an investigation of clinical treatment field delivery …...imrt field delivery, vertical and...

An investigation of clinical treatment field delivery verification usingcherenkov imaging: IMRT positioning shifts and field matching

Paul J. Blacka)Department of Radiation Oncology, Columbia University, New York, NY 10032, USADepartment of Radiation Oncology, Novant Health, Winston-Salem, NC 27103, USA

Christian Velten, Yi-Fang Wang, and Yong Hum NaDepartment of Radiation Oncology, Columbia University, New York, NY 10032, USA

Cheng-Shie WuuDepartment of Radiation Oncology, Columbia University, New York, NY 10032, USADepartment of Applied Physics and Applied Mathematics, Columbia University, New York, NY 10027, USA

(Received 29 May 2018; revised 1 October 2018; accepted for publication 10 October 2018;published xx xxxx xxxx)

Purpose: Cherenkov light emission has been shown to correlate with ionizing radiation dose deliv-ery in solid tissue. An important clinical application of Cherenkov light is the real-time verificationof radiation treatment delivery in vivo. To test the feasibility of treatment field verification, Cheren-kov light images were acquired concurrent with radiation beam delivery to standard and anthropo-morphic phantoms. Specifically, we tested two clinical treatment scenarios: (a) Observation of fieldoverlaps or gaps in matched 3D fields and (b) Patient positioning shifts during intensity modulatedradiation therapy (IMRT) field delivery. Further development of this technique would allow real-timedetection of treatment delivery errors on the order of millimeters so that patient safety and treatmentquality can be improved.Methods: Cherenkov light emission was captured using a PI-MAX4 intensified charge coupled device(ICCD) system (Princeton Instruments). All radiation delivery was performed using a Varian Trilogy lin-ear accelerator (linac) operated at 6 MVor 18 MV for photon and 6 MeVor 16 MeV for electron stud-ies. Field matching studies were conducted with photon and electron beams at gantry angles of 0°, 15°,and 45°. For each modality and gantry angle, a total of three data sets were acquired. Overlap and gapdistances of 0, 2, 5, and 10 mm were tested and delivered to solid phantom material of30 9 30 9 5 cm3. Phantom materials used were white plastic water and brown solid water. Tests wereadditionally performed on an anthropomorphic phantomwith an irregular surface. Positioning shift stud-ies were performed using IMRT fields delivered to a thoracic anthropomorphic phantom. For thoracicphantom measurements, the camera was placed laterally to observe the entire right side of the phantom.Fields were delivered with known translational patient positioning shifts in four directions. Changes inthe Cherenkov fluence were evaluated through the generation of difference maps from unshifted Cheren-kov images. All images were evaluated using ImageJ, Python, and MATLAB software packages.Results: For matched fields, Cherenkov images were able to quantitate matched field separationswith discrepancies between 2 and 4 mm, depending on gantry angle and beam energy or modality.For all photon and electron beams delivered at a gantry angle of 0°, image analysis indicated averagediscrepancies of less than 2 mm for all field gaps and overlaps, with 83% of matched fields exhibit-ing discrepancies less than 1 mm. Beams delivered obliquely to the phantom surface exhibited aver-age discrepancies as high as 4 mm for electron beams delivered at large oblique angles. Finally, forIMRT field delivery, vertical and lateral patient positioning shifts of 2 mm were detected in somecases, indicating the potential detectability threshold of using this technique alone.Conclusions: Our study indicates that Cherenkov imaging can be used to support and bolster currenttreatment delivery verification techniques, improving our ability to recognize and rectify millimeter-scale delivery and positioning errors. © 2018 American Association of Physicists in Medicine[https://doi.org/10.1002/mp.13250]

Key words: cherenkov, real-time monitoring, treatment verification

1. INTRODUCTION

In the practice of clinical radiation therapy, it is vital thatpatient positioning and beam targeting match the simulated

situation used by the treatment planning system. Large errorsin patient positioning or treatment delivery parameters couldresult in serious patient injury or inappropriate dose deliveredto the disease site. Because of this, a number of techniques

1 Med. Phys. 0 (0), xxxx 0094-2405/xxxx/0(0)/1/xx © 2018 American Association of Physicists in Medicine 1

https://doi.org/10.1002/mp.13250

and safeguards have been established to prevent these poten-tial errors. These include image guidance,1,2 immobilizationdevices,3,4 procedure timeouts, treatment plan checks, andmachine quality assurance (QA).5 Recently, more advancedverification techniques have included 4D targeting verifica-tion.6 Examples of these techniques have been the introduc-tion of respiratory gating,7,8 motion tracking,9,10 and livepatient positioning verification based on patient surface map-ping using optical light.11 EPID based patient position track-ing has also made great advancement in recent years. Thistechnique allows for real-time monitoring of radiation fluenceafter exiting a patient and techniques have been developed toreconstruct the dose delivered to a patient at each fraction.12

This technique is an important competitor to Cherenkov-based treatment verification but facing difficulty when tryingto determine surface dose and cannot be used in any clinicalsituation in which an imaging panel cannot be employed.Moreover, EPID based dosimetry systems are limited to pho-ton irradations, as electron fields would not have sufficientfluence upon exiting a patient to be used by a portal imagingsystem. With the exception of the more recently developed4D systems, conventional treatment verification techniquesoccur prior to radiation delivery and cannot detect deviationsfrom simulated setup during treatment delivery. Real-timepositioning verification techniques are currently available,such as fluoroscopy, optical, and magnetic resonance linearaccelerator (linac) systems. The use of fluoroscopy in anexternal beam setting is presently a matter of debate due tothe increased patient dose.13 Magnetic resonance linacs canprovide real time treatment verification with good tissue con-trast; however, their use in radiation oncology is still low.14

Optical verification techniques are the most widely adoptedto date, however, they are only capable of tracking patientposition in real time and do not directly verify radiation flu-ence during treatment. Of the 4D techniques available, all aredesigned to verify the patient’s position based on the estab-lished coordinate system, and do not evaluate live changes inradiation deposition during treatment.6 Cherenkov imaginghas the capability to supplant or enhance current real-timeverification techniques through its unique ability to monitorradiation fluence in real time.15

Since its discovery and characterization, Cherenkov radia-tion has been observed in many situations in which ionizingradiation interacts with matter.16–19 An important characteris-tic of Cherenkov radiation is the finding that its intensity canbe directly correlated with absorbed dose under controlledconditions or most easily in cases of irradiated homogeneousmaterials.20–22 In the context of radiation therapy, Cherenkovradiation has the unique capability of observing live changesin the radiation fluence during treatment delivery.23,24 Earlywork on potential clinical applications of Cherenkov radiationwas originally demonstrated in water and tissue equivalentphantoms.19,22,25,26 Later patients receiving radiation therapyhave been used for proof of concept studies.15,27 Medicalphysics applications focused on Cherenkov radiation,described as Cherenkoscopy originally by Jarvis et al.,15,27

have yielded a number of important clinical applications.

