sides of reflectance data. It

suggests that reasonable contrast can be recovered given only one side of reflectance

data, if the inclusion/tumor has modest distance from the boundary.

Table 5.4 provides the comparison of the sensitivity of inclusion for four subsets

of measurement data. Subset A has the highest inclusion sensitivity at all depths, while

subset D (transmission data only) has the lowest inclusion sensitivity. Also, measurement

data using two sides of reflectance data (B) has similar sensitivity to that using one side

of reflectance data, since the inclusion is far away from the lower side of the mesh

boundary. As we can see from Table 5-3, reconstruction using only transmission data

gives the lowest contrast among all groups. This can be explained by the fact that

transmission data only (D) has the lowest inclusion sensitivity among all groups, as

shown in Table 5.4.

98

Table 5.4. Comparison of the sensitivity of the inclusion (%) using different subsets of measurement data acquired at different depths. Four subsets of measurement data are compared: (A) full dataset, i.e., both transmission and reflectance; (B) two sides of reflectance data; (C) upper side of reflectance data and (D) only transmission data.

99

Chapter 6: In vivo Collagen Quantification in Breast Tissue

6.1 Introduction

Separation of multiple chromophores in breast tissue has been one of the major

challenges in NIRS and NIRST. Limited by the lack of data at wavelengths longer than

850nm, most near infrared imaging/spectroscopy systems target on the quantification of

up to four major absorbers of oxy- and deoxy-hemoglobin, water and lipid in breast. The

existence of other chromophores/absorbers is usually neglected during the imaging

reconstruction/fitting procedure. This is mainly due to the most higher gain PMTs suffer

from photocathodes that have poor performance above this wavelength range, and may

result in overestimation/underestimation in the above four major absorbers. Among all

the possible chromophores, collagen is one main constituent of soft tissues and its

quantification in breast tissue will be of interest to many researchers. Collagen seems to

be involved in the development of breast cancer [145], and expected to be related to

breast density [146]. Taroni et al has recently shown collagen as one important biomarker

in lesion classification, as concluded from a clinical study using an optical

mammography system operating with 7 wavelengths (635nm to 1060nm) [147]. Taroni

and coworkers did an extensive set of studies on collagen quantification in human breast,

who first measured collagen absorption spectrum, quantified collagen in vivo [148] and

showed collagen images [149]. However, there are only few publications regarding non-

invasive quantification of collagen content in breast tissue [111, 147, 150, 151], and there

are no publications testing the quantification of collagen in breast tissues with

tomographic capabilities.

100

Using the hybrid 12-wavlength FD-CW NIRST system, with wavelength

coverage from 661nm to 1064nm, tomographic images of collagen content in breast

tissue have been extracted for the first time. Section 6.2 shows the simulation studies

using both homogeneous and heterogeneous phantom, and section 6.3 discusses collagen

quantification in normal subjects and cancer patient, and further investigates the effect of

presence of collagen on the reconstruction of other chromophores.

6.2 Simulation

The absorption spectrum of collagen content was encoded into the reconstruction

procedure as well, which enabled tomographic recovery of collagen content, in addition

to the other four chromophores.

6.2.1 Homogeneous phantom simulation

First a homogeneous mesh with HbT of 20M, StO2 of 70%, Water of 50%,

Lipid of 50%, SA of 0.8 and SP of 0.3 was assigned for each node. A certain amount of

collagen (0-10%) was added into the mesh as well. The total amount of water, lipid and

collagen is kept as 100%. Then the forward dataset was generated, based on which the

images were recovered for all chromophores except for collagen. As a result, some levels

of overestimate or underestimate were expected in the recovered images, because of the

contribution of collagen in the forward data.

101

Figure 6.1. Recovered HbT (a), StO2 (b), water (c) and lipid (d) in a simulated homogeneous phantom, with collagen content increased from 0 to 10%.

Background collagen concentration with a range of 0 to 8.5% was reported in

human subjects by Taroni et al in Reference [111]. A slightly higher value of 10% has

been used in the simulation studies. Figure 6.1 shows the recovered chromophore

concentration of four major absorbers in a simulated homogeneous phantom when

background collagen content increased from 0 to 10%. When collagen content increased

from 0% to 10%, overestimation in HbT, water and lipid, as well as underestimation in

StO2 can be observed from the recovered values. The overestimate increases linearly with

the concentration of collagen added. With 10% collagen, there was an overestimate of

8.9uM, 1.8% and 15.8% in the recovered HbT, water and lipid, respectively. These

overestimates stayed the same, regardless of the actual assigned HbT, water and lipid

102

changed in the range of 10uM - 30uM, 30% - 70% and 30% - 70% respectively.

Meanwhile, the underestimate in StO2 decreased nonlinearly versus increasing collagen

concentration. With 10% collagen, there was an underestimate of 12.6%, 9.5% and 6.5%

in the recovered StO2 for actual StO2 assigned at 80%, 70% and 60%, respectively.

The absorption due to collagen content would contribute to the recovery of other

chromophores, and results in an overestimation in all other four chromophores, when

collagen was not included in the reconstruction procedure, as shown in Fig. 6.1.

Depending on the absorption spectrum of each chromophore and the wavelengths utilized

to generate forward data, the amount of overestimate varied between chromophores. As

shown in Fig. 6.1(b), deoxy-hemoglobin is more sensitive to the presence of collagen

than oxy-hemoglobin, which leads to the underestimate in oxygen saturation. In addition,

with the same amount of added collagen content, recovered StO2 suffers more

underestimate in the case of higher StO2. Comparing Fig. 6.1(c) and Fig. 6.1(d), it was

found that given the same amount of 10% collagen added into the background, the

overestimate in water (1.8%) is much less than that in lipid (15.8%). This may be because

one absorption peak of lipid (at 926nm) is very close to one peak of collagen (at 913nm)

and the ignored absorption from collagen was accounted in lipid.

6.2.2 Heterogeneous phantom simulation

Insufficient recovery of tumor to normal-surrounding-tissue contrast is always a

challenge in diffuse tomographical image reconstruction. In addition to the diffusive

nature of photon transport in biological tissue, the existence of chromophores other than

four major absorbers may also make a contribution. To further investigate the effect of

existence of collagen on the recovered contrast in HbT and other chromophores, a two-

103

region heterogeneous mesh was generated with a circular inclusion centered at (70mm,

40mm) with a diameter of 20mm. An inclusion/background contrast of 2 in HbT was

assigned to the mesh, with the same concentration of other chromophores as the

homogeneous phantom described above. Similarly, a certain amount (0-10%) of collagen

was uniformly added to the heterogeneous mesh. The images were reconstructed using

the simulated forward data for all chromophores without collagen included. The

recovered inclusion/background contrast was investigated versus different amounts of

collagen.

Figure 6.2. Reconstructed images of a simulated heterogeneous phantom. (a) Images with true values. The diameter of the circular inclusion is 20 mm. An inclusion/background contrast of 2 is assigned to HbT, with homogeneous background value of 75%, 45%, 45%, 10%, 0.8 and 0.3 assigned for StO2, water, lipids, collagen, SA and SP, respectively. Reconstructed images without collagen (b) and with collagen included (c).

Figure 6.2 showed the reconstructed images without collagen (Fig. 6.2(b)) and

with collagen (Fig. 6.2(c)). An inclusion/background contrast of 2 was assigned to HbT,

with true values of 40M and 20M assigned in the inclusion and background,

respectively. Homogeneous background values of 75%, 45%, 45%, 10%, 0.8 and 0.3

were assigned for StO2, water, lipids, collagen, SA and SP, respectively. As shown in

Fig. 6.2(b), the recovered HbT of 31.0M in the background was significantly

overestimated, resulting in an underestimated recovered contrast of 1.3. In contrast, with

104

collagen included in the reconstruction, the recovered contrast of HbT increased to 1.6.

Considering that the assigned collagen content was homogenous, the

inclusion/background contrast in reconstructed collagen image of 1.2 indicates possible

crosstalk between HbT and collagen.

The same procedure was repeated for StO2, water, and lipid, respectively, in the

presence of 10% of background collagen. When collagen content was not included in the

reconstruction process, the recovered contrast of StO2, water and lipid decreased by

8.3%, 4.5% and 14.3%, respectively, compared with the case where collagen was

included in the reconstruction, given true contrast of 1.3, 1.5 and 1.5.

Figure 6.3. Image reconstruction in the presence of collagen, which was not included in the reconstruction. A similar simulation setup was used as Figure 2, except that the contrast was assigned in HbT (a), StO2 (b), water (c), and lipid (d), respectively. The extracted inclusion/background contrast was plotted versus collagen concentration accordingly.

105

Figure 6.3 shows the extracted inclusion/background contrast plotted versus collagen

concentration, with contrast assigned in HbT (a), StO2(b), water (c) and lipid (d),

respectively. For all chromophores, the recovered contrast decreases versus increasing

amount of collagen concentration. The recovered contrast clearly presents different

collagen dependences between collagen less than 2% and collagen higher than 2%. The

recovered HbT contrast drops dramatically from 0% to 2% collagen, after which it keeps

dropping at a lower rate. With 10% collagen added into the background, the recovered

inclusion/background contrast in HbT drops by 28.2% from 1.9, 27.4% from 1.7, and

15.4% from 1.3, respectively for assigned contrast of 2.5, 2.0 and 1.5. For StO2 and

Lipid, the dependence of recovered contrast on collagen concentration is not obvious

with collagen less than 2%. While there is a sharp drop in the recovered contrast around

2% collagen, after which the recovered contrast tends to be stable. In the presence of

collagen, the recovered water contrast was not significantly affected, despite some

amounts of suppression.

6.3 In vivo collagen quantification in breast tissue

6.3.1 In vivo collagen quantification in normal subject

Figure 6.4. Contents of breast tissue recovered for HbT (a), StO2 (b), Water (c) and Lipids (d), with and without collagen included in reconstruction. The radiographic density type of subject #1 and #2 is heterogeneously dense (HD) and scattered fibroglandular dense (Scattered), respectively.

106

Figure 6.4 shows the comparison of estimated chromophore concentrations of two

normal subjects with and without collagen included in the reconstruction process. The

radiographic density type of subject #1 and #2 is heterogeneously dense (HD) and

scattered fibroglandular dense (Scattered), respectively. The recovered background

collagen content of subject #1 and #2 is 4.8% and 1.3%, respectively. Compared with

reconstruction without collagen included, reconstruction with collagen yields in 6.7uM &

1.6uM lower HbT, 17.8% & 4.7% higher StO2, 2.8% & 0.5% lower water, and 6.7% &

3.8% lower lipids, for subject #1 & #2, respectively.

A total number of nine normal subjects were imaged, and similar comparison was

performed for each subject. Tomographic images were reconstructed of collagen and the

other four chromophores using spectral dataset acquired from the FD-CW NIRST system.

Average background collagen concentration was then calculated for each subject using

the reconstructed collagen image. Average collagen content with a range of 0 to 4.8%

was found in breast tissue among all the subjects, together with an increase in HbT from

0 to 6.7 M, 0 to 2.8% in water, and 0 to 6.7% in lipid, and decrease in StO2 from 0 to

17.8%, when collagen was included in the reconstruction.

Comparing the reconstructions with and without collagen, there was lower HbT,

water and lipid, and higher StO2 observed when collagen was included into the

reconstruction estimation process. This result is consistent with the simulation results,

and the patient data reported by Taroni et al [148, 152]. However, the change in lipid

between with and without collagen in the image reconstruction in this study is

significantly higher than that reported by Taroni. This may be explained by the fact that

different wavelengths have been used to extract chromophore contents in our NIRST

107

system, and there are major differences in the nature of the signal acquisition (time

domain system vs FD + CW system here), as well as large potential differences in the

fitting algorithms.

6.3.2 In vivo collagen quantification in cancer patient

Figure 6.5. MRI T2 images of a patient with invasive cancer in the left breast: (a) Axial view, (b) sagittal view and (c) coronal view. Reconstructed optical images without (d) and with (e) collagen included. Recovered optical images are displayed in the same orientation in (d) and (e) as in (c).

Figure 6.5 shows the MRI and reconstructed optical images of a breast cancer

patient imaged with the NIRST system. This 63-year old woman has a 2-cm invasive

ductal carcinoma (IDC) tumor in her left breast. The position of the tumor was marked in

advance on the breast surface according to her prior MRI to guide placement of the fiber-

breast interface during optical imaging. The optical images were reconstructed without

(Fig. 6.5(d)) and with (Fig. 6.5(e)) collagen included. Orientation of the breast mesh (2-8

o’clock) was adjusted to coincide with the coordinate system in MRI, as shown in Figs.

6.5(d)-6.5(e). The recovered inclusion/background contrast in HbT increased from 1.5 to

1.7 when collagen was included in the reconstruction.

108

HbT contrast has been always presented as the most important indicator in breast

cancer diagnosis and tumor response monitoring to neoadjuvant chemotherapy (NAC). A

more accurate quantification of the HbT contrast is important to provide better separation

between malignant and benign lesions, and this can also provide more robust prediction

of tumor response to NAC at the early stages. Furthermore, a contrast of 1.2 was found in

the recovered collagen image, with an average background collagen concentration of

2.3%. This contrast may bring in a new biomarker for breast cancer detection and/or

treatment monitoring, though more patient data is needed to validate that this collagen

contrast does not come from the crosstalk between collagen and other chromophores.

6.4 Discussions

In this chapter, tomographic images of breast collagen content have been

recovered for the first time, and image reconstruction approaches with and without

collagen content included have been validated in simulation studies and normal subject

exams. Simulations indicate that including collagen content into the reconstruction

procedure can significantly reduce the overestimation in total hemoglobin, water and

lipid by 8.9 M, 1.8% and 15.8%, respectively, and underestimates in oxygen saturation

by 9.5%, given an average 10% background collagen content. A breast cancer patient

with invasive ductal carcinoma was imaged and the reconstructed images show that the

recovered tumor/background contrast in total hemoglobin increased from 1.5 to 1.7 when

collagen was included in reconstruction.