These have included verification of whole breast treatmentdelivery in live patients,15 verification of dose homogeneityfor total body irradiation and total skin electron therapy,28 3Dreconstruction of intensity modulated radiation therapy(IMRT) treatment fields,21,29 quantitation of tissue oxygena-tion using Cherenkov excited phosphorescence,25 Cherenkovemission from positron emitting radiotracers,30,31 and poten-tial applications in molecular imaging.32,33

In addition, many of the limitations associated with Cher-enkov imaging have been addressed and mitigated. The lowrelative intensity of Cherenkov emission to ambient lighthas been improved through the introduction of gating thecamera’s electronic shutter directly off of the linear accelera-tor target current.19 Variation in Cherenkov intensity due toangle of observation has been improved through image postprocessing corrections.22 Low frame rate Cherenkov imageacquisition has been improved through better optimized cam-era systems and sensor binning.34 In addition, work has beenperformed to determine optimal Cherenkov imaging systems,including camera selection and lens choice.35 This study wasdesigned to improve upon the current developments in Cher-enkoscopy. All work was focused on external beam photonand electron delivery using a clinical linear accelerator (linac)as a radiation source.

The first major focus of this study was the clinical applica-tion of matched fields. Field matching is used in many clini-cal scenarios in radiation oncology. These include cases suchas comprehensive breast irradiation with lymph node involve-ment. In these cases, up to five fields are matched on the sur-face of the skin to give uniform coverage of disease sitesinternally.36 In addition, craniospinal irradiation (CSI)involves the matching of two whole brain fields and one totwo spinal fields, depending on the anatomy of the patient.37

Here, it is crucial to cover the brain and spinal cord withoutdelivering hot spots to nerve tissue. Doing so could result inmyelopathy or even fatal nerve toxicities. Due to the potentialdangers associated with setup errors with this technique, clin-ical setups are performed with physician and physicist assis-tance. Furthermore, gaps between the treatment fields areestablished with an inter-fraction feathering technique to min-imize the possibility of an accidental hotspot in the spinalcord.37 With this in mind, we explored the clinical feasibilityof using Cherenkov imaging to detect potential errors in fieldmatching, which can be acquired during treatment field deliv-ery. Given that typical field junctions of 10 mm are employedclinically, the ability to verify field positioning within 2 mmis ideal, with only stereotactic radiosurgery or stereotacticbody radiotherapy (SBRT) treatments requiring positioningcertainty within 1 mm. Therefore, this has the potential todetect field matching errors live in vivo and enable thepatient’s plan to be adjusted to prevent potential injury.

The second major focus of this study was the clinicalapplication of verification of patient positioning duringIMRT delivery. IMRT is employed in cases where target cov-erage becomes challenging while trying to meet organ at risk(OAR) constraints set by the physician. IMRT uses multi leafcollimator (MLC) systems to modulate the radiation fluence

Medical Physics, 0 (0), xxxx

2 Black et al.: Field delivery verification using cherenkov 2

during treatment delivery to achieve optimal target coveragewhile adequately sparing OARs.38 This technique requiresaccurate patient positioning within 2 mm of error in anydirection to avoid potential injury or misadministration.5 InIMRT cases, treatment delivery verification occurs thoughaforementioned established techniques in addition to patientspecific QA performed before treatment begins.39

It is important to note that patient positioning shift detec-tion using Cherenkov imaging has been investigated previ-ously by other groups. One such work, published in 2014used comprehensive breast irradiation fields, utilizing bothphantom and live patient treatment deliveries to quantitativelydetermined shifts based on a rigid matrix registration usingstandard 3 mm translational tolerances, and 3° rotational tol-erances. Their results demonstrate that these shifts weredetectable. This study used Cherenkoscopy with AlignRTused as a verification technique, an optical patient positioningverification system, to verify positioning shift errors real timeduring patient treatment.27 This study was focused on wholebreast irradiation. The goal of our investigation was to deter-mine the smallest detectable positioning shift using Cheren-kov imaging alone. In addition, this study employed a largervariety of treatment plans, specifically lung SBRT using 6MV beams, brain IMRT using 6 MV beams, and prostateIMRT using 18 MV beams. In addition, we developed a tech-nique for evaluating difference maps of the Cherenkov radia-tion fluence to improve our ability to observe shifts whenthey occur.

This work explores the feasibility and technical limitationsof using Cherenkov imaging to aid in real time in vivo verifi-cation of external beam radiation therapy delivery in clini-cally relevant scenarios. Results presented here establish adetection threshold when using Cherenkov imaging alone toevaluate field matching and patient positioning errors.

2. MATERIALS AND METHODS

2.A. Radiation delivery

All radiation treatment fields were delivered using a Var-ian Trilogy linear accelerator (linac) operated at 6 MV or18 MV accelerating potential for photons; 6 MeV or16 MeV accelerating potential for electrons. All matchedfield results presented in this work were performed using a5 cm thick plastic water phantom, which is off-white incolor. Additional preliminary results were acquired in solidwater phantoms (Best Medical Company, Nashville, TN,USA), which is brown in color, and a thorax anthropomor-phic phantom (CIRS, Norfolk, VA, USA). Photon fieldswere delivered using collimator jaws set to 10 9 10 cm2 atisocenter, 100 cm Source to Surface Distance (SSD) to thephantom. An image of this setup for this set of experimentsis provided in Fig. 1(a). It should be noted that the gantryangle selected in Fig. 1(a) was not an angle selected forirradiation, and was chosen to make all element of thesetup easily visible. All electron fields were delivered usinga 10 9 10 cm2 cone, 110 cm SSD to the surface. The

larger SSD for electron field delivery was selected to main-tain a camera field of view that included the entire top sur-face of the plastic water phantom. For all matched fieldexperiments, 100 Monitor Units (MU) were delivered perfield at a rate of 600 MU/min. Photon-photon and electron-electron matched fields were delivered with known fieldspacing established using projected light fields and calcu-lated table shifts. The following matched field separationswere tested: 10, 5, 2, and 0 mm gaps or overlaps. Prelimi-nary data testing 1 mm gaps and overlaps revealed thatthese distances were not detectable; as such, this distancewas not included in this investigation. In addition, theeffect of oblique gantry angles on the quantitation of radia-tion field spacing was tested. On top of anterior-posterior(AP) field deliveries (gantry angle = 0°), 15°, and 45° gan-try angles were tested as well. This simulates AP and AP-oblique treatment deliveries. For all matched field experi-ments, Cherenkov images were acquired of the phantomsurface that the radiation beam was entering.