109

Chapter 7: Imaging Normal Subjects

7.1 Introduction

The breast is a turbid, light scattering medium with a complex combination of

layers, bands, sheets and nodules of absorbers, scatters and fluorophores [153]. In

general, the breast is richly supplied with blood but more vessels are found in the

fibroglandular than in the adipose tissues. Postmenopausal breast shows a reduction in

the fibroglandular volume and increase in the adipose volume [154]. Breast density

reflects variations in breast tissue composition and is strongly associated with breast

cancer risk [155]. Currently the gold standard in telling the breast density is x-ray

mammography. NIRS/NIRST methods have shown success in providing optical

biomarkers which are correlated with breast density [156-158].

The performance of the hybrid NIRST system has been validated on phantom

experiments as discussed in Chapter 5. However, the complexities of the composition,

size and shape, of the human breast with irregular-shape tumors cannot be completely

mimicked by tissue phantoms. To better evaluate the accessibility of the breast-fiber

interface to human breasts, a normal subject imaging study has been carried and the

results are presented in this section, before imaging cancer patients with the NIRST

system. The performance of the imaging system and reconstruction algorithm were

validated through comparison of reconstructed chromophore concentrations of normal

subjects with those reported in literature.

All human subjects imaging was carried out under a protocol approved by the

Committee for the Protection of Human Subjects (CPHS) at Dartmouth-Hitchcock

110

Medical Center. Written consent was obtained for each subject and the nature of

procedure was fully explained.

7.2 Imaging setup

A major improvement in the current system was the breast interface, designed to

fit different breast sizes and shapes easily. We have shown that measurement sensitivity,

or tumor coverage, plays a critical role in accurate, spatial reconstruction of optical

properties [56]. In previous studies, the fiber-breast interface has been investigated

extensively for different diffuse optical tomography systems [51, 56]. A common

disadvantage of these breast interface geometries was their lack of mobility and the

requirement for the patient to be positioned prone on a specific imaging bed during data

acquisition. For the purpose of monitoring patient response during neoadjuvant

chemotherapy, where the intention is to examine an individual frequently at different

time points during treatment, the added convenience is significant. As shown in Fig. 1,

the subjects were seated in an adjustable chair in an examination room, and the NIRST

system was placed outside.

Figure 7.1. The setup of human subject imaging. (a) The NIRST system placed outside the exam/infusion room. (b) Exam room. (c) A female subject being imaged on the left breast.

111

7.3 Imaging normal subjects with various breast sizes

Figure 7.2 shows recovered images of HbT, StO2, water, lipid, scattering

amplitude and scattering power for three normal subjects. Two interfaces were evaluated

- one with a deeper and one with a flatter surface curvature. The first (C cup, Fig. 7.2(a))

and second (D cup, Fig. 7.2(b)) subject were imaged using the breast interface with a

deeper curvature, and had separations between the two fiber holders of 63mm and 85mm,

respectively. The third subject had smaller breasts (A cup) and was imaged using the

interface with flat curvature and a maximum separation of 40mm. All sixteen fibers

contacted the breast well in the three cases. Heterogeneity in the recovered images

appears to arise from differences in fibroglandular and adipose content in the breast,

which varied case by case. No common pattern was found in the recovered images, which

suggests that the NIRST system does not introduce systematic bias. The interface with

flatter curvature was used for smaller breasts, and enabled better coupling at all fibers.

The other interface with deeper curvature yielded slightly better performance when all

fibers were in contact, which mostly the case for the larger breasts was considered here.

As shown in Fig. 7.2, we were able to image both small and large breasts. The

recovered lipid content for the participant imaged with a 63mm interface separation was

35.8%, which was lower than the other two participants.

112

Figure 7.2. Reconstructed optical images of three normal subjects. Maximum separation between the two fiber holders in the interface was 63mm (a), 85mm (b), and 40mm (c), respectively.

7.4 Continuous imaging of normal subjects

One major goal of this thesis is to dynamically monitor tumor response to NAC

during one single infusion and the whole course of treatment. Figure 7.3 shows the

recovered optical parameters of two normal subjects measured continuously for 30

minutes. The standard deviation of 8 continuous acquisitions was calculated as 4.5%,

3.8%, 4.2%, 2.3%, 3.8% and 4.1% for HbT, StO2, water, lipid, water and lipid,

respectively. Several factors contribute to the temporal variation including breathing

pattern and patient movement, and the effects of these variations on the recovery of

tumor/background optical contrast needs further investigation.

113

Figure 7.3. Continuous measurements of (a) HbT; (b) StO2; (c) water; (d) lipid; (e) SA and (f) SP for two normal subjects.

7. 5 Intra-subject and inter-subject variations

A group of ten normal subjects were imaged on both sides of the breast.

Corresponding ages, breast sizes, mammographic breast densities were recorded.

Subjects were divided into high and low radiographic density groups based on their

recent mammograms. Specifically, fatty and scattered breasts were categorized as low

density, and heterogeneously dense (HD) and extremely dense (ED) breasts were

considered as high density. Inter-subject and intra-subject variations were compared with

previous studies to validate the performance of the current NIRST system as well.

Normalized standard deviation was calculated to evaluate tissue heterogeneity in

healthy breast tissue. Table 7.1 shows the mean value and standard deviation of both

sides of the breast. The average spatial variance in HbT and StO2 across the recovered

tomographic images of each subject was 13.4% and 7.1%, respectively. Larger variance

of 14.6% and 13.7% was found for water and lipid, respectively. The larger variation in

the latter two chromophore concentrations arises mainly from the non-uniform

114

distribution of glandular structures in the breast. The low variation in StO2 and HbT

agrees with our previous study reported by Wang et al [51], and results presented by

Shan et al [159]. Compared with Wang’s data, the results here indicated less variation in

water and lipid which may be a consequence of having more CW channels in the longer

wavelength range that lead to more accurate reconstruction of water and lipid.

Table 7.1. Mean and standard deviation of optical parameters of both sides of the breast.

Inter-subject variation for age, body mass index (BMI), HbT, StO2, water and

lipid appears in Table 7.2. Mean HbT values within the breast ranged from 10.0 to

26.8μM with an overall subject mean of 18.1μM. Mean StO2 within the breast varied

from 58.5% to 88.0% with an overall group mean value of 70.5%. The results reported

here are consistent with several other studies of asymptomatic breast tissue [45, 121,

160], which also indicate that optical and physiological parameters are significantly

affected by biological factors such as age, menopausal status, hormone use, and BMI.

A substantial amount of variation was found in the recovered lipid among these

10 normal subjects, with total range from 17.2% to 62.2%, and mean value of 40.2% with

standard deviation of 15.1%. The recovered physiologically relevant values of HbT,

115

StO2, water and lipid all fell into reasonable ranges, and were comparable to previous

studies [152, 161, 162].

In addition to the inter- and intra-subject variation, differences between left and

right sides of the breast were assessed by calculating |left-right|/average*100% for all

optical parameters, which were 13.2% for HbT, 5.4% for StO2, 8.3% for water, and

12.9% for lipid. No statistically significant differences were found between the left and

right values for breasts imaged.

Table 7.2. Mean, standard deviation and total range of physiological and optical parameters of 10 normal subjects

A Student’s t-test determined whether different breast density groups could be

separated given the recovered optical parameters. Significance was achieved at the 95%

confidence interval using a two-tailed distribution. Figure 7.4 shows a comparison, in

which significantly (p<0.05) higher HbT and water contents (<0.03) were found in the

high-density group relative to the low-density group. No trends in oxygen saturation were

identified between the two groups, each being near 70% oxygenated. A strong correlation

between NIRST recovered properties and radiographic breast density was observed, as

shown previously [161, 162].

116

Figure 7.4. Comparison between radiographic dense and non-dense groups in terms of HbT (a), StO2 (b), Water (c), Lipid (d), SA (e) and SP (e).

7.6 Discussions

HbT has been shown to be an indicator of tissue malignancy, widely used to

assess changes in tumor physiology during neoadjuvant chemotherapy [2, 117, 163]. StO2

may also be an important index for predicting tumor responses, since hypoxic tumors

have been found to be more resistant to chemotherapy [164]. Water, lipid and scatter

components have also shown potential to correlate with patient response [165]. Intra- and

inter-subject variations were calculated and compared to our previous work. Temporal

variations were also investigated through continuous measurements of 30 minutes. Less

than 5% standard deviation in 8 continuous measurements was observed, suggesting the

data are stable. No significant difference was found between the two sides of the breast

for all optical parameters, which supports use of the average of the contralateral breast

imaged before treatment to highlight the tumor/background contrast relative to the

surrounding tissue over the course of therapy. Furthermore, these physiological

parameters were compared between high and low density groups, and significantly higher

117

HbT and water were found in the high density breasts, consistent with earlier studies [51].

No significant differences were found in the other optical indicators, partly because of the

modest sample size in the normal subject groups.

118

Chapter 8: Towards monitoring breast cancer response to neoadjuvant

chemotherapy 8.1 Introduction

The optimal management of locally advanced breast cancer (LABC) has been a

challenging problem [166]. Preoperative systemic (neoadjuvant) chemotherapy (NAC)

has become an important option in the treatment of LABC, which sometimes allow for

breast-conserving surgery [167]. Studies indicate that tumor response to NAC correlate

with clinical outcomes, and patients with pathologic complete response (pCR) have

higher long-term survival rate than those with pathologic incomplete response (pIR)

[168]. However, due to the difficulty in evaluating patient response, there are still

controversies in optimizing the choice of intensity and duration of NAC for LABC [169].

In the case where a patient can demonstrate pCR/pIR at early stages during the treatment,

a more customized treatment plan can be expected to maximize clinical outcome.

NIRS/NIRST has also been used in predicating and monitoring breast tumor

response to NAC, besides its application in breast cancer diagnosis. Several research

groups have published promising results [43, 93, 116, 165, 170]. The independently-

executed, prospective multicenter ACRIN 6691 trial led by Tromberg was carried out

several years ago [43]. In this study, the diffuse optical spectroscopic imaging (DOSI)

systems were developed at University of California, Irvine, and delivered to six

institutions. A total of 60 patients undergoing NAC were enrolled over a 2-year period.

The results suggest that subjects who presented a greater drop in %TOITN from baseline

measurement to mid-therapy measurement, were more likely to have a pCR to NAC.

More specifically, in the subgroup of 17 patients who have baseline tumor StO2 greater

than the median population, a AUC value of 0.83 was obtained in %TOITN.

119

In another recent study led by Zhu [171], the Hb content of 32 subjects, with 20

Miller-Payne grade 1-3 tumors and 14 grade 4 to 5 tumors, were assessed at different

stages during the NAC treatment, using a near-infrared imager coupled with an

ultrasonography (US) system. There were significantly higher HbT, oxy- and deoxy-

hemoglobin in the grade 4 and grade 5 tumors than in grade 1-3 tumors, with p-values of

0.005, 0.008 and 0.017 respectively. There was also significantly higher mean percent

HbT in grade 4-5 tumors at the end of treatment cycles 1-3, with p-values of 0.009, 0.004

and <0.001, respectively.

In the optics in medicine group at Dartmouth College, a series of clinical trials

have been conducted by Jiang et al, using the standard-alone NIRST system for

monitoring treatment response to NAC [30, 57, 114]. In an earlier study published in

2009 [30], average normalized change in HbT showed to be the only significant DOS

parameter in separating pCR from pIR group in seven patients. In a more recent study [2]

which involved 19 patients with locally advanced breast cancer undergoing NAC, it has

shown that pretreatment HbT inside the tumor ROI relative to that in the contralateral

breast, and the change in HbT after the first cycle of NAC were significant predictors of a

pCR. Therefore, HbT of both cancerous and contralateral breasts before the start of first

cycle of NAC, and change in HbT in the involved breast within the first cycle of

treatment could be used as prognostic indicators of NAC response.

The requirement of breast imaging several times with additional contrast MRI

scan may inhibit enrollment of women, based on our previous experience. A portable

NIRST system with easy setup of the fiber-breast interface could ease the burden of

participation, which allows for patient imaging during the process of chemotherapy

120

infusion within the infusion suite. In addition, a larger clinical trial involving more

patients is needed for further validation and identification of the best potential imaging

biomarkers. We expect that the combination of optical recovered biomarkers and other

clinical parameters, will add more predictive power of tumor response to NAC, with

more patients enrolled. This will help individualize patient treatment and hopefully

reduce treatment length for some patients by early identification of pIR group, which are

of great importance considering the cost and complexity of NAC procedures. In this

chapter, the hybrid FD-CW NIRST system will be used to monitor tumor response to

NAC, and case studies are discussed.

8.2 Optimal workflow of NIRST breast imaging

Figure 8.1. Optimal workflow for NIRST patient imaging.

121

Figure 8.1 outlines the workflow for NIRST patient breast imaging, which has

been optimized to get reliable optical recovery in the clinical conditions. First, the

LABVIEW acquisition program was customized for completely automatic NIRST data

acquisition, and a user friendly custom LABVIEW GUI was developed as well. Each

time a patient was imaged, the patient relevant information such as patient ID, age, body

mass index (BMI), mammographic breast density, visit number, interface orientation and

size were input directly through the acquisition GUI. Measurement data of multiple visits

were grouped together with easy lookup. Before each patient imaging session, a reference

homogeneous silicone phantom was imaged for calibration. This calibration was used to

calibrate system variation and to reconstruct patient images.

During the patient imaging session, measurement plane/orientation was picked up

and recorded according to the tumor location and size obtained from the pre-

treatment/diagnostic MRI/mammography images. After data acquisition at all

wavelengths, a 2D mesh was chosen from the mesh library based on given separation

between two plates of the breast interface.

With both homogeneous phantom and heterogeneous patient data, the nonlinear

FEM based reconstruction strategy and parameters were optimized to get accurate optical

properties from patient imaging. Individual regularization parameter and stopping

criterion, which decides number of iterations before the reconstruction terminates, were

chosen to maximize the ratio of the tumor to contralateral breast.

Displaying the optical images in a way that the tumor location can be easily

recognized and correlated to MR images is a challenging and critical task due to different

imaging geometries and breast deformation during the optical data acquisition. Such

122

visualization is even more important in monitoring tumor response to NAC since the

breast/tumor size may change significantly during the long course of NAC. A user-

friendly GUI was developed to display optical images for easy interpretation by clinical

personnel.

For the purpose of ROI analysis, the tumor ROI was defined with the help of

contrast MRI images of coronal plane acquired before the initiation of therapy. The entire

tumor area, the entire area outside of the tumor, and the entire area of the contralateral

breast, were defined as tumor, nontumor, and contralateral ROIs, respectively. The mean

values of HbT, StO2, water, lipid, SA and SP of each ROI were extracted from the

reconstructed images.

8.3 Case studies of imaging breast cancer patients

8.3.1 Imaging breast cancer patients

To validate the performance of the NIRST system, espacially the flexibility of the

fiber-breast interface in the case of various breast sizes, shapes and tumor locations, a

group of 6 breast cancer patients have been imaged by the NIRST system.

123

Figure 8.2. Case study #1. Dynamic Contrast Enhanced MR Images (DCE-MRI) of a patient with invasive cancer in the right breast: (a) Axial view, (b) sagittal view and (c) coronal view. Recovered optical images of HbT, StO2, water, lipid, SA, and SP for right breast (d) and contralateral breast (e).

Case study #1: This case was a 67-year old woman with a 1.1×0.9×1.2 cm3

invasive ductal carcinoma in her right breast. The position of the tumor was marked in

advance on the breast surface to guide placement of the fiber interface during optical data

acquisition. Dynamic Contrast Enhanced MR Images (DCE-MRI) of the patient were

also displayed in Figs. 8.2(a)-8.2(c). Registration between MRI and reconstructed optical

images was accomplished, with the prior knowledge of the fiber bundle positions relative

to the tumor location (marked on the breast surface), using MR images acquired prior to

the optical imaging session. Orientation of the breast mesh (1-7 o’clock) was adjusted to

coincide with the coordinate system in MRI, as shown in Fig. 8.2(d). The tumor region

was segmented from the recovered HbT image. The average tumor to background

contrast was calculated to be 1.41X for HbT, 0.92X for StO2, 1.23X for water, 0.78X for

lipid, 0.95X for scattering amplitude, and 1.08X for scattering power, respectively.

124

Figure 8.3. Case study #2. DCE-MRI of a patient with invasive cancer in the right breast: (a) Axial view, (b) sagittal view and (c) coronal view. Recovered optical images of HbT, StO2, water, lipid, SA, and SP for cancerous breast (d) and contralateral breast (e).

Case study #2: This case was a 39-year-old female subject with a 2.1x1.5x1.9cm

invasive ductal carcinoma and ductal carcinoma in situ in her right breast. MRI images of

the patient were also acquired and displayed in Figs. 8.3(a)-8.3(c). Orientation of the

breast mesh (2-8 o’clock) was adjusted to coincide with the coordinate system in MRI, as

shown in Fig. 8.3(d). The HbT concentration at the tumor region was 27.8μM on average,

while the background value was about 17.7μM. The average tumor to background

contrast was calculated to be 1.57X for HbT, 1.08X for StO2, 0.92X for water, 1.07X for

lipid, 1.50X for SA, and 1.16X for SP, respectively.

Figure 8.4. Case study #3. DCE-MRI images of a patient with invasive cancer in the left breast: (a) Axial view, (b) sagittal view and (c) coronal view. Recovered optical images of HbT, StO2, water, lipid, SA, and SP for cancerous breast (d) and contralateral breast (e).

Case study #3: This case was a 47-year-old female subject with a 12x8x5 cm

invasive ductal carcinoma in her left breast. MRI images of the patient were also

displayed in Figs. 8.4(a)-8.4(c). Orientation of the breast mesh (2-8 o’clock) was adjusted

125

to coincide with the coordinate system in MRI, as shown in Fig. 8.4(d). It’s clearly seen

from the MRI images that the tumor region occupies almost the whole breast. The HbT

concentration at the tumor region was 37.1μM on average, while the background value

was about 23.9μM. The average tumor to background contrast was calculated to be 1.55X

for HbT, 1.02X for StO2, 1.39X for water, 0.77X for lipid, 1.37X for SA, and 0.49X for

SP, respectively.

Figure 8.5. Case study #4. DCE-MRI images of a patient with invasive cancer in the left breast: (a) Axial view, (b) sagittal view and (c) coronal view. Recovered optical images of HbT, StO2, water, lipid, SA, and SP for cancerous breast (d) and contralateral breast (e).

Case study #4: This case was a 49-year-old female subject with 5.3x3.2x1.4 cm

invasive ductal carcinoma in her right breast. MRI images of the patient were also

acquired and displayed in Figs. 8.5(a)-8.5(c). Orientation of the breast mesh (2-8 o’clock)

was adjusted to coincide with the coordinate system in MRI, as shown in Fig. 8.5(d). A

modest amount of pressure was added onto the breast, which ensured that all the fibers

were in good contact with the breast tissue. The tumor was pushed towards the center of

the breast, as shown in the reconstructed HbT image in Fig. 8.5. The HbT concentration

126

at the tumor region was 32.4μM on average, while the background value was about

24.1μM. The average tumor to background contrast was calculated to be 1.46X for HbT,

1.03X for StO2, 1.52X for water, 0.82X for lipid, 1.24X for SA, and 1.27X for SP,

respectively.

8.3.2 Monitoring breast response to NAC

Figure 8.6. Case study #5. Clinical images of a 63-year-old patient with pathological confirmed pIR. Postcontrast T2-weighted MRI images prior to initiation: (a) axial view, (b) sagittal view and (c) coronal view. Pathological findings showed pIR to neoadjuvant chemotherapy. Reconstructed optical images of HbT (uM), StO2 (%), water (%), lipid (%), scattering amplitude (SA) and scattering power (SP) before treatment (d), on day 9 of cycle 1 (e) and after therapy (29 days prior to surgery, (f)) are shown for abnormal breast. The recovered images of contralateral breast are shown in (g).

Case study #5: This case was a 63-year-old woman who had a 2.0x2.2x2.2cm

tumor in her left breast which was proved to be DC, ER-positive, PR-positive and HER2-

127

negative. The chemotherapy regime consisted of dose dense AC (DD-AC, Doxorubicin +

Cyclophosphamide) and weekly trastuzumab. Figure 8.6 shows postcontrast MRI images

obtained prior to initiation, as well as reconstructed optical images of HbT, StO2, water,

lipid, SA and SP 14 days before the first treatment, on day 9 of cycle 1, and 6 days after

the last cycle (29 days before surgery) for the left breast.

Table 8.1. Reconstructed optical contrast in terms of HbT, StO2, water, lipid, SA and SP for three visits of before treatment, on day 9 of cycle 1, and after therapy, respectively.

As shown in Table 8.1, the contrast was defined as the ratio between tumor ROI

and pretreatment contralateral breast in HbT, StO2, water, lipid, SA and SP, respectively

for three measurements. There has been an increase in the contrast of HbT between pre-

treatment measurement and (Day 9, cycle 1), from 1.4 to 1.6. Such increase in HbT

contrast has also been found in previous studies as a strong indicator to pIR [30].

128

Figure 8.7. Case study #6. Clinical images of a 54-year-old patient with radiologic findings and pIR. Postcontrast MRI images prior to initiation: (a) axial view, (b) sagittal view and (c) coronal view. Pathological findings showed pIR to neoadjuvant chemotherapy. Reconstructed optical images of HbT (uM), StO2 (%), water (%), lipid (%), scattering amplitude (SA) and scattering power (SP) before treatment (d), on day 19 of cycle 1 (f), and after therapy (24 days prior to surgery, (g)) are shown for abnormal breast. The recovered images of contralateral breast are shown in (g).

Case study #6: This case was a 54-year-old woman who had a 3.8x2.4x3.4cm

tumor in her left breast which was proved to be IDC, DCIS, ER-negative, PR-negative

and HER2-negative. The chemotherapy regime consisted of four cycles of Carboplatin

and Taxol. Figure 8.7 shows postcontrast MRI images obtained prior to initiation, as well

as reconstructed optical images of HbT, StO2, water, lipid, SA and SP 2 days before the

first treatment, on day 19 of cycle 1, and 19 days after the last cycle (24 days before

surgery) for the left breast.

129

Table 8.2. Reconstructed optical contrast in terms of HbT, StO2, water, lipid, SA and SP for three visits of before treatment, on day 19 of cycle 1, and after therapy, respectively.

Table 8.2 shows the extracted contrasts in HbT, StO2, water, lipid, SA and SP,

respectively for three visits. There is an early percentage change of -16.7% in HbT

between (Day 19, cycle 1) and pretreatment measurement. The pathology results on the

surgical specimen revealed that the residual IDC and DCIS were present in the lesions,

which confirmed the case as pIR, although with some treatment effects. This patient has

Triple-negative breast cancer (TNBC), by the lack of estrogen receptor (ER),

progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER-2)

expression. TNBC are biologically aggressive and studies have suggested that they

respond to NAC better than other types of breast cancer [172]. However, this patient still

had partial response to NAC, which might be related to the negative early percentage

change in HbT contrast.

8.4 Discussions

HbT has shown to be the most important optical biomarker in breast diagnosis

and treatment monitoring [2, 45]. The contrast in HbT indicates an overall increase in

tissue vascular density in the tumor region. The formation of new blood vessels is

stimulated by the production of various growth factors caused by the carcinoma cells

[173]. Hemoglobin saturation is also reflective of their physiologic composition in the

breast. Oxy hemoglobin is more related to global vascular structures, while deoxy-Hb is

130

more sensitive to cellular oxygen consumption and local metabolism. Low oxygen

tension or hypoxia is capable of facilitating tumor growth [174], and it has been related to

decreased survival rate and increased risk of local recurrence in soft tissue sarcomas

[175]. Therefore, the measurement of pretreatment/baseline StO2 in breast can be

attractive due to its potential predictive value. However, there is no obvious change in

StO2 in the tumor, as shown in the case studies, which is consistent with other studies

[27].

Besides oxy- and deoxy-Hb, water and lipids are another two major absorbers in

the breast tissue, which have critical biochemical significances [165]. MRI studies of the

apparent diffusion coefficient of water (ADCw) provide an insight into the distribution of

water, which relates to the cellular pathology and cellular density in several tissues. The

increase in water of the tumor might indicate variations in tumor cell density and edema.

Meanwhile, the reductions in tumor water might be caused by cell deaths. Cerussi

reported that there is a significantly higher pretreatment tumor to normal water ratio in

the pCR group [165]. The quantification of lipid can also be helpful indicating lesions in

the breast. Several studies on normal breast tissues have shown the lipid properties in

breast are strongly affected by age and menopausal status [159, 176].

Besides the absorption-derived chromophore concentrations, scattering-derived

parameters such as scattering amplitude (SA) and scattering power (SP) provide

additional information about tissue cellularity [177], composition [162], and disease state

[178]. SA and SP are related to the number density and size of scatters, respectively. A

large variation in SA and SP was reported among subjects with different radiographic

densities [179], which was most likely related to the dominant compositional changes.

131

Further microscopic study of the scattering properties of tumors has the potential to

explain the exact features that contribute to the scattering properties.

132

Chapter 9: Conclusions and Future Directions

9.1 Completed work

This thesis focused on improving the performance of near infrared spectral

tomography (NIRST) for diagnostic imaging and treatment monitoring of breast cancer,

from both the developments of imaging system and the nonlinear image reconstruction.

Some major results and conclusions are recapped here.

9.1.1 Optimization of MRI-guided NIRST image reconstruction

Chapter 3 introduced a robust optimization algorithm for optimal choice of the

regularization parameters for “hard prior” image reconstruction based on exam-specific

data acquired during MRI/NIRST examination of the breast. The optical contrast values

for HbT, StO2, TOI, and scattering parameters were estimated by applying the

optimization algorithm to amplitude data only, and both amplitude and phase data. A

statistical difference with p<0.05 indicates that absorption-derived contrasts in HbT and

TOI as well as reduced-scattering derived contrast in SP can be used to differentiate

malignant from benign lesions.

To the best of our knowledge, these results represent the first time an extensive

study of regularization has been conducted on a relatively large amount of clinical breast

exam data with the MRI-NIRST multi-modality imaging approach. The optimization

algorithm has better performance in differentiating malignant from benign cases,

compared to a fixed regularization parameter. The best diagnostic performance occurred

with optimal regularization values selected from individual’s exam data, and when

combining HbT and TOI estimated using both amplitude and phase data as the diagnostic

indicator. These results were published by Zhao et al [113].

133

9.1.2 A hybrid FD/CW system with simultaneous measurements at twelve wavelengths

Chapter 4 summarized the unique features of the 12-wavelength hybrid FD-CW

NIRST system developed in this thesis. The system was developed for quantifying

changes in HbT, StO2, water, lipid, scattering amplitude and scattering power in the

breast during neoadjuvant chemotherapy. One six-wavelength FD source module and one

six-wavelength CW source module were built to provide the wavelength coverage in the

range of 661–1064nm. A full data acquisition was completed by sequentially acquiring

two sets of data, each of which consisting of simultaneous acquisition of three FD and

three CW wavelengths. Using a novel gain adjustment scheme in the PMTs, the data

acquisition time for each simultaneous acquisition has been reduced from 90 to 55

seconds, while signal variation was also reduced from 2.1% to 1.1%. An adjustable

interface was designed to fit different breast shapes and sizes. All components were

integrated into a portable system, which allows robust measurements in the infusion unit.