For experiments focused on the real-time detectabilitylimit of patient positioning errors, IMRT fields selected fordelivery were chosen to sample the variety of plan parame-ters delivered in the clinic. This included a standard frac-tionation 6 MV brain IMRT plan, a 6 MV lung SBRTplan, and a standard fractionation 18 MV prostate IMRTplan. These selected plans enabled the effect of the follow-ing parameters on detection threshold to be evaluated: (a)The two available photon energies on this linac, 6 and18 MV and (b) Higher MU fields vs standard MU fields.IMRT field delivery was performed using an anthropomor-phic thorax phantom. In these experiments, the phantomwas centered at the linac mechanical isocenter. Fields weredelivered using the treatment parameters originally used totreat the patient. This includes gantry angles, collimatorangles, beam energy (6 or 18 MV accelerating potential),dose rate (600 MU/min), and MLC motions. Figure 1(b)provides a photo of the setup used for this set of experi-ments. In addition, the field of view was set to incorporatethe entire right side of the phantom by placing the cameralateral to the phantom with a downward facing angle. Fig-ure 1 provides an image of the setup for both the flat phan-tom and anthropomorphic phantom experiments.

Plans were delivered at a known starting position with themechanical isocenter set to the phantom center. These imageswere designated origin images and used for all shift compar-isons. Plans were then redelivered with known positioningshifts along three independent directions: anterior, superior,and left lateral. Additional plan deliveries were performedwith a combined positioning shift in all three directions. Posi-tioning shift distances tested were 10, 5, 3, and 2 mm.Finally, after acquiring Cherenkov images of IMRT fieldsdelivered at the designated shifted positions, the thorax phan-tom was repositioned to its origin position. This was done toestablish control images to account for image differences dueto inherent table positioning errors. All phantom shifts wereperformed by moving the treatment couch. All Cherenkovimages were acquired with the room lights off.



2.B. Image acquisition

All images were obtained using a PIMAX-4 camera sys-tem (Princeton Instruments, Princeton, NJ, USA) featuring anintensified charge coupled device (ICCD) sensor capable of aminimum gate width of 4 ns and a minimum gate delay of25 ns. For our purposes, the gate width was matched to thepulse width used by our linac (2.5|ls) and the gate delay wasset to the minimum. This was accomplished by coupling thetrigger port of our camera via a triaxial cable connector to thetarget current or forward power port of the linac for photon orelectron delivery, respectively. To optimize the gate widthand delay, a small gate width of 1 ls was selected and thegate delay was increased until the Cherenkov emission inten-sity was reduced. Next, the gate delay was reduced to a mini-mum to confirm that no loss of signal intensity occurred.This verified that a minimum gate delay of 25 ns was optimalto capture a maximum amount of Cherenkov signal. Finally,the gate width was increased until maximum signal intensitywas observed; verifying that the entire Cherenkov pulse wascaptured. Since the effective frame rate of this technique isprincipally limited by the readout time of the CCD sensorarray, it was necessary to bin the array for an optimal framerate. To acquire images at an optimal frame rate without com-promising image quality, the ICCD sensor array was binned2 9 2, effectively lowering the resolution by a factor of two.These settings resulted in an image acquisition frame rate of20|fps. Images for field matching experiments were acquiredat an angle of approximately 40° to a beam delivered at 0°gantry angle. For positioning shift experiments, an angle ofapproximately 70° was selected to maximize the visible areaof the phantom while avoiding collision and direct irradiationissues with the camera system.

Linear accelerators operate at an internal frequency, whichis also the maximum pulse frequency. Our Varian Trilogyaccelerator operates at 180 Hz. At 600 MU

min it takes 18|cyclesto deliver one MU; when gating using the signal from linacforward power, which was employed for electron fields, wefound that using less than six triggered exposures per frameyields missing pulses, that is, frames containing no Cheren-kov light, while more than six yielded both missing and

summed pulses. Thus, for all modalities we used six triggeredexposures to constitute one frame. For our purposes, wefound it more useful to think of frame rate in the context ofMU delivered. While some minor variance was observed fordifferent IMRT fields, these image acquisition parametersresulted in an effective frame rate of ~ 2 frames/MU.

The camera was positioned on the treatment table for all5 cm plastic water slab experiments inferior to the phantom,with a downward facing angle of approximately 40° relativeto the incident beam. This angle has been determined to beoptimal for Cherenkov light emission intensity.40 Thisenables a fixed camera position relative to the phantom for allmatched field shifts. Matched field studies were repeated fora total of three data sets for all parameters tested. For all tho-rax phantom experiments the camera was placed off the table,lateral to the phantom with a downward facing angle ofapproximately 70° relative to an incident beam delivered at0° gantry angle. This camera positioning allowed acquisitionof the entire right side of the phantom surface, which was notideal for Cherenkov emission intensity but was necessary toobtain the proper field of view. Using this fixed positioningresults in the acquisition of Cherenkov images produced by abeam entering a surface or a beam exiting a surface, depend-ing on the gantry angle of the treatment field. All fields thatused angles in which the gantry physically obscured the phan-tom or caused direct irradiation of the camera system wereexcluded and not delivered.