The performance of the hybrid NIRST system was validated in phantom experiments

(Chapter 5), normal subject exams (Chapter 7) and breast cancer exams (Chapter 8). The

design and characterization of the 12-wavelength system was published by Zhao et al

[180].

9.1.3 Silicone soft-gel based tissue mimicking phantom for NIRST imaging

In Chapter 5, several major tissue-mimicking phantoms were compared, and a

novel silicone soft-gel based tissue mimicking phantom was developed and characterized.

The proposed silicone soft gel phantom used A-341 silicone as base material. Different

coloring paints were used to provide scattering and absorption, respectively. The

preparation procedure is significantly simplified compared with that of making gelatin

134

and traditional resin/RTV phantoms, since it does not require either vacuum or

microwave. The whole procedure can be completed within one hour. The phantom

presents stable optical properties over several months, which makes it an ideal candidate

in repetitive measurement for quality assurance and routine calibration.

The concentration of pink paint in the phantom affects both a and s in a linear

way, while that of white paint only affects s linearly. By altering the concentration of

white and pink paints, specific a and s can be obtained. A series of large breast tissue

mimicking phantoms were made as well, with sphere-shape inclusions of different sizes.

Such phantom was imaged at different depths, and the reconstructed images presented a

strong depth dependence of recovered contrast. Eventually, the heterogenous phantom

was used to validate the feasibility of NIRST imaging with partial

transmission/reflectance data. The experiments results suggest that reasonable image

reconstruction can be achieved using only one side of reflectance data.

9.1.4 Collagen quantification using the NIRST system

In Chapter 6, tomographic images of breast collagen content have been recovered

for the first time, and image reconstruction approaches with and without collagen content

included have been validated in simulation studies and normal subject exams. When

collagen was not included into the reconstruction, overestimate in recovered HbT, water

and lipid, as well as underestimate in StO2, have been observed. Both simulation and

patient studies showed that the recovered inclusion/background contrast in HbT increased

when collagen was included into the reconstruction, in the presence of background

collagen.

9.1.5 Imaging normal subjects

135

Chapter 7 discussed imaging of normal subjects using the hybrid NIRST system.

A group of ten normal subjects were imaged on both sides of the breast. The adjustable

breast-fiber interface proved to work well for various breast sizes and breast densities. A

strong correlation between NIRST recovered properties and radiographic breast density

was observed, where significantly (p<0.05) higher HbT and water contents were found in

the high-density group relative to the low-density group. Two normal subjects were

measured sequentially for 30 minutes and the standard deviation of 8 continuous

acquisitions was found to be less than 5%.

9.1.6 Imaging breast cancer patients

Chapter 8 presented case studies of monitoring tumor response to NAC using the

NIRST system. An optimized workflow of data acquisition and image reconstruction was

developed to get reliable optical recovery in clinical conditions.

9.2 Future Directions

The work presented in this thesis improved NIRST breast imaging from different

perspectives. In this section, three potential directions are proposed based on preliminary

results, to further improve the performance of NIRST in breast cancer diagnosis and

treatment monitoring.

9.2.1 Optimization of NIRST system for monitoring patient response to NAC

The design of a multi-channel hybrid FD-CW NIRST system with simultaneous

acquisition has been introduced in this thesis. The performance of the NIRST system may

be further improved with the addition of more FD channels (phase data). Optical image

reconstruction using FD measurement with both amplitude and phase data has shown

superior performance to that using CW measurement with only amplitude data, in the

136

case of single wavelength NISRT image reconstruction. However, as shown in Chapter 3,

the improvement in the spectral image reconstruction with the addition of phase data was

not significant in MRI-guided NIRST using hard-prior image reconstruction, since the

effect of phase data can be partially replaced by the spatial prior information obtained

from MRI images. As a result, better localization of tumor in the reconstructed images

can be expected with the addition of more phase data, when no spatial prior information

is available.

As shown in Chapter 8, the shape of the breast was approximated by a 2D

football-shape FEM mesh, which was created using the separation between two sides of

the fiber-breast interface. Such approximation is valid only when certain boundary

conditions were satisfied. In the case of small breast imaging, the simplified 2D

approximation might be insufficient for accurate modeling of the breast tissue.

Multimodality NIRST imaging with MRI/Ultrasound can be used to provide 3D patient

specific mesh for NIRST image reconstruction, which is not feasible in the current setup.

Instead, thanks to the rapid progress in consumer-grade electronics, range sensing devices

such as time-of-flight (ToF) camera and structured light imaging system, have the

potential to provide high-resolution 3D point cloud map of the surrounding object, from

which 2D/3D breast FEM mesh can be created.

9.2.2 Optimization of sampling geometry in MRI-guided NIRST

PMT detectors are usually expensive and bulky in size. In a typical fiber based

FD NIRST system, a large array of PMT detectors are needed to sample enough source-

detector pairs in terms of both amplitude and phase data. The reduction of phase data

137

without significantly sacrificing system performance can help simply the design of next-

generation FD NIRST systems.

It has been found in section 3.2 that we can still differentiate the malignant vs.

benign lesions based on recovered optical contrast, using phase data for only

homogeneous fitting of initial guess, but not necessary in the reconstruction. This can be

explained by the assumption that in MRI guided NIRS tomography, hard priors

segmented from MRI images can provide similar spatial information phase data used to

provide. As a result, phase data is not necessary during the reconstruction procedure in

the case of hard-prior reconstruction. However, phase data is still needed to get initial

guess of SA and SP. In this section, we systematically investigated the NIRST

reconstruction with limited phase data.

Figure 9.1. Flowchart outlining the sequence for two reconstruction methods.

As shown in Fig. 9.1, when both amplitude and phase data are used to construct

the Jacobian matrix, noted as AMPL/PH, both absorption derived parameters such as

chromophore concentrations of HbT, StO2, water and lipid, and also scattering derived

parameters including SA and SP, can be recovered; when only amplitude data is used to

construct the Jacobian matrix, only absorption derived parameters will be recovered.

Note, even for the AMPL reconstruction method, scattering parameters are still needed to

be assigned in the initial guess to initialize the reconstruction procedure. For either

138

AMPL/PH or AMPL method, both amplitude and phase data are fitted into the model to

get a complete set of initial guesses by assuming the tissue is homogeneous. The process

has historically provided a reasonable starting point for the image reconstruction

algorithm [45].

(a) (b) (c) Figure 9.2. Triangular interface with different sampling geometries. Three strategies of choosing phase measurements, (a) with full transmission across many sources and detectors, (b) with just partial reflectance data, and (c) with just partial transmittance data. Fiber locations are shown as blue dots.

The utilization of triangular patient interface has been discussed previously by

Mastanduno et al [96]. The adjustable triangular interface is designed to place 16 fiber

optical bundles via position 1 through 16, as shown in Fig. 9.2. There are three strategies

of taking phase measurements, (a) with full transmission across many sources and

detectors, (b) with just partial reflectance data, and (c) with just partial transmittance data.

To better understand how the number of detectors and sources affects the accuracy of

initial guess of scattering properties, initial guesses with a series of detector and source

combinations are obtained and further compared.

139

Figure 9.3. (a) Illustration of triangular patient interface for optical fiber placement. (b) Relative difference versus detector number.

Both the triangular interface and patient breast do not have geometrical symmetry

at different optical bundles. As a result, different fiber bundles and thus corresponding

PMT/PD detectors can have various importance from the image reconstruction

perspective. To evaluate the importance of a single detector, the relative error in terms of

SA and SP between homogeneous fitting with phase data from all 15 PMT detectors and

that using only one single PMT detector are plotted in Fig. 9.3(b).

Taking SA for instance, local minimum of fitting difference occurs at detector #

3 and #10, which indicates that they are the most critical positions for detector placement.

Similarly, the other positions can be sorted form high to low priority based on the fitting

difference. Moreover, we also can notice that SA (blue) and SP (red) shows similar

pattern with respect to detector position. Eventually, all the positions are sorted with

corresponding priority.

140

Figure 9.4. (a) Relative difference of optical contrast versus number of FD detectors used for homogeneous fitting of initial guess. ROC curves with 4 detectors (b), 6 detectors (c) and 15 detectors (d) for estimating initial guess.

Next, we tried to use phase data from limited number of PMT detectors for

estimating of initial guess of scattering properties, and compare with that using complete

phase data in terms of tum/ad contrast. As shown in Fig. 9.4(a), the relative difference of

tumor/adipose contrast is plotted versus number of PMT detectors. From 0 to 14, phase

data of a given number of sorted PMT detectors with highest priority are used for

estimating initial guess. Taking the number of 6 FD detectors for instance, PMT detectors

of position #2, #3, #4, #9, #10, and #11 were included. With increasing number of FD

detectors included, the relative difference in terms of contrast decreases due to the

141

addition of more phase data. Note that there exists a significant drop between 4 and 6 FD

detectors. 4 detectors are still not enough to get accurate initial guess, with a difference of

more than 10%, while 6 detectors yield in a relatively small difference of less than 3%.

As a result, we are able to get small enough variation in tumor contrast to the surrounding

normal tissue, by reducing the number of FD detectors from 16 to 6, showing the

potential of reducing the FD detectors. By comparing Fig. 9.4(b) through Fig. 9.4(d), we

can also find that 4 detectors and 15 detectors show similar statistical performance, p-

value=0.005 & AUC=0.82, versus p-value=0.004 & AUC=0.84, respectively in terms of

HbT. Meanwhile, 4 detectors correspond to a much worse AUC of 0.71.

Table 9.1. Initial guess of SA and SP for four categories grouped by MRI-identified breast density.

As we have discussed, the number of FD detectors could be as low as 6 without

sacrificing the reconstruction accuracy. This could be further improved with a lookup

table based approach. Mean values of four groups with different breast densities, and all

patients are listed in terms of SA and SP, as shown in Table 9.1.

Reconstruction of tumor to adipose contrast with initial guess from two lookup

tables, and that fitted with phase data (AMPL), are compared in Fig. 9.5 in terms of p

value and AUC for HbT and TOI. For HbT, both Table I and Table II have similar

performances (p value below 1% and close AUC), and are better than the first method of

AMPL (AUC of 0.88). Similar pattern exists for TOI among three methods. Therefore,

we can differentiate between malignant and benign patients without using any phase data

142

for either estimating initial guess or reconstruction. Such conclusion shows the potential

of hardware simplification by removing FD modules in MRI guided NIRS tomography,

while keeping a reasonable statistical performance.

Figure 9.5. ROC curves of AMPL reconstruction (a), Lookup table I (b), and lookup table II (c) for HbT, and AMPL reconstruction (d), lookup table I (e), and lookup table II (f) for TOI.

In this section, MRI guided NIRS optical tomography with limited phase data was

investigated. The optical contrast values for HbT, StO2, TOI, and scattering parameters

were estimated using amplitude only and both amplitude and phase data. A systematic

optimization of the system hardware design has been conducted as well. A statistical

difference (p<0.05) occurred between malignant and benign groups in terms of HbT and

TOI. We are able to get less than 3% variation in tumor contrast to the surrounding

normal tissue, by reducing the number of FD detectors from 16 to 6, showing the

potential of reducing the FD detectors. Furthermore, a lookup table of the scattering

143

properties has been made to replace that fitted from measured phase data. To the best of

our knowledge, these results represent the first time an extensive study of reconstruction

with limited phase data has been conducted on a relatively large amount of clinical breast

exam data with the MRI/NIRST multi-modality imaging approach. Such result would

potentially help in the design of next generation MRI coupled NIRS system and

optimized reconstruction method.

9.2.3 Imaging small tumors using MRI-guided NIRST

It’s always challenging to diagnose small size breast tumor with MRI-guided

NIRST. In this section, the limitation of the detectible size of tumor in MRI-guided

NIRST was investigated through simulation studies. The tumor size was defined as the

equivalent tumor diameter. The tumor region was first segmented from the MRI T1

images, and the whole tumor volume was calculated by adding all the tumor elements

with corresponding sub volume. Then equivalent tumor diameter was calculated as the

diameter of an imaginary sphere with the same volume of the tumor. The tumor region

can be dilated/eroded in the MRI mesh to increase/decrease the equivalent tumor

diameter. 2D mesh of equivalent tumor diameter of 12.2mm, 37.7mm and 46.8mm are

shown in Fig. 7(a)-(c), respectively.

Figure 9.6. Equivalent tumor diameter of 12.2mm (a), 37.7mm(b) and 46.8mm (c). Three regions of tumor, fibroglandular and adipose are represented in white (region 3), yellow (region 2) and red (region 1), respectively.

144

Tumor sensitivity is another critical indicator characterizing the relative coverage

of optical measurement of the tumor region, which was defined by Mastanduno et al. [56]

and suggests that no diagnostic significance was found when measured amplitude and

phase data with very low region of interest sensitivity were included in the analysis. In

practice, a minimum relative tumor sensitivity of 1% was required for filtering patient

data.

As shown in Fig. 9.7, the recovered HbT of tumor (black asteroid) is plotted

versus equivalent tumor diameter, with corresponding tumor sensitivity represented by

red square. The tumor sensitivity increases when equivalent tumor diameter increases for

both patients, since larger tumor tends to be covered more given the fixed sampling

geometry. The recovered HbT of tumor is close to that of background with small

equivalent tumor diameter (<9mm), shown in Fig. 9.7(a). For the other patient, the

minimum equivalent tumor diameter becomes 5mm (Fig. 9.7(b)).

Figure 9.7. Recovered HbT (black asteroid) of tumor and adipose (black circle) are plotted versus equivalent tumor contrast, for patient #51 (a) and #30 (b), respectively. The corresponding tumor sensitivity is represented by red square.

Figure 9.8 shows the recovered tumor/adipose contrast versus actual

tumor/adipose contrast for the tumors with equivalent tumor diameter of 8mm, 10mm,

12mm, 20mm and 40mm, respectively. The recovered tumor/adipose contrast presents

145

similar pattern when the tumor size is relatively large (20 and 40 mm), while the

recovered contrast increases much slower versus actual contrast for small tumor with

equivalent tumor diameter of 8-12mm. Aggressive regularization techniques can be used

to recover objective/maximum contrast in the case of small tumor.

Figure 9.8. Recovered tumor/adipose contrast versus actual tumor/adipose contrast for the tumor with equivalent tumor diameter of 8mm, 10mm, 12mm, 20mm and 40mm, respectively. A fixed regularization parameter of 1 was used in the image reconstruction.

Finally, the role of phase data in the recovery of HbT of tumor was investigated

for tumor with small size. The recovered HbT of tumor and adipose was plotted versus

equivalent tumor diameter, using FD/CW reconstruction (Fig. 9.9(a)) and CW

reconstruction (Fig. 9.9(b)), respectively for the same patient #30. Using FD/CW

reconstruction with both amplitude and phase data, the minimum equivalent tumor

diameter with recoverable contrast is 5mm, and the recovered HbT of tumor stays

relatively stable when tumor diameter increases higher than 5mm (Fig. 9.9(a)). By

contrast, the minimum equivalent tumor diameter increases to 8mm using CW

146

reconstruction, and the recovered HbT of tumor has much larger variation as well (Fig.

9.9(b)).

Figure 9.9. Comparison between FD/CW reconstruction (a) and CW reconstruction (b), with increasing equivalent tumor diameter. The recovered HbT of tumor and adipose are represented by blue asteroid and red square, respectively.

To conclude, the limit on the size of tumor with recoverable optical contrast in

MRI-guided NIRST was investigated through simulation studies, using the patient MRI

mesh with actual optical sampling geometry acquired from our clinical dataset. The

tumor was dilated from the original tumor size, corresponding to increasing equivalent

tumor diameter and increasing tumor sensitivity. Minimum equivalent tumor diameter of

5mm was found. Moreover, FD/CW reconstruction shows better performance than CW

reconstruction in the case of small tumor.

147

Appendix A: LabVIEW Acquisition Program

A.1 Front panel: Program initialization

(1) Program initialization

(2) Patient/phantom measurement mode (default)

(3) Calibration of PMT/PD detectors

148

A.2 Front panel: Data acquisition

(1) Switch to the “Acquisition” tab

(2) Choose the path of the folder where patient data will be saved

(3) Type patient ID

(4) Choose one from the measurement categories: “homo”, “hetero”, “left”, “right”

(5) Set visit number of a specific patient

(6) Choose one acquisition number: 13 and 14 for a complete measurement

(7) Click “Start Acquisition” once (1) to (6) have been set

149

Appendix B: Matlab Codes

Function Name Description

calibration_FD_wavelength Calibrate the AC amplitude and phase

using the PMT calibration file

calibration_CW_wavelength Calibrate the AC amplitude using the PD

calibration file

calibration_all_wavelength Calibrate amplitude and phase of all the

wavelengths

plot_lnri_fd Plot lnri and phase vs. s-d distance for FD

measurement.

plot_lnri_cw Plot lnri vs. s-d distance for CW

measurement.

reconstruction_spectral_fdcw Reconstruction of patient/phantom optical

images using hybrid FD-CW data.

reconstruction_single_wavelength Reconstruction of absorption/scattering

images using single wavelength data.

visualization_reconstructed_images Visualization and image processing of the

reconstructed optical images

para_mesh_creation Make 2D football shape mesh with various

separations between two plates

reconstruct_spectral_fdcw_n

Reconstruction of optical images using

hybrid data set. The choice of

regularization, stopping criterion, and

other constraints can be adjusted.

recon_GUI GUI for data calibration, image

reconstruction and visualization.

tumor_dilation_MRI Varying the size of tumor in MRI breast

images.

Optimal_regualrization_L_curve Get optimal regularization parameter using

L-curve analysis

150

Appendix C: Itemized Components List Itemized Component Quantity

Multi-channel RF synthesizer (HS2004, Holzworth Instruments) 1

6x1 fiber optic combiner (Fiberguide, Striling, New Jersey) 2

Photomultiplier tube (PMT, H9305-3, Hamamatsu, Japan) 15

Photodiode module (PD, C10439-03, Hamamatsu, Japan) 15

Bidirectional RF switch (401-220802A-R0HS, Dow-Key Microwave) 3

Coaxial frequency mixer (ZP-1-S+, Mini-Circuits) 3

Coaxial Bias-Tee (ZFBT-282-1.5A+, Mini-Circuits) 3

Power Splitter (ZFSC-2-1-S+, Mini-Circuits) 3

DC power supply (LS35-5, TDK-Lambda) 4

4-Channel SPDT Relay Board (RLY104, Winford, USA) 4

Laser Diode Driver (IP250-BV, Thorlabs) 6

Quadruple 2-Input Positive-AND Gates (SN74LS08N, Texas Instruments)

6

Laser Diode Driver (LD-1255, Thorlabs) 6

Collimation lenses (F220SMA-B, Thorlabs) 12

Microwave cable (Thorlabs) 30

Laser Diodes at 661nm, 730nm, 785nm, 808nm, 830nm, 850nm, 852nm, 905nm, 915nm, 940nm, 975nm and 1064nm

12

Computer System (SL-2U-AH110M-WD, SuperLogics) 1

16 Ch Voltage Output Module for USB (NI 9264, National Instruments) 1

Multifunction I/O Device (USB-6255, National Instruments) 1

Multifunction I/O Device (USB-6259, National Instruments) 1

3-meter long bifurcated fiber bundles (Customized) 16

151

References

1. T. Durduran, R. Choe, W. Baker, and A. Yodh, "Diffuse optics for tissue

monitoring and tomography," Reports on Progress in Physics 73, 076701 (2010).

2. S. Jiang, B. W. Pogue, P. A. Kaufman, J. Gui, M. Jermyn, T. E. Frazee, S. P. Poplack, R. DiFlorio-Alexander, W. A. Wells, and K. D. Paulsen, "Predicting breast tumor response to neoadjuvant chemotherapy with Diffuse Optical Spectroscopic Tomography prior to treatment," Clinical Cancer Research 20, 6006-6015 (2014).

3. F. El-Ghussein, M. A. Mastanduno, S. Jiang, B. W. Pogue, and K. D. Paulsen, "Hybrid photomultiplier tube and photodiode parallel detection array for wideband optical spectroscopy of the breast guided by magnetic resonance imaging," J Biomed Opt 19, 011010 (2014).

4. L. Tabar, M.-F. Yen, B. Vitak, H.-H. T. Chen, R. A. Smith, and S. W. Duffy, "Mammography service screening and mortality in breast cancer patients: 20-year follow-up before and after introduction of screening," The Lancet 361, 1405-1410 (2003).

5. P. Skaane, S. Hofvind, and A. Skjennald, "Randomized trial of screen-film versus full-field digital mammography with soft-copy reading in population-based screening program: follow-up and final results of Oslo II study," Radiology 244, 708-717 (2007).

6. A. Lucassen, E. Watson, and D. Eccles, "Evidence based case report: advice about mammography for a young woman with a family history of breast cancer," BMJ: British Medical Journal 322, 1040 (2001).

7. L. W. Bassett, and C. Kimme-Smith, "Breast sonography," AJR. American journal of roentgenology 156, 449-455 (1991).

8. W. A. Berg, L. Gutierrez, M. S. NessAiver, W. B. Carter, M. Bhargavan, R. S. Lewis, and O. B. Ioffe, "Diagnostic accuracy of mammography, clinical examination, US, and MR imaging in preoperative assessment of breast cancer," Radiology 233, 830-849 (2004).

9. S. Harms, D. Flamig, K. Hesley, M. Meiches, R. Jensen, W. Evans, D. Savino, and R. Wells, "MR imaging of the breast with rotating delivery of excitation off resonance: clinical experience with pathologic correlation," Radiology 187, 493-501 (1993).

10. C. Boetes, J. O. Barentsz, R. D. Mus, R. Van Der Sluis, L. van Erning, J. Hendriks, R. Holland, and S. Ruys, "MR characterization of suspicious breast lesions with a gadolinium-enhanced TurboFLASH subtraction technique," Radiology 193, 777-781 (1994).

11. H. T. Le‐Petross, G. J. Whitman, D. P. Atchley, Y. Yuan, A. Gutierrez‐Barrera, G. N. Hortobagyi, J. K. Litton, and B. K. Arun, "Effectiveness of alternating

152

mammography and magnetic resonance imaging for screening women with deleterious BRCA mutations at high risk of breast cancer," Cancer 117, 3900-3907 (2011).

12. A. J. Rijnsburger, I.-M. Obdeijn, R. Kaas, M. M. Tilanus-Linthorst, C. Boetes, C. E. Loo, M. N. Wasser, E. Bergers, T. Kok, and S. H. Muller, "BRCA1-associated breast cancers present differently from BRCA2-associated and familial cases: long-term follow-up of the Dutch MRISC Screening Study," Journal of clinical oncology 28, 5265-5273 (2010).

13. F. Sardanelli, F. Podo, G. D'Agnolo, A. Verdecchia, M. Santaquilani, R. Musumeci, G. Trecate, S. Manoukian, S. Morassut, and C. de Giacomi, "Multicenter comparative multimodality surveillance of women at genetic-familial high risk for breast cancer (HIBCRIT study): interim results," Radiology 242, 698-715 (2007).

14. M. T. Alexander, C. E. Loo, J. Wesseling, R. M. Pijnappel, and K. G. Gilhuijs, "Association between rim enhancement of breast cancer on dynamic contrast-enhanced MRI and patient outcome: impact of subtype," Breast cancer research and treatment 148, 541-551 (2014).

15. S. G. Orel, and M. D. Schnall, "MR imaging of the breast for the detection, diagnosis, and staging of breast cancer," Radiology 220, 13-30 (2001).

16. T. Sørlie, C. M. Perou, R. Tibshirani, T. Aas, S. Geisler, H. Johnsen, T. Hastie, M. B. Eisen, M. Van De Rijn, and S. S. Jeffrey, "Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications," Proceedings of the National Academy of Sciences 98, 10869-10874 (2001).

17. B. Fisher, A. Brown, E. Mamounas, S. Wieand, A. Robidoux, R. G. Margolese, A. Cruz, E. R. Fisher, D. L. Wickerham, and N. Wolmark, "Effect of preoperative chemotherapy on local-regional disease in women with operable breast cancer: findings from National Surgical Adjuvant Breast and Bowel Project B-18," Journal of Clinical Oncology 15, 2483-2493 (1997).

18. K. W. Hance, W. F. Anderson, S. S. Devesa, H. A. Young, and P. H. Levine, "Trends in inflammatory breast carcinoma incidence and survival: the surveillance, epidemiology, and end results program at the National Cancer Institute," Journal of the National Cancer Institute 97, 966-975 (2005).

19. P. Rastogi, S. J. Anderson, H. D. Bear, C. E. Geyer, M. S. Kahlenberg, A. Robidoux, R. G. Margolese, J. L. Hoehn, V. G. Vogel, and S. R. Dakhil, "Preoperative chemotherapy: updates of national surgical adjuvant breast and bowel project protocols B-18 and B-27," Journal of Clinical Oncology 26, 778-785 (2008).

20. A. B. Chagpar, L. P. Middleton, A. A. Sahin, P. Dempsey, A. U. Buzdar, A. N. Mirza, F. C. Ames, G. V. Babiera, B. W. Feig, and K. K. Hunt, "Accuracy of physical examination, ultrasonography, and mammography in predicting residual

153

pathologic tumor size in patients treated with neoadjuvant chemotherapy," Annals of surgery 243, 257 (2006).

21. C. E. Loo, M. E. Straver, S. Rodenhuis, S. H. Muller, J. Wesseling, M.-J. T. V. Peeters, and K. G. Gilhuijs, "Magnetic resonance imaging response monitoring of breast cancer during neoadjuvant chemotherapy: relevance of breast cancer subtype," Journal of Clinical Oncology 29, 660-666 (2011).

22. B. T. Miller, A. M. Abbott, and T. M. Tuttle, "The influence of preoperative MRI on breast cancer treatment," Ann Surg Oncol 19, 536-540 (2012).

23. D. Groheux, S. Giacchetti, M. Espié, D. Rubello, J.-l. Moretti, and E. Hindié, "Early monitoring of response to neoadjuvant chemotherapy in breast cancer with 18F-FDG PET/CT: defining a clinical aim," European journal of nuclear medicine and molecular imaging 38, 419-425 (2011).

24. M.-L. W. Ah-See, A. Makris, N. J. Taylor, M. Harrison, P. I. Richman, R. J. Burcombe, J. J. Stirling, J. A. d'Arcy, D. J. Collins, and M. R. Pittam, "Early changes in functional dynamic magnetic resonance imaging predict for pathologic response to neoadjuvant chemotherapy in primary breast cancer," Clinical Cancer Research 14, 6580-6589 (2008).

25. A. Berriolo-Riedinger, C. Touzery, J.-M. Riedinger, M. Toubeau, B. Coudert, L. Arnould, C. Boichot, A. Cochet, P. Fumoleau, and F. Brunotte, "[18F] FDG-PET predicts complete pathological response of breast cancer to neoadjuvant chemotherapy," European journal of nuclear medicine and molecular imaging 34, 1915-1924 (2007).

26. M. Cutler, "Transillumination of the breast," Annals of surgery 93, 223 (1931).

27. A. Cerussi, N. Shah, D. Hsiang, A. Durkin, J. Butler, and B. J. Tromberg, "In vivo absorption, scattering, and physiologic properties of 58 malignant breast tumors determined by broadband diffuse optical spectroscopy," J Biomed Opt 11, 044005 (2006).

28. A. Corlu, R. Choe, T. Durduran, M. A. Rosen, M. Schweiger, S. R. Arridge, M. D. Schnall, and A. G. Yodh, "Three-dimensional in vivo fluorescence diffuse optical tomography of breast cancer in humans," Opt Express 15, 6696-6716 (2007).