2.C. Image analysis

For each data set, an acquired field-of-view image wasused to extract reference points (edges of the phantom) usedto perform perspective transformations of acquired images tosimulate a beam’s-eye-view. The raw stack was flattenedusing either a median or an average projection type; eachyielded the median or average value, respectively, of eachrespective pixel in the stack.41 Using the previously extractedreference points, a least-square perspective transformation(landmark correspondences) to a reference image is per-formed. The result is cropped to the size of the phantom andsaved to file. From this, a 2D profile was extracted and saved

FIG. 1. Setup photo for (a) 5 cm plastic water slab beam delivery, and (b) 3D thorax phantom beam delivery. In both images the PIMAX-4 camera system ismounted on a tripod to the left and the phantom is positioned on the treatment table to the right.



using a centered rectangle sized to encompass the full fieldwidth and one-twelfth of the field height. These steps wereperformed in ImageJ (Fiji) and automated in a macro requir-ing only the reference data and image directory as input. Inaddition, dialogs were used to enable the selection of differ-ent parameters including an optional background subtractionusing a 100 px radius rolling ball filter. This method uses alarge ball centered on each pixel to average the backroundaround that pixel. For our purposes, we found a rolling ballradius of 100 pixels to be effective.42 Each image is indepen-dently normalized to unity based on the image maximumvalue.

Using a Python script, lateral intensity profiles includingpixel value uncertainties were calculated from the 2D pro-files. These were then normalized such that the minimum andmaximum pixel values corresponded to 0 and 1, respectively.The parameters of the Cherenkov radiation-defined fieldedges were extracted by performing fits to the lateral intensityprofiles using a multi-parameter (pi) function utilizing thePython bindings for ROOT (PyROOT),43

f x; p0; p1; . . .; p8ð Þ ¼ p02

1� tanhx� p1p2

� ��

þX5i¼0

piþ3xi

The first part is a hyperbolic tangent,44,45 modified torange from 0 to p0, describing the left (+) and right (-)edges. The parameters p0 through p2 describe the edgescaling, edge position (point of highest gradient), and edgebroadening, respectively; the range of values for these wereset to [0.1, 1.25], [xmin, xmax], and [0.1, 30]. An eighthorder polynomial enabled increased conformity of the fitfunction in regions left and right from the edge, which wasimportant to retrieve consistent parameter values for thehyperbolic tangent. Its parameters (p3 through p8) were leftto vary freely except for the constant term, which is limitedto [0, 0.3]. The fit range was automatically chosen to bethe central 40% of the profile (in pixels), excluding pointsoutside of this range. These choices in fit ranges yieldedthe most robust results across datasets and modalities.Automatic distinction between left and right field edgeswas performed by v2 discrimination. Fitting was repeatedmultiple times to stabilize the minimization. Using anotherscript, this process was repeated for all profiles and a com-bined result file generated.

To obtain a proper pixel size, an image of a 10 9 10 cm2

light field projected onto a surface at 100 SSD was acquired.Perspective transformation as outlined above was applied tothis image. Finally, a pixel-to-millimeter conversion factorwas calculated for each dataset using the full width at halfmaximum (FWHM) intensity in lateral direction of the trans-formed light field image. With this factor, the edge locationresults of each dataset were converted into millimeter andsubsequently, measured overlaps and gaps calculated. Thismethod was used as it would be an easy and reliable methodfor calibrating distances in a clinical situation without the

need for an additional device. To test the reliability of thismethod of pixel size calibration, we acquired images of threeUS quarter dollars at different positions on the phantom sur-face. Using the known size of these objects we determinedthe pixel calibration to be 1.1 mm/pixel. Using the light fieldmethod of calibrating pixel size, a calibration factor of1.2 mm/pixel was determined. We consider the difference inthese two calibration factors to be within the error of mea-surement of this study and are confident the pixel size cali-bration based on projected light field can be consideredreliable. In the results provided below, all values provided areabsolute discrepancy in mm.

The means and uncertainties of the means of each overlapor gap were calculated for all modalities. For measurementsunder zero gantry angle, the absolute errors (measured lesssetup) and total uncertainties were calculated. The latter aregiven by rD;i ¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffir2i þ r2setup

q, where ri is the uncertainty of

the mean overlap or gap and rsetup is the statistical uncertaintyof the setup. The setup uncertainty is given by the resolutionð0:1 cmÞ of lateral couch positions and jaw-defined fieldsize, contributing 1mmffiffiffiffi

12p each. This yields a setup uncertainty of

rsetup ¼ 1mmffiffi3

p � 0:58mm.

3. RESULTS

3.A. Field matching studies

For all field matching studies, fields of known size weredelivered so that their diverging edges were separated or over-lapped by known distances. Matching photon fields withknown gap or overlap distances were imaged to determine thedetectability limit of small field separations and their agree-ment with the setup parameters. For 6 MV photon fields, dis-tances of 10, 5, 2, and 0 mm were tested. Figure 2 providesCherenkov images of photon field matches with 5 mm gapsor overlaps for 6 MV fields. Figure 3 provides similar resultsfor 2 mm gaps and overlaps, demonstrating a qualitativedetectability limit of 2 mm. Field gaps or overlaps of 1 mmwere not detectable using this imaging method (data notshown). Similarly, examples of results from electron-electronfield matching studies are provided in Figs. 3 and 4. Figure 5illustrates the method we used to fit field edges from theacquired images.

The signal-to-noise ratio (SNR) was exemplarily evaluatedunder 0° gantry angle for 6 MeV, 16 MeV, 6 MV, and18 MV. The signal mean in the central part of the fielddivided by the standard deviation in an equally sized area out-side the field is taken as SNR measure. Using the medianstack flattening method, the SNR for 6 and 16 MeV electronbeams is 1222.2 � 164.2 and 775.56 � 118.2, while for 6and 18 MV photon beams the SNR is 475.3 � 19.7 and158.8 � 23.3, respectively. This shows that the SNR for pho-ton beams decreases with increasing beam energy (� 70%),likely due to the same factors that influence the larger penum-bra for higher energy beams, increased scatter and range ofprimary electrons and to a smaller extent, differences in



electron contamination. A similar effect is seen for electronswith the high-energy SNR being approximately 60% lower.

Using a light field pixel size calibration previouslydescribed, FWHM calculation was performed on all gap andoverlap experiments. These results determined consistent dis-crepancy between known distances and calculated distances asthe gaps and overlaps across all tested distances. In addition,we observed similar discrepancies for 18 MV matched fieldswhen compared to 6 MV matched fields. Table I provides asummary of all gap and overlap distance calculations, whileFigs. 6 and 7 provide a graphical illustration of that data.

Quantitation of photon-photon and electron-electronmatched fields indicated strong agreement between expectedand measured field separations for all AP delivered beams asprovided in Table I; all discrepancies were within 2 mm, with83% of measurements within 1 mm of expected values. Aver-age discrepancies by modality were highest for 6 MeV elec-trons, with 0.91 mm � 0.2 mm. In contrast, there appearedto be no energy dependence for photon-photon matchedfields, with average discrepancies of 0.64 mm � 0.17 mmand 0.35 mm � .07 mm for 6 and 18 MV beams,respectively.

(a) (b)

(c) (d)

FIG. 2. Cherenkov images acquired in real time of 6 MV photon-photon matched fields delivered with AP beams and individual field sizes of 10 9 10 cm2 atthe phantom surface. Matched field gaps were 0 mm at the surface (a), 2 (b), 5 (c), and 10 (d) mm.

(a) (b)

(c) (d)

FIG. 3. Cherenkov images acquired in real time of 6 MV photon-photon matched fields delivered with AP beams and individual field sizes of 10 9 10 cm2 atthe phantom surface. Matched field overlaps were (a) 0 mm at the surface, (b) 2 mm, (c) 5 mm, and (d) 10 mm.



Experiments performed utilizing both small and largeangle oblique fields revealed larger discrepancies, particu-larly with electron-electron matched fields. For photon-photon matched fields, small angle oblique beam deliveriesagreed well with expected field separations. These resultsalso demonstrated minor energy dependence for small angleoblique photon fields. For 6 MV beams, the average dis-crepancy between expectation and measurement was0.48 mm � 0.17 mm while for 18 MV beams at this anglethe average discrepancy was 1.07 mm � 0.24 mm. Asdemonstrated in Table I, all 6 MV photon beams for smallangle oblique beams had discrepancies within 2 mm.18 MV beams exhibited a maximum discrepancy of2.1 mm, however, all other discrepancies were within 2 mm.This observed energy dependence for photon-photonmatched fields had a larger effect for large angle obliquebeams. Specifically, the average discrepancy for 6 MVmatched fields was 2.06 mm � 0.08 mm, with the majority

of these discrepancies having been larger than 2 mm. For18 MV matched fields, these discrepancies increase, with anaverage discrepancy of 2.47 mm � 0.30 mm. It is importantto note that these discrepancies were not random, but sys-tematic for all values, with all measurements based on Cher-enkov images appearing as if fields were 2–3 mm closer toone another than setup.

As demonstrated in Table I, Cherenkov images ofelectron-electron matched fields exhibited larger discrep-ancies compared to photon-photon matched fields forboth small and large angle oblique beams. Specifically,the average discrepancy for electron-electron matchedfields ranged from 3 to 4 mm in the direction of fieldoverlap, and did not appear to have a large angle orenergy dependence. This could be due to the largeheterogeneity in measured Cherenkov light at the surfaceobserved for oblique electron fields, as demonstrated inFig. 8. These findings could indicate that the

(a) (b) (c)

(d) (e) (f)

FIG. 4. Cherenkov images acquired in real time of 6 MeV electron-electron matched fields delivered with AP beams and individual field sizes of 11 9 11 cm2

at the phantom surface. Matched field separations were (a) and (d) 0 mm at the surface, (b) 5 mm gap, (c) 10 mm gap, (e) 5 mm overlap, and (f) 10 mm overlap.

FIG. 5. An example Cherenkov image (left) and the associated lateral profile (points) and fit (line) (right) based on the ROI indicated as a white box in the leftCherenkov image pane.



determination of a systematic correction factor couldaccount for the discrepancies observed for oblique deliv-ered fields for both photon-photon and electron-electronmatched fields.

3.B. IMRT field studies

The Cherenkov image of one delivered IMRT field is pro-vided in Fig. 9. This Cherenkov image is an IMRT field

TABLE I. Quantitative analysis for all AP (0° gantry angle), small oblique (15° gantry angle), and large oblique (45° gantry angle) field images acquired for thisstudy. Discrepancies were evaluated between field edge distance measurements and known gap and overlap distances. All values provided are in units of mm andare difference from expected value. Note that overlaps are all indicated by positive values while gaps are indicated by negative values.

6 MV 18 MV 6 MeV 16 MeV

AP (0° gantry angle)

0 mm 0.06 � 0.47 0.19 � 0.11 1.06 � 0.83 0.54 � 0.94 0 mm

2 mm OVLP 0.09 � 0.67 0.42 � 0.73

2 mm GAP 1.13 � 0.15 0.46 � 0.03

5 mm OVLP 0.75 � 0.50 0.26 � 0.04 1.52 � 0.55 0.47 � 0.33 5 mm OVLP

5 mm GAP 0.70 � 0.36 0.32 � 0.41 0.94 � 0.38 1.99 � 0.28 5 mm GAP

10 mm OVLP 1.28 � 0.27 0.14 � 0.12 0.46 � 0.71 0.11 � 0.92 10 mm OVLP

10 mm GAP 0.60 � 0.28 0.64 � 0.42 0.47 � 0.56 0.07 � 0.66 10 mm GAP

Small oblique (15° gantry angle)

0 mm 0.26 � 0.44 0.29 � 0.52 3.77 � 0.30 3.45 � 0.11 0 mm

2 mm OVLP 0.04 � 0.23 2.61 � 0.39

2 mm GAP 0.9 � 0.43 0.78 � 0.39

5 mm OVLP 0.21 � 0.15 1.46 � 0.54 3.94 � 0.51 3.51 � 0.31 5 mm OVLP

5 mm GAP 0.59 � 0.38 2.07 � 0.64 2.70 � 0.72 2.47 � 0.45 5 mm GAP

10 mm OVLP 0.11 � 0.35 0.49 � 0.48 3.58 � 0.38 3.08 � 0.10 10 mm OVLP

10 mm GAP 1.24 � 0.44 1.37 � 0.31 3.82 � 0.35 3.48 � 0.12 10 mm GAP

Large Oblique (45° Gantry Angle)

0 mm 2.05 � 0.30 2.87 � 0.25 3.80 � 0.83 3.99 � 0.23 0 mm

2 mm OVLP 1.98 � 0.25 2.60 � 0.58

2 mm GAP 1.77 � 1.22 0.79 � 0.15

5 mm OVLP 1.95 � 0.63 3.33 � 0.33 2.67 � 1.06 2.60 � 0.45 5 mm OVLP

5 mm GAP 2.39 � 0.58 2.70 � 0.47 3.33 � 1.00 3.43 � 0.38 5 mm GAP

10 mm OVLP 2.04 � 0.81 2.42 � 0.63 2.81 � 0.68 3.13 � 0.12 10 mm OVLP

10 mm GAP 2.25 � 0.50 2.60 � 0.57 3.45 � 1.20 3.48 � 0.51 10 mm GAP

FIG. 6. Graphical representation of photon data provided in Table I. Generally higher obliquity resulted in larger absolute discrepancies from known fieldspacing.



delivered to the thorax phantom at the origin position exitingthe surface of the phantom. Here, the composite imageincludes an overlay with the field of view of the camera whenwith normal room lighting. Figure 10(a) illustrates the detailsof a difference map, where a positioning shift is detected.Figure 10(b) provides the difference maps created from allpositioning shifts in the anterior direction for an 18 MVIMRT field, including the difference map created from a non-shifted image. In this example, the beam exit side of thephantom surface is imaged. You can see in this figure that allshifts are quantitatively detectable down to 2 mm. For eachIMRT field and shift direction, we determined a detectabilitylimit in mm. This was established using the origin repositionimage as a control. Shifts were identified as detected if thedifference map exhibited a recognizable difference pattern inthe region of the field fluence that could not be attributed tothe random differences observed in the control image. Specif-ically, threshold values were determined based on “return toorigin” images. These images were acquired after a numberof table movements and a return of the table to the originposition. Differences observed here should be due only to thepositioning error of the treatment table. Differences observedin all shifted images that were not greater than these thresholdvalues were not considered to be detected as they could notbe distinguished from table positioning error. Table II pro-vides the detectability limit for all fields tested in all four shiftdirections. We found a number of conditions that had a sig-nificant effect on the signal to noise and the subsequentdetectability limit of a shift. The first important parameterwas energy. 18 MV beams produced Cherenkov images with~29 greater signal to noise than 6 MV beams. Second,whether the surface observed had a beam entering or exitinghad a large effect on signal-to-noise for all energies with

average SNRs observed to be ~39 greater for an exitingbeam compared to an entering beam. In addition, our setupwas most sensitive to anterior shift detection, followed by leftlateral shift detection. This setup was least sensitive to shiftsin the longitudinal direction. This established an ideal casefor shift detection using this method and setup: 18 MV fieldsexiting an observed surface with increased MU. These obser-vations were derived from the beam parameters of the differ-ent plan sites that were tested. For six of the measured beamsfeaturing some or all of these ideal conditions, 2 mm shiftswere detectable. We also observed that combined table shiftshad similar detectability limits to the anterior table shifts,indicating that this observation angle and phantom geometrywere most sensitive to anterior shift detection.

4. DISCUSSION

4.A. Matched field studies

The first of the two clinical scenarios investigated in thisstudy was photon-photon and electron-electron field match-ing. While this study employed only phantom materials, ourfindings can be used in future work focused on photon andelectron field matching. Earlier work by Zhang et al. haseffectively demonstrated that the use of phantom materials inCherenkov studies can be effectively translated to clinical sit-uations. This was demonstrated through a Monte Carlo analy-sis of the optical and radiation transport properties ofphantom materials compared to various modeled tissues.22

Future work will extend these findings to more biologicallyderived systems. Field matching is used clinically in the twotreatment scenarios, we focused on: 3D breast irradiationplans with nodal involvement and cerebral spinal irradiation

FIG. 7. Graphical representation of the electron data provided in Table I. This illustrates significant discrepancy increases with increasing obliquity of incidentbeam angle.



(CSI) plans. In both scenarios, photon-photon field matchingor electron-electron field matching could be performeddepending on the case. In the case of CSI plans, feathering isemployed at the field junction site, using a field spacing oftypically 10 mm at the patient surface.37 This is designed toplace the overlap point and the consequent hot spot anteriorto the spinal cord to avoid possible myelopathy due to spinalcord toxicities. Setup is performed by marking the spacing onthe patient based on light field position. Cherenkov imagingcould provide an improvement to the workflow by verifyingfield junction spacing during beam delivery. This could

detect a patient shift that occurs after setup that the conven-tional method would not detect. Our experiments focused onthe ability of Cherenkov imaging to (a) verify intended fieldjunction gaps or identify insufficient gaps and (b) detect fieldoverlap, which would result in undesired hotspots in thespinal cord. Quantitation of the separation or overlap of Cher-enkov imaged fields provides the detectability limit for detect-ing beam delivery errors. With a robust method ofquantifying field separation errors, it could be possible toextrapolate the dosimetric error for a single treatment and cor-rect for that error in future fractions. These specific clinical

FIG. 8. (a) Cherenkov images acquired in real time of 6 MeV electron-electron matched fields delivered with varying gantry angle and field separation. Allbeams were delivered with a 10 9 10 cm2 cone with 110 SSD to the phantom surface for an AP beam. Gantry angles were 0° (left), 15° (center), and 45° (right).Field separations were matched at the surface (top), 5 mm gap (middle), and 5 mm overlap (bottom). (b) 1D profile comparisons of Gafchromic film vs Cheren-kov intensity.



field verification situations are equally relevant to 3D com-prehensive breast plans in which field matching is crucial togetting full dose coverage of the breast as well as the axillary,internal-mammary, and supraclavicular lymph nodes.36

Our quantitative analysis of matched field Cherenkovimaging yielded results that were overall consistent with theexpected field positions. Using data from three data sets alsoenabled us to provide uncertainties on the calculated meanvalues to verify both, accuracy and precision of the experi-ments and analyses. Under 0° gantry angle all averages werewithin 2 mm, with all but two being within 1 mm; and novalue further from the expected value than three standarddeviations. The distribution of absolute differences betweenmeasured and setup field matching showed no consistent

FIG. 9. A composite Cherenkov image featuring an 18 MV IMRT fieldexiting an anthropomorphic thorax phantom.

FIG. 10. (a) Analysis of all shifted images was performed by creating difference maps for each shift. This provides a visual explanation for how difference mapsshould be analyzed. (b) An original IMRT field delivered to a 3D anthropomorphic thorax phantom (top left) and the difference maps associated with anteriorshifts over distances of 10, 5, 3, and 2 mm for an 18 MV IMRT delivered field. The difference map for no shift is provided (top right) to illustrate the differencesobserved due to random scatter and mechanical positioning errors in treatment table movement.



modality or energy correlations, despite the higher SNR inelectron beam acquisitions. This indicates that the fit functionused to calculate the position of the field edge is robust withrespect to overall signal noise and broadening of the fieldedge, which is observed primarily for electron and high-energy photon beams. These results could change if a non-ideal camera angle is used, which may be the case in a realclinical scenario. In future studies involving live patients, thedifferences in field gap quantification as a function of cameraangle relative to the incident beam will be explored.

To broaden the study beyond simple AP beams to a flatsurface, we repeated our field matching experiments usingoblique beam angles of 15° and 45°. These results yielded anumber of interesting observations. First, the beam modalityplayed a role in the observed field position discrepancies.Namely, photons exhibited higher discrepancies with increas-ing beam obliquity. In contrast, this behavior was not strictlyobserved for electron fields. In the case of electron-electronmatched fields, a marked increase in field position discrep-ancy was observed with oblique beams when compared toAP beams. This discrepancy, however, did not increase withincreasing beam angle or energy.

At nonzero gantry angles, the agreement between setupand measured values was reduced. The Cherenkov light-defined field edges were shifted laterally away from the gan-try head yielding overall larger values than setup. This is dueto the oblique beam delivery which extends the high-intensityfield area in the direction of the beam and broadens the fieldedge. This could be caused by the same beam properties thatinfluence penumbra, such as more laterally scattered elec-trons from further downstream or lower energy electrons,which would then be close enough to the surface such thattheir Cherenkov light emissions could be more easilydetected. This effect would be more pronounced for electronsas the dose buildup region decreases. Photons, in contrast,need to first undergo charged particle producing interactions,

which reduces the density of Cherenkov light-producing par-ticles close to the surface. This explains the observation thatlow-energy electrons exhibit a more spatially restrictedheterogeneity as their shorter range leads to Cherenkov lightgenerated only in a limited portion of the next field.

In addition to the observed angular dependence, it wasfound that the beam energy can have an effect on the mea-sured field position discrepancies in some situations. Thiswas most evident in the case of photons delivered at obliqueangles. 18 MV beams delivered at both 15° and 45° exhibitedhigher field spacing discrepancies than 6 MV beams deliv-ered at identical angles. This could be explained by the dosedeposition qualities of each beam. In the case of photonfields, electrons must be generated within the irradiated mate-rial in order for Cherenkov light to be created. In the case ofAP delivered beams, Cherenkov light is primarily generatedby backscattered and laterally scattered electrons near the sur-face. In the case of oblique beam angles, effective buildupdistance is reduced at the surface. In addition, electrons gen-erated downstream at the edge of the first field generate Cher-enkov light further from the field edge, effectively blurringthe Cherenkov field edge. This phenomenon is likely to bemore pronounced for higher energy beams in which photonshave a longer range than that of a 6 MV beam.

Interestingly, the energy dependence observed for photonsis not observed in the case of electron-electron matchedfields. It is likely that the justification involving the largerrange for higher energy beams is still present, but the effectof this property on field size discrepancy could be smallercompared to other factors. One of these factors is likely theobserved heterogeneity of Cherenkov light intensity observedfor oblique beams. This property is evident in Fig. 8, whichprovides Cherenkov images of 6 MeV beams delivered at allangles tested. It is readily apparent in this figure that thisheterogeneity becomes more pronounced with increasingangle. To address this, a film irradiation was performed by

TABLE II. Detectability thresholds for all treatment fields employed for this study. This setup was most sensitive to shifts in the anterior direction.

Field parameters Detectability limit (mm)Field ID Energy Surface SBRT/Standard Anterior Superior Lateral Combined

P1 18 MV ENTRANCE STANDARD >10 >10 >10 >10

P4 18 MV ENTRANCE STANDARD 5 >10 10 3

P5 18 MV EXIT STANDARD 2 >10 5 2

P6 18 MV EXIT STANDARD 2 >10 3 3

P7 18 MV EXIT STANDARD 5 >10 >10 5

L1 6 MV ENTRANCE SBRT >10 >10 >10 >10



L5 6 MV ENTRANCE SBRT 2 >10 >10 2

L6 6 MV EXIT SBRT 3 >10 5 3

L7 6 MV EXIT SBRT 5 >10 5 2

B1 6 MV ENTRANCE STANDARD >10 >10 >10 >10

B2 6 MV EXIT STANDARD 10 >10 10 10

B3 6 MV EXIT STANDARD 5 10 >10 5

B4 6 MV EXIT STANDARD 3 5 5 3



placing the film at the surface of the phantom and repeatingthe 6 MeV field deliveries at a gantry angle of 45°. Filmanalysis revealed the same heterogeneity in dose distributionat the phantom surface, as demonstrated in Fig. 8(b), but thenormalized dose gradient was larger than the normalizedCherenkov light intensity gradient observed in the sameregion. This indicates that the Cherenkov light intensity isfollowing dosimetric behavior at the surface, however, otherfactors are blurring out this effect somewhat. This translatesto a larger overall discrepancy when quantifying field edgesdefined using Cherenkov light imaging.

It is important to note that for all oblique field data pro-vided in this study, discrepancies were consistently in thedirection of field overlap. Simply put, for oblique beamsimaged with Cherenkov light, fields appear closer togetherthan what the projected light field or the dose distributionwould indicate. This systematic discrepancy was observed forall modalities and energies, while the size of the discrepancywas in some cases affected by beam energy and beam angleas detailed above. In the case of photon-photon matchedfields, this systematic error appeared to have a direct depen-dence on beam angle. This indicates that a correction factorcould be determined with a follow-up study in which moreangles are tested. In contrast, electron-electron fields alsoexhibited a systematic discrepancy favoring overlap, howeverthis effect did not appear to depend directly on the gantryangle or beam energy. This could indicate that a more sim-plistic correction factor could be identified, but again thiswould require further data taken at a larger number of beamangles.

It should be noted that while these results comprehen-sively explored a number of different modalities, energies,and beam geometries, all data was acquired on phantommaterial that was white in color. As such we have to expectthat while these results establish detectability limits and tech-niques for evaluating them, it is possible that results couldvary in cases of different surface color or material withchanges in optical properties. Furthermore, these results willneed to be followed up with studies on phantoms with irregu-lar surfaces. To provide preliminary data on these topics, wecompared Cherenkov penumbra and intensity values betweentwo materials: (a) the white colored plastic water that hasbeen used for all previous field matching experiments pre-sented in this work, and (b) Solid water which is brown incolor. This will give us an idea of how we can expect resultsto change when the optical properties of the surface beingobserved change. Notably, our experiments revealed that thepenumbra is affected by surface color as expected. Specifi-cally, the Cherenkov images acquired from solid water irradi-ations had a penumbra that was 9.0 mm � 0.03 mm. Thiswas larger than the compared plastic water which had apenumbra of 4.9 mm � 0.02 mm. It should be noted thatlarger penumbra does not guarantee larger field discrepan-cies, and we believe that these changes in penumbra can becorrected for in future studies. As expected, the average pixelvalue of Cherenkov images acquired on irradiated solid waterwas lower intensity when compared to plastic water. Solid

water average signal intensity was 18.7% lower. This couldagain affect field spacing measurements, and will be exploredin detail in future studies. Finally, irregular surface geometrywas tested using a 5 mm field gap for both 6 and 18 MVphotons. Measured field gap of acquired Cherenkov imageswas 6.5 and 4.7 mm, respectively. This was encouraging, asthe difference was similar to observed discrepancies for18 MV fields provided earlier. The increase in discrepancyfor 6 MV photons compared to earlier results was unex-pected, but was still quantitatively less than 2 mm. Futurestudies will test more irregular surfaces to further explore theeffect they have on evaluating surface fields spacing usingCherenkov imaging.

4.C. IMRT field studies

The second major focus of this work centered around thedelivery of IMRT fields to a 3D anthropomorphic thoraxphantom. The goal of these experiments was to determine thedetectability limit of positioning shifts along all three Carte-sian axes using established patient IMRT plan fields. A totalof three actual patient plans were employed for this part ofthe study, selected to cover a range of beam parameters. Thisincluded two 6 MV treatment plans, one of which was stan-dard fractionation with approximately 90 MU per field andthe other SBRT with approximately 200 MU per field.Finally, a high-energy (18 MV) IMRT prostate plan wasselected to determine the possible influence of energy on shiftdetectability.

Early studies focused on IMRT field delivery determineda significantly improved signal to noise, typically a factor of2 improvement, when observing a surface through which abeam is exiting compared to a surface upon which a beam isentering. This observation is supported by Monte Carlostudies of a variety of factors affecting Cherenkov intensity,in which Zhang et al. predicts a 50% difference betweenentering vs exiting geometry.46 This makes intuitive sense asthe surface dose is relatively low compared to maximumdose. Specifically, surface dose for a 6 V beam is as can beless than 50% of maximum dose. This behavior is evenmore apparent for higher energy beams. Here, relative sur-face dose for the 18 MV beam can be as low as 30% ofmaximum dose. The observation, however, that exit surfaceCherenkov signal exhibits improved signal-to-noise overentrance surface cannot be solely attributed to the depthdose characteristics. For example, the anthropomorphicphantom used in this study has a maximum diameter of30 cm at midline. For an 18 MV beam, the percent depthdose curves tell us that surface dose is relatively equivalentto surface dose at 30 cm but we observe improved field def-inition and signal-to-noise for an 18 MV beam exiting at30 cm than a beam entering the phantom surface. Moreover,we see improved signal-to-noise for an identical setup usinga 6 MV IMRT field. Depth dose characteristics would pre-dict the opposite to be true as the dose at 30 cm is approxi-mately 25% maximum dose compared to a relative surfacedose of ~50%.



These observations require the consideration of additionalphysical characteristics of Cherenkov light production in thissituation to explain the improved signal-to-noise on beam exitcompared to beam entrance. Zhang et al. propose that themain property contributing to this factor is the difference insampling depth.46 It is also likely that a significant contribut-ing factor to this phenomenon is the directionality of scat-tered electrons produced in the irradiated medium. SinceCherenkov light is produced only by high-energy particles inmatter, we must consider that electrons produced by photonsin these energy ranges are primarily forward scattered.47,48

Indeed as beam energy is increased electrons are further for-ward scattered.47,48 This is likely a contributor to theobserved Cherenkov emission at a beam exiting surface com-pared to a beam entering surface. It may be useful, therefore,to focus on exit surfaces clinically, as beam characteristics atthese surfaces are more easily discernable.

This observation was further explored through attemptinggamma analysis of IMRT fields exiting or entering a plasticwater slab. In both cases Cherenkov images were captured ofthe top surface of the plastic water during IMRT field deliv-ery. Gamma analysis was performed after landscape corre-spondence transformation was performed to minimizegeometric perspective distortions due to the camera observa-tion angle. These experiments revealed that a relative gammaanalysis comparison of our Cherenkov image to the planardose export showed the highest level of agreement for an exit-ing IMRT field for all fields tested, with some results reach-ing the acceptable level of 90% for a 3%, 3 mm threshold.We quickly realized, however, that a robust gamma analysiswould be difficult to achieve due to the nonisotropic emissionof Cherenkov light at the surface and the inherent variancesdue to field modulation, field size, observation angle, andbeam energy.

The principal focus of the experiments performed usingIMRT fields was positioning shift detection through Cheren-kov imaging of an irregular 3D surface. Here, rather thanfocusing on the Cherenkov emission and surface dose charac-teristics of individual fields, we instead concentrated ondetermining the detectability limit of Cherenkov emission dif-ferences observed during a wide variety of situations. To testas clinically relevant a situation as possible, we chose to use afixed camera position relative to the phantom while deliver-ing a full IMRT plan at the original gantry positions. We thenapplied a number of known positioning shifts and reimagedthe same surface during additional IMRT plan deliveries. Thegoal was to determine how small a shift in each direction wasdetectable using this fixed camera setup. As expected fromprevious observations, exiting beams had a lower detectabil-ity limit, meaning smaller shifts were detectable, when com-pared to entering beams. This camera setup was mostsensitive to anterior and combined shifts. As seen inFig. 10(b) and Table II, shifts as small as 2 mm were detect-able in certain situations