29. S. P. Poplack, K. D. Paulsen, A. Hartov, P. M. Meaney, B. W. Pogue, T. D. Tosteson, M. R. Grove, S. K. Soho, and W. A. Wells, "Electromagnetic breast imaging: average tissue property values in women with negative clinical findings," Radiology 231, 571-580 (2004).

30. S. Jiang, B. W. Pogue, C. M. Carpenter, S. P. Poplack, W. A. Wells, C. A. Kogel, J. A. Forero, L. S. Muffly, G. N. Schwartz, K. D. Paulsen, and P. A. Kaufman, "Evaluation of breast tumor response to neoadjuvant chemotherapy with tomographic diffuse optical spectroscopy: case studies of tumor region-of-interest changes," Radiology 252, 551-560 (2009).

154

31. M. Bartek, X. Wang, W. Wells, K. D. Paulsen, and B. W. Pogue, "Estimation of subcellular particle size histograms with electron microscopy for prediction of optical scattering in breast tissue," J Biomed Opt 11, 064007 (2006).

32. X. Wang, B. W. Pogue, S. Jiang, H. Dehghani, X. Song, S. Srinivasan, B. A. Brooksby, K. D. Paulsen, C. Kogel, S. P. Poplack, and W. A. Wells, "Image reconstruction of effective Mie scattering parameters of breast tissue in vivo with near-infrared tomography," J Biomed Opt 11, 041106 (2006).

33. M. L. Flexman, M. A. Khalil, R. Al Abdi, H. K. Kim, C. J. Fong, E. Desperito, D. L. Hershman, R. L. Barbour, and A. H. Hielscher, "Digital optical tomography system for dynamic breast imaging," Journal of biomedical optics 16, 076014-076014-076016 (2011).

34. S. van de Ven, S. Elias, A. Wiethoff, M. van der Voort, A. Leproux, T. Nielsen, B. Brendel, L. Bakker, M. van der Mark, W. Mali, and P. Luijten, "Diffuse optical tomography of the breast: initial validation in benign cysts," Mol Imaging Biol 11, 64-70 (2009).

35. H. Jiang, N. V. Iftimia, Y. Xu, J. A. Eggert, L. L. Fajardo, and K. L. Klove, "Near-infrared optical imaging of the breast with model-based reconstruction," Academic radiology 9, 186-194 (2002).

36. V. Krishnaswamy, K. E. Michaelsen, B. W. Pogue, S. P. Poplack, I. Shaw, K. Defrietas, K. Brooks, and K. D. Paulsen, "A digital x-ray tomosynthesis coupled near infrared spectral tomography system for dual-modality breast imaging," Opt Express 20, 19125-19136 (2012).

37. S. R. Arridge, and W. R. Lionheart, "Nonuniqueness in diffusion-based optical tomography," Optics letters 23, 882-884 (1998).

38. P. Taroni, A. Torricelli, L. Spinelli, A. Pifferi, F. Arpaia, G. Danesini, and R. Cubeddu, "Time-resolved optical mammography between 637 and 985 nm: clinical study on the detection and identification of breast lesions," Phys Med Biol 50, 2469-2488 (2005).

39. D. Grosenick, K. T. Moesta, M. Möller, J. Mucke, H. Wabnitz, B. Gebauer, C. Stroszczynski, B. Wassermann, P. M. Schlag, and H. Rinneberg, "Time-domain scanning optical mammography: I. Recording and assessment of mammograms of 154 patients," Physics in medicine and biology 50, 2429 (2005).

40. T. D. O’Sullivan, A. E. Cerussi, D. J. Cuccia, and B. J. Tromberg, "Diffuse optical imaging using spatially and temporally modulated light," Journal of biomedical optics 17, 0713111-07131114 (2012).

41. D. Roblyer, T. D O’Sullivan, R. V. Warren, and B. J. Tromberg, "Feasibility of direct digital sampling for diffuse optical frequency domain spectroscopy in tissue," Measurement Science and Technology 24, 045501 (2013).

155

42. H. Ban, M. Schweiger, V. Kavuri, J. Cochran, L. Xie, D. Busch, J. Katrašnik, S. Pathak, S. Chung, and K. Lee, "Heterodyne frequency‐domain multispectral diffuse optical tomography of breast cancer in the parallel‐plane transmission geometry," Medical Physics 43, 4383-4395 (2016).

43. B. J. Tromberg, Z. Zhang, A. Leproux, T. D O'Sullivan, A. E. Cerussi, P. Carpenter, R. S. Mehta, D. Roblyer, W. Yang, and K. D. Paulsen, "Predicting Responses to Neoadjuvant Chemotherapy in Breast Cancer: ACRIN 6691 Trial of Diffuse Optical Spectroscopic Imaging (DOSI)," Cancer Research, canres. 0346.2016 (2016).

44. Q. Zhu, A. Ricci Jr, P. Hegde, M. Kane, E. Cronin, A. Merkulov, Y. Xu, B. Tavakoli, and S. Tannenbaum, "Assessment of functional differences in malignant and benign breast lesions and improvement of diagnostic accuracy by using US-guided diffuse optical tomography in conjunction with conventional US," Radiology 280, 387-397 (2016).

45. M. A. Mastanduno, J. Xu, F. El-Ghussein, S. Jiang, H. Yin, Y. Zhao, K. Wang, F. Ren, J. Gui, and B. W. Pogue, "MR-guided Near Infrared Spectral Tomography Increases Diagnostic Performance of Breast MRI," Clinical Cancer Research, clincanres. 2546.2014 (2015).

46. B. Brooksby, S. Jiang, H. Dehghani, B. W. Pogue, K. D. Paulsen, C. Kogel, M. Doyley, J. B. Weaver, and S. P. Poplack, "Magnetic resonance-guided near-infrared tomography of the breast," Review of Scientific Instruments 75, 5262-5270 (2004).

47. T. O. McBride, B. W. Pogue, S. Poplack, S. Soho, W. A. Wells, S. Jiang, U. L. Osterberg, and K. D. Paulsen, "Multispectral near-infrared tomography: a case study in compensating for water and lipid content in hemoglobin imaging of the breast," J Biomed Opt 7, 72-79 (2002).

48. S. Srinivasan, B. W. Pogue, C. Carpenter, P. K. Yalavarthy, and K. Paulsen, "A boundary element approach for image-guided near-infrared absorption and scatter estimation," Med Phys 34, 4545-4557 (2007).

49. H. Dehghani, M. E. Eames, P. K. Yalavarthy, S. C. Davis, S. Srinivasan, C. M. Carpenter, B. W. Pogue, and K. D. Paulsen, "Near infrared optical tomography using NIRFAST: Algorithm for numerical model and image reconstruction," Communications in numerical methods in engineering 25, 711-732 (2009).

50. M. Jermyn, H. Ghadyani, M. A. Mastanduno, W. Turner, S. C. Davis, H. Dehghani, and B. W. Pogue, "Fast segmentation and high-quality three-dimensional volume mesh creation from medical images for diffuse optical tomography," J Biomed Opt 18, 86007 (2013).

51. J. Wang, S. Jiang, Z. Li, R. M. diFlorio-Alexander, R. J. Barth, P. A. Kaufman, B. W. Pogue, and K. D. Paulsen, "In vivo quantitative imaging of normal and cancerous breast tissue using broadband diffuse optical tomography," Med Phys 37, 3715-3724 (2010).

156

52. J. Wang, B. W. Pogue, S. Jiang, and K. D. Paulsen, "Near-infrared tomography of breast cancer hemoglobin, water, lipid, and scattering using combined frequency domain and cw measurement," Opt Lett 35, 82-84 (2010).

53. C. M. Carpenter, B. W. Pogue, S. Jiang, J. Wang, B. A. Hargreaves, R. Rakow-Penner, B. L. Daniel, and K. D. Paulsen, "MR water quantitative priors improves the accuracy of optical breast imaging," IEEE Trans Med Imaging 30, 159-168 (2011).

54. C. M. Carpenter, R. Rakow-Penner, S. Jiang, B. W. Pogue, G. H. Glover, and K. D. Paulsen, "Monitoring of hemodynamic changes induced in the healthy breast through inspired gas stimuli with MR-guided diffuse optical imaging," Med Phys 37, 1638-1646 (2010).

55. C. M. Carpenter, S. Srinivasan, B. W. Pogue, and K. D. Paulsen, "Methodology development for three-dimensional MR-guided near infrared spectroscopy of breast tumors," Opt Express 16, 17903-17914 (2008).

56. M. A. Mastanduno, J. Xu, F. El-Ghussein, S. Jiang, H. Yin, Y. Zhao, K. E. Michaelson, K. Wang, F. Ren, and B. W. Pogue, "Sensitivity of MRI-guided near-infrared spectroscopy clinical breast exam data and its impact on diagnostic performance," Biomedical optics express 5, 3103-3115 (2014).

57. S. Jiang, B. W. Pogue, K. E. Michaelsen, M. Jermyn, M. A. Mastanduno, T. E. Frazee, P. A. Kaufman, and K. D. Paulsen, "Pilot study assessment of dynamic vascular changes in breast cancer with near-infrared tomography from prospectively targeted manipulations of inspired end-tidal partial pressure of oxygen and carbon dioxide," J Biomed Opt 18, 76011 (2013).

58. H. M. Kuerer, A. A. Sahin, K. K. Hunt, L. A. Newman, T. M. Breslin, F. C. Ames, M. I. Ross, A. U. Buzdar, G. N. Hortobagyi, and S. E. Singletary, "Incidence and impact of documented eradication of breast cancer axillary lymph node metastases before surgery in patients treated with neoadjuvant chemotherapy," Annals of surgery 230, 72 (1999).

59. H. M. Kuerer, L. A. Newman, T. L. Smith, F. C. Ames, K. K. Hunt, K. Dhingra, R. L. Theriault, G. Singh, S. M. Binkley, and N. Sneige, "Clinical course of breast cancer patients with complete pathologic primary tumor and axillary lymph node response to doxorubicin-based neoadjuvant chemotherapy," Journal of Clinical Oncology 17, 460-460 (1999).

60. D. A. Boas, D. H. Brooks, E. L. Miller, C. A. DiMarzio, M. Kilmer, R. J. Gaudette, and Q. Zhang, "Imaging the body with diffuse optical tomography," IEEE signal processing magazine 18, 57-75 (2001).

61. A. Ishimaru, Wave propagation and scattering in random media (Academic press New York, 1978).

157

62. S. R. Arridge, M. Cope, and D. T. Delpy, "The theoretical basis for the determination of optical pathlengths in tissue: temporal and frequency analysis," Phys Med Biol 37, 1531-1560 (1992).

63. S. R. Arridge, M. Schweiger, M. Hiraoka, and D. T. Delpy, "A finite element approach for modeling photon transport in tissue," Med Phys 20, 299-309 (1993).

64. P. Gonzalez-Rodriguez, and A. D. Kim, "Comparison of light scattering models for diffuse optical tomography," Opt Express 17, 8756-8774 (2009).

65. S. T. Flock, M. S. Patterson, B. C. Wilson, and D. R. Wyman, "Monte Carlo modeling of light propagation in highly scattering tissues. I. Model predictions and comparison with diffusion theory," IEEE Transactions on Biomedical Engineering 36, 1162-1168 (1989).

66. M. J. Saxton, "Anomalous diffusion due to obstacles: a Monte Carlo study," Biophysical journal 66, 394-401 (1994).

67. X. Wang, and L. V. Wang, "Propagation of polarized light in birefringent turbid media: a Monte Carlo study," J Biomed Opt 7, 279-290 (2002).

68. S. Arridge, M. Schweiger, M. Hiraoka, and D. Delpy, "A finite element approach for modeling photon transport in tissue," Medical physics 20, 299-309 (1993).

69. S. R. Arridge, P. van der Zee, M. Cope, and D. T. Delpy, "Reconstruction methods for infrared absorption imaging," in Optics, Electro-Optics, and Laser Applications in Science and Engineering(International Society for Optics and Photonics1991), pp. 204-215.

70. A. N. Tikhonov, V. I. A. k. Arsenin, and F. John, Solutions of ill-posed problems (Winston Washington, DC, 1977).

71. P. C. Hansen, Rank-deficient and discrete ill-posed problems: numerical aspects of linear inversion (SIAM, 1998).

72. C. Groetsch, "The theory of Tikhonov Regularization for Fredholm Equations," 104p, Boston Pitman Publication (1984).

73. S. Srinivasan, B. W. Pogue, S. Jiang, H. Dehghani, and K. D. Paulsen, "Spectrally constrained chromophore and scattering near-infrared tomography provides quantitative and robust reconstruction," Appl Opt 44, 1858-1869 (2005).

74. B. Brooksby, S. Srinivasan, S. Jiang, H. Dehghani, B. W. Pogue, K. D. Paulsen, J. Weaver, C. Kogel, and S. P. Poplack, "Spectral priors improve near-infrared diffuse tomography more than spatial priors," Opt Lett 30, 1968-1970 (2005).

75. B. J. Pichler, A. Kolb, T. Nägele, and H.-P. Schlemmer, "PET/MRI: paving the way for the next generation of clinical multimodality imaging applications," Journal of Nuclear Medicine 51, 333-336 (2010).

158

76. R. L. Barbour, H. L. Graber, J. Chang, S.-L. Barbour, P. C. Koo, and R. Aronson, "MRI-guided optical tomography: prospects and computation for a new imaging method," IEEE computational science and engineering 2, 63-77 (1995).

77. B. W. Pogue, and K. D. Paulsen, "High-resolution near-infrared tomographic imaging simulations of the rat cranium by use of a priori magnetic resonance imaging structural information," Opt Lett 23, 1716-1718 (1998).

78. V. Ntziachristos, A. G. Yodh, M. D. Schnall, and B. Chance, "MRI-guided diffuse optical spectroscopy of malignant and benign breast lesions," Neoplasia 4, 347-354 (2002).

79. L. Zhang, Y. Zhao, S. Jiang, B. W. Pogue, and K. D. Paulsen, "Direct regularization from co-registered anatomical images for MRI-guided near-infrared spectral tomographic image reconstruction," Biomedical optics express 6, 3618-3630 (2015).

80. J. Feng, S. Jiang, J. Xu, Y. Zhao, B. W. Pogue, and K. D. Paulsen, "Multiobjective guided priors improve the accuracy of near-infrared spectral tomography for breast imaging," Journal of biomedical optics 21, 090506-090506 (2016).

81. S. L. Jacques, "Optical properties of biological tissues: a review," Physics in medicine and biology 58, R37 (2013).

82. W.-F. Cheong, S. A. Prahl, and A. J. Welch, "A review of the optical properties of biological tissues," IEEE journal of quantum electronics 26, 2166-2185 (1990).

83. V. Peters, D. Wyman, M. Patterson, and G. Frank, "Optical properties of normal and diseased human breast tissues in the visible and near infrared," Physics in medicine and biology 35, 1317 (1990).

84. D. A. Boas, A. M. Dale, and M. A. Franceschini, "Diffuse optical imaging of brain activation: approaches to optimizing image sensitivity, resolution, and accuracy," Neuroimage 23 Suppl 1, S275-288 (2004).

85. S. R. Arridge, and M. Schweiger, "Photon-measurement density functions. Part 2: Finite-element-method calculations," Applied Optics 34, 8026-8037 (1995).

86. H. J. Van Staveren, C. J. Moes, J. van Marie, S. A. Prahl, and M. J. Van Gemert, "Light scattering in lntralipid-10% in the wavelength range of 400–1100 nm," Applied optics 30, 4507-4514 (1991).

87. J. R. Mourant, T. Fuselier, J. Boyer, T. M. Johnson, and I. J. Bigio, "Predictions and measurements of scattering and absorption over broad wavelength ranges in tissue phantoms," Appl Opt 36, 949-957 (1997).

88. J. Wang, S. C. Davis, S. Srinivasan, S. Jiang, B. W. Pogue, and K. D. Paulsen, "Spectral tomography with diffuse near-infrared light: inclusion of broadband frequency domain spectral data," J Biomed Opt 13, 041305 (2008).

159

89. P. K. Yalavarthy, B. W. Pogue, H. Dehghani, and K. D. Paulsen, "Weight-matrix structured regularization provides optimal generalized least-squares estimate in diffuse optical tomography," Med Phys 34, 2085-2098 (2007).

90. B. A. Brooksby, H. Dehghani, B. W. Pogue, and K. D. Paulsen, "Near-infrared (NIR) tomography breast image reconstruction with a priori structural information from MRI: algorithm development for reconstructing heterogeneities," IEEE Journal of selected topics in quantum electronics 9, 199-209 (2003).

91. P. C. Hansen, "Truncated singular value decomposition solutions to discrete ill-posed problems with ill-determined numerical rank," SIAM Journal on Scientific and Statistical Computing 11, 503-518 (1990).

92. P. C. Hansen, "Analysis of discrete ill-posed problems by means of the L-curve," SIAM review 34, 561-580 (1992).

93. B. J. Tromberg, B. W. Pogue, K. D. Paulsen, A. G. Yodh, D. A. Boas, and A. E. Cerussi, "Assessing the future of diffuse optical imaging technologies for breast cancer management," Med Phys 35, 2443-2451 (2008).

94. V. Ntziachristos, A. G. Yodh, M. Schnall, and B. Chance, "Concurrent MRI and diffuse optical tomography of breast after indocyanine green enhancement," Proc Natl Acad Sci U S A 97, 2767-2772 (2000).

95. B. Brooksby, B. W. Pogue, S. Jiang, H. Dehghani, S. Srinivasan, C. Kogel, T. D. Tosteson, J. Weaver, S. P. Poplack, and K. D. Paulsen, "Imaging breast adipose and fibroglandular tissue molecular signatures by using hybrid MRI-guided near-infrared spectral tomography," Proc Natl Acad Sci U S A 103, 8828-8833 (2006).

96. M. A. Mastanduno, F. El-Ghussein, S. Jiang, R. DiFlorio-Alexander, X. Junqing, Y. Hong, B. W. Pogue, and K. D. Paulsen, "Adaptable Near-Infrared Spectroscopy Fiber Array for Improved Coupling to Different Breast Sizes During Clinical MRI," Acad Radiol 21, 141-150 (2014).

97. S. R. Arridge, "Optical tomography in medical imaging," Inverse problems 15, R41 (1999).

98. J. Singer, F. A. Grunbaum, P. Kohn, and J. P. Zubelli, "Image reconstruction of the interior of bodies that diffuse radiation," Science 248, 990-993 (1990).

99. B. W. Pogue, T. O. McBride, J. Prewitt, U. L. Österberg, and K. D. Paulsen, "Spatially variant regularization improves diffuse optical tomography," Applied optics 38, 2950-2961 (1999).

100. K. D. Paulsen, and H. Jiang, "Spatially varying optical property reconstruction using a finite element diffusion equation approximation," Medical Physics 22, 691-701 (1995).

101. J. P. Culver, R. Choe, M. J. Holboke, L. Zubkov, T. Durduran, A. Slemp, V. Ntziachristos, B. Chance, and A. G. Yodh, "Three-dimensional diffuse optical tomography in the parallel plane transmission geometry: evaluation of a hybrid

160

frequency domain/continuous wave clinical system for breast imaging," Med Phys 30, 235-247 (2003).

102. R. J. Gaudette, D. H. Brooks, C. A. DiMarzio, M. E. Kilmer, E. L. Miller, T. Gaudette, and D. A. Boas, "A comparison study of linear reconstruction techniques for diffuse optical tomographic imaging of absorption coefficient," Phys Med Biol 45, 1051-1070 (2000).

103. A. Li, E. L. Miller, M. E. Kilmer, T. J. Brukilacchio, T. Chaves, J. Stott, Q. Zhang, T. Wu, M. Chorlton, R. H. Moore, D. B. Kopans, and D. A. Boas, "Tomographic optical breast imaging guided by three-dimensional mammography," Appl Opt 42, 5181-5190 (2003).

104. B. Brooksby, S. Jiang, H. Dehghani, B. W. Pogue, K. D. Paulsen, J. Weaver, C. Kogel, and S. P. Poplack, "Combining near-infrared tomography and magnetic resonance imaging to study in vivo breast tissue: implementation of a Laplacian-type regularization to incorporate magnetic resonance structure," J Biomed Opt 10, 051504 (2005).

105. J. Carpenter, and J. Bithell, "Bootstrap confidence intervals: when, which, what? A practical guide for medical statisticians," Statistics in medicine 19, 1141-1164 (2000).

106. E. R. DeLong, D. M. DeLong, and D. L. Clarke-Pearson, "Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach," Biometrics, 837-845 (1988).

107. F. S. Azar, K. Lee, A. Khamene, R. Choe, A. Corlu, S. D. Konecky, F. Sauer, and A. G. Yodh, "Standardized platform for coregistration of nonconcurrent diffuse optical and magnetic resonance breast images obtained in different geometries," J Biomed Opt 12, 051902 (2007).

108. R. De Blasi, S. Fantini, M. Franceschini, M. Ferrari, and E. Gratton, "Cerebral and muscle oxygen saturation measurement by frequency-domain near-infra-red spectrometer," Medical and Biological Engineering and Computing 33, 228-230 (1995).

109. T. J. Farrell, M. S. Patterson, and B. Wilson, "A diffusion theory model of spatially resolved, steady‐state diffuse reflectance for the noninvasive determination of tissue optical properties invivo," Medical physics 19, 879-888 (1992).

110. C. D'Andrea, L. Spinelli, A. Bassi, A. Giusto, D. Contini, J. Swartling, A. Torricelli, and R. Cubeddu, "Time-resolved spectrally constrained method for the quantification of chromophore concentrations and scattering parameters in diffusing media," Opt Express 14, 1888-1898 (2006).

111. P. Taroni, D. Comelli, A. Pifferi, A. Torricelli, and R. Cubeddu, "Absorption of collagen: effects on the estimate of breast composition and related diagnostic implications," J Biomed Opt 12, 014021 (2007).

161

112. Y. Zhao, M. A. Mastanduno, S. Jiang, F. El-Ghussein, J. Xu, J. Gui, B. W. Pogue, and K. D. Paulsen, "Systematic optimization of MRI guided near infrared diffuse optical spectroscopy in breast," in SPIE BiOS(International Society for Optics and Photonics2015), pp. 931605-931605-931608.

113. Y. Zhao, M. A. Mastanduno, S. Jiang, E.-G. Fadi, J. Gui, B. W. Pogue, and K. D. Paulsen, "Optimization of image reconstruction for magnetic resonance imaging–guided near-infrared diffuse optical spectroscopy in breast," Journal of biomedical optics 20, 056009-056009 (2015).

114. S. Jiang, B. W. Pogue, A. M. Laughney, C. A. Kogel, and K. D. Paulsen, "Measurement of pressure-displacement kinetics of hemoglobin in normal breast tissue with near-infrared spectral imaging," Appl Opt 48, D130-136 (2009).

115. (!!! INVALID CITATION !!! [13, 14]).

116. H. Soliman, A. Gunasekara, M. Rycroft, J. Zubovits, R. Dent, J. Spayne, M. J. Yaffe, and G. J. Czarnota, "Functional imaging using diffuse optical spectroscopy of neoadjuvant chemotherapy response in women with locally advanced breast cancer," Clin Cancer Res 16, 2605-2614 (2010).

117. D. Roblyer, S. Ueda, A. Cerussi, W. Tanamai, A. Durkin, R. Mehta, D. Hsiang, J. A. Butler, C. McLaren, and W.-P. Chen, "Optical imaging of breast cancer oxyhemoglobin flare correlates with neoadjuvant chemotherapy response one day after starting treatment," Proceedings of the National Academy of Sciences 108, 14626-14631 (2011).

118. M. A. Mastanduno, S. Jiang, R. Diflorio-Alexander, B. W. Pogue, and K. D. Paulsen, "Automatic and robust calibration of optical detector arrays for biomedical diffuse optical spectroscopy," Biomed Opt Express 3, 2339-2352 (2012).

119. M. A. Mastanduno, S. Jiang, R. DiFlorio-Alexander, B. W. Pogue, and K. D. Paulsen, "Remote positioning optical breast magnetic resonance coil for slice-selection during image-guided near-infrared spectroscopy of breast cancer," J Biomed Opt 16, 066001 (2011).

120. M. A. Mastanduno, F. El-Ghussein, S. Jiang, R. DiFlorio-Alexander, X. Junqing, Y. Hong, B. W. Pogue, and K. D. Paulsen, "Adaptable near-infrared spectroscopy fiber array for improved coupling to different breast sizes during clinical MRI," Academic radiology 21, 141-150 (2014).

121. D. R. Leff, O. J. Warren, L. C. Enfield, A. Gibson, T. Athanasiou, D. K. Patten, J. Hebden, G. Z. Yang, and A. Darzi, "Diffuse optical imaging of the healthy and diseased breast: a systematic review," Breast cancer research and treatment 108, 9-22 (2008).

122. J. Wang, S. Jiang, Z. Li, R. J. Barth, P. A. Kaufman, B. W. Pogue, and K. D. Paulsen, "In vivo quantitative imaging of normal and cancerous breast tissue

162

using broadband diffuse optical tomography," Medical physics 37, 3715-3724 (2010).

123. S. Jiang, B. W. Pogue, C. M. Carpenter, S. P. Poplack, W. A. Wells, C. A. Kogel, J. A. Forero, L. S. Muffly, G. N. Schwartz, and K. D. Paulsen, "Evaluation of breast tumor response to neoadjuvant chemotherapy with tomographic diffuse optical spectroscopy: Case studies of tumor region-of-interest changes 1," Radiology 252, 551-560 (2009).

124. C. M. Carpenter, B. W. Pogue, S. Jiang, H. Dehghani, X. Wang, K. D. Paulsen, W. A. Wells, J. Forero, C. Kogel, J. B. Weaver, S. P. Poplack, and P. A. Kaufman, "Image-guided optical spectroscopy provides molecular-specific information in vivo: MRI-guided spectroscopy of breast cancer hemoglobin, water, and scatterer size," Opt Lett 32, 933-935 (2007).

125. Y. Zhao, B. W. Pogue, S. J. Haider, J. Gui, K. D. Paulsen, and S. Jiang, "Portable, parallel 9-wavelength near-infrared spectral tomography (NIRST) system for efficient characterization of breast cancer within the clinical oncology infusion suite," Biomedical Optics Express 7, 2186-2201 (2016).

126. E. Gratton, S. Fantini, M. A. Franceschini, G. Gratton, and M. Fabiani, "Measurements of scattering and absorption changes in muscle and brain," Philosophical Transactions of the Royal Society B: Biological Sciences 352, 727-735 (1997).

127. J. C. Hebden, "Advances in optical imaging of the newborn infant brain," Psychophysiology 40, 501-510 (2003).

128. B. Chance, S. Nioka, J. Zhang, E. F. Conant, E. Hwang, S. Briest, S. G. Orel, M. D. Schnall, and B. J. Czerniecki, "Breast cancer detection based on incremental biochemical and physiological properties of breast cancers: a six-year, two-site Study1," Academic radiology 12, 925-933 (2005).

129. B. C. Wilson, P. J. Muller, and J. C. Yanch, "Instrumentation and light dosimetry for intra-operative photodynamic therapy (PDT) of malignant brain tumours," Phys Med Biol 31, 125-133 (1986).

130. C. H. Contag, and B. D. Ross, "It's not just about anatomy: in vivo bioluminescence imaging as an eyepiece into biology," Journal of magnetic resonance imaging 16, 378-387 (2002).

131. J. G. Fujimoto, C. Pitris, S. A. Boppart, and M. E. Brezinski, "Optical coherence tomography: an emerging technology for biomedical imaging and optical biopsy," Neoplasia 2, 9-25 (2000).

132. B. W. Pogue, and M. S. Patterson, "Review of tissue simulating phantoms for optical spectroscopy, imaging and dosimetry," J Biomed Opt 11, 041102 (2006).

133. L. I. Grossweiner, "Optical dosimetry in photodynamic therapy," Lasers in surgery and medicine 6, 462-466 (1986).

163

134. H. Wang, T. B. Huff, D. A. Zweifel, W. He, P. S. Low, A. Wei, and J.-X. Cheng, "In vitro and in vivo two-photon luminescence imaging of single gold nanorods," Proceedings of the National Academy of Sciences of the United States of America 102, 15752-15756 (2005).

135. V. Ntziachristos, "Fluorescence molecular imaging," Annu. Rev. Biomed. Eng. 8, 1-33 (2006).

136. D. Huang, E. A. Swanson, C. P. Lin, J. S. Schuman, W. G. Stinson, W. Chang, M. R. Hee, T. Flotte, K. Gregory, and C. A. Puliafito, "Optical coherence tomography," Science (New York, NY) 254, 1178 (1991).

137. E. L. Hull, M. G. Nichols, and T. H. Foster, "Quantitative broadband near-infrared spectroscopy of tissue-simulating phantoms containing erythrocytes," Physics in medicine and biology 43, 3381 (1998).

138. M. Firbank, and D. Delpy, "A design for a stable and reproducible phantom for use in near infra-red imaging and spectroscopy," Physics in medicine and biology 38, 847 (1993).

139. U. Sukowski, F. Schubert, D. Grosenick, and H. Rinneberg, "Preparation of solid phantoms with defined scattering and absorption properties for optical tomography," Phys Med Biol 41, 1823-1844 (1996).

140. R. Bays, G. Wagnières, D. Robert, J. F. Theumann, A. Vitkin, J. F. Savary, P. Monnier, and H. van den Bergh, "Three‐dimensional optical phantom and its application in photodynamic therapy," Lasers in surgery and medicine 21, 227-234 (1997).

141. G. Beck, N. Akgün, A. Rück, and R. Steiner, "Design and characterisation of a tissue phantom system for optical diagnostics," Lasers in medical science 13, 160-171 (1998).

142. B. W. Pogue, C. Willscher, T. O. McBride, U. L. Osterberg, and K. D. Paulsen, "Contrast-detail analysis for detection and characterization with near-infrared diffuse tomography," Med Phys 27, 2693-2700 (2000).

143. S. Jiang, B. W. Pogue, T. O. McBride, and K. D. Paulsen, "Quantitative analysis of near-infrared tomography: sensitivity to the tissue-simulating precalibration phantom," J Biomed Opt 8, 308-315 (2003).

144. Y. Zhao, W. R. Burger, M. Zhou, B. W. Pogue, and K. D. P. S. Jiang, "A portable, 12-wavelength parallel near-infrared spectral tomography (NIRST) system for efficient characterization of breast cancer during neoadjuvant chemotherapy," in Proc. of SPIE Vol(2017), pp. 100590R-100591.

145. S. Alowami, S. Troup, S. Al-Haddad, I. Kirkpatrick, and P. H. Watson, "Mammographic density is related to stroma and stromal proteoglycan expression," Breast Cancer Research 5, R129 (2003).

164

146. Y.-P. Guo, L. J. Martin, W. Hanna, D. Banerjee, N. Miller, E. Fishell, R. Khokha, and N. F. Boyd, "Growth factors and stromal matrix proteins associated with mammographic densities," Cancer Epidemiology and Prevention Biomarkers 10, 243-248 (2001).

147. P. Taroni, A. M. Paganoni, F. Ieva, A. Pifferi, G. Quarto, F. Abbate, E. Cassano, and R. Cubeddu, "Non-invasive optical estimate of tissue composition to differentiate malignant from benign breast lesions: A pilot study," Scientific Reports 7, 40683 (2017).

148. P. Taroni, D. Comelli, A. Pifferi, A. Torricelli, and R. Cubeddu, "Absorption of collagen: effects on the estimate of breast composition and related diagnostic implications," Journal of Biomedical Optics 12, 014021-014021-014024 (2007).

149. G. Quarto, L. Spinelli, A. Pifferi, A. Torricelli, R. Cubeddu, F. Abbate, N. Balestreri, S. Menna, E. Cassano, and P. Taroni, "Estimate of tissue composition in malignant and benign breast lesions by time-domain optical mammography," Biomedical optics express 5, 3684-3698 (2014).

150. P. Taroni, A. Pifferi, E. Salvagnini, L. Spinelli, A. Torricelli, and R. Cubeddu, "Seven-wavelength time-resolved optical mammography extending beyond 1000 nm for breast collagen quantification," Optics express 17, 15932-15946 (2009).

151. P. Taroni, A. Pifferi, G. Quarto, A. Farina, F. Ieva, A. M. Paganoni, F. Abbate, E. Cassano, and R. Cubeddu, "Time domain diffuse optical spectroscopy: In vivo quantification of collagen in breast tissue," in SPIE Optical Metrology(International Society for Optics and Photonics2015), pp. 952910-952910-952918.

152. P. Taroni, G. Danesini, A. Torricelli, A. Pifferi, L. Spinelli, and R. Cubeddu, "Clinical trial of time-resolved scanning optical mammography at 4 wavelengths between 683 and 975 nm," J Biomed Opt 9, 464-473 (2004).

153. J. E. Johnson, T. Takenaka, and T. Tanaka, "Two-dimensional time-domain inverse scattering for quantitative analysis of breast composition," IEEE transactions on biomedical engineering 55, 1941-1945 (2008).

154. C. Byrne, C. Schairer, J. Wolfe, N. Parekh, M. Salane, L. A. Brinton, R. Hoover, and R. Haile, "Mammographic features and breast cancer risk: effects with time, age, and menopause status," JNCI: Journal of the National Cancer Institute 87, 1622-1629 (1995).

155. N. F. Boyd, L. J. Martin, M. Bronskill, M. J. Yaffe, N. Duric, and S. Minkin, "Breast tissue composition and susceptibility to breast cancer," Journal of the National Cancer Institute 102, 1224-1237 (2010).

156. T. D O'Sullivan, A. Leproux, J.-H. Chen, S. Bahri, A. Matlock, D. Roblyer, C. E. McLaren, W.-P. Chen, A. E. Cerussi, and M.-Y. Su, "Optical imaging correlates with magnetic resonance imaging breast density and revealscomposition changes during neoadjuvant chemotherapy," Breast Cancer Research 15, R14 (2013).

165

157. K. M. Blackmore, J. A. Knight, J. Walter, and L. Lilge, "The Association between Breast Tissue Optical Content and Mammographic Density in Pre-and Post-Menopausal Women," PloS one 10 (2015).

158. Q. Fang, S. A. Carp, J. Selb, G. Boverman, Q. Zhang, D. B. Kopans, R. H. Moore, E. L. Miller, D. H. Brooks, and D. A. Boas, "Combined optical imaging and mammography of the healthy breast: optical contrast derived from breast structure and compression," IEEE Trans Med Imaging 28, 30-42 (2009).

159. N. Shah, A. E. Cerussi, D. Jakubowski, D. Hsiang, J. Butler, and B. J. Tromberg, "Spatial variations in optical and physiological properties of healthy breast tissue," J Biomed Opt 9, 534-540 (2004).

160. L. Spinelli, A. Torricelli, A. Pifferi, P. Taroni, G. M. Danesini, and R. Cubeddu, "Bulk optical properties and tissue components in the female breast from multiwavelength time-resolved optical mammography," J Biomed Opt 9, 1137-1142 (2004).

161. S. Srinivasan, B. W. Pogue, S. Jiang, H. Dehghani, C. Kogel, S. Soho, J. J. Gibson, T. D. Tosteson, S. P. Poplack, and K. D. Paulsen, "Interpreting hemoglobin and water concentration, oxygen saturation, and scattering measured in vivo by near-infrared breast tomography," Proc Natl Acad Sci U S A 100, 12349-12354 (2003).

162. B. J. Tromberg, N. Shah, R. Lanning, A. Cerussi, J. Espinoza, T. Pham, L. Svaasand, and J. Butler, "Non-invasive in vivo characterization of breast tumors using photon migration spectroscopy," Neoplasia 2, 26-40 (2000).

163. R. Choe, A. Corlu, K. Lee, T. Durduran, S. D. Konecky, M. Grosicka-Koptyra, S. R. Arridge, B. J. Czerniecki, D. L. Fraker, A. DeMichele, B. Chance, M. A. Rosen, and A. G. Yodh, "Diffuse optical tomography of breast cancer during neoadjuvant chemotherapy: a case study with comparison to MRI," Med Phys 32, 1128-1139 (2005).

164. A. M. Shannon, D. J. Bouchier-Hayes, C. M. Condron, and D. Toomey, "Tumour hypoxia, chemotherapeutic resistance and hypoxia-related therapies," Cancer treatment reviews 29, 297-307 (2003).

165. A. Cerussi, D. Hsiang, N. Shah, R. Mehta, A. Durkin, J. Butler, and B. J. Tromberg, "Predicting response to breast cancer neoadjuvant chemotherapy using diffuse optical spectroscopy," Proc Natl Acad Sci U S A 104, 4014-4019 (2007).

166. R. W. Carlson, and A. M. Favret, "Multidisciplinary management of locally advanced breast cancer," The breast journal 5, 303-307 (1999).

167. L. Gianni, W. Eiermann, V. Semiglazov, A. Manikhas, A. Lluch, S. Tjulandin, M. Zambetti, F. Vazquez, M. Byakhow, and M. Lichinitser, "Neoadjuvant chemotherapy with trastuzumab followed by adjuvant trastuzumab versus neoadjuvant chemotherapy alone, in patients with HER2-positive locally

166

advanced breast cancer (the NOAH trial): a randomised controlled superiority trial with a parallel HER2-negative cohort," The Lancet 375, 377-384 (2010).

168. B. Fisher, J. Bryant, N. Wolmark, E. Mamounas, A. Brown, E. R. Fisher, D. L. Wickerham, M. Begovic, A. DeCillis, and A. Robidoux, "Effect of preoperative chemotherapy on the outcome of women with operable breast cancer," Journal of Clinical Oncology 16, 2672-2685 (1998).

169. G. Hortobagyi, "Comprehensive management of locally advanced breast cancer," Cancer 66, 1387-1391 (1990).

170. M. Schegerin, A. N. Tosteson, P. A. Kaufman, K. D. Paulsen, and B. W. Pogue, "Prognostic imaging in neoadjuvant chemotherapy of locally-advanced breast cancer should be cost-effective," Breast Cancer Res Treat 114, 537-547 (2009).

171. Q. Zhu, P. A. DeFusco, A. Ricci Jr, E. B. Cronin, P. U. Hegde, M. Kane, B. Tavakoli, Y. Xu, J. Hart, and S. H. Tannenbaum, "Breast cancer: assessing response to neoadjuvant chemotherapy by using US-guided near-infrared tomography," Radiology 266, 433-442 (2013).

172. R. Dent, M. Trudeau, K. I. Pritchard, W. M. Hanna, H. K. Kahn, C. A. Sawka, L. A. Lickley, E. Rawlinson, P. Sun, and S. A. Narod, "Triple-negative breast cancer: clinical features and patterns of recurrence," Clin Cancer Res 13, 4429-4434 (2007).

173. S. Thomsen, and D. Tatman, "Physiological and pathological factors of human breast disease that can influence optical diagnosis," Annals of the New York Academy of Sciences 838, 171-193 (1998).

174. M. Hockel, and P. Vaupel, "Tumor hypoxia: definitions and current clinical, biologic, and molecular aspects," J Natl Cancer Inst 93, 266-276 (2001).

175. D. M. Brizel, S. P. Scully, J. M. Harrelson, L. J. Layfield, J. M. Bean, L. R. Prosnitz, and M. W. Dewhirst, "Tumor oxygenation predicts for the likelihood of distant metastases in human soft tissue sarcoma," Cancer Res 56, 941-943 (1996).

176. S. Kennedy, J. Geradts, T. Bydlon, J. Q. Brown, J. Gallagher, M. Junker, W. Barry, N. Ramanujam, and L. Wilke, "Optical breast cancer margin assessment: an observational study of the effects of tissue heterogeneity on optical contrast," Breast Cancer Res 12, R91 (2010).

177. J. R. Mourant, M. Canpolat, C. Brocker, O. Esponda-Ramos, T. M. Johnson, A. Matanock, K. Stetter, and J. P. Freyer, "Light scattering from cells: the contribution of the nucleus and the effects of proliferative status," Journal of biomedical optics 5, 131-137 (2000).

178. R. Hornung, T. H. Pham, K. A. Keefe, M. W. Berns, Y. Tadir, and B. J. Tromberg, "Quantitative near-infrared spectroscopy of cervical dysplasia in vivo," Hum Reprod 14, 2908-2916 (1999).

167

179. B. W. Pogue, S. Jiang, H. Dehghani, C. Kogel, S. Soho, S. Srinivasan, X. Song, T. D. Tosteson, S. P. Poplack, and K. D. Paulsen, "Characterization of hemoglobin, water, and NIR scattering in breast tissue: analysis of intersubject variability and menstrual cycle changes," J Biomed Opt 9, 541-552 (2004).

180. Y. Zhao, W. R. Burger, M. Zhou, E. B. Bernhardt, P. A. Kaufman, R. R. Patel, C. V. Angeles, B. W. Pogue, K. D. Paulsen, and S. Jiang, "Collagen quantification in breast tissue using a 12-wavelength near infrared spectral tomography (NIRST) system," Biomedical Optics Express 8, 4217-4229 (2017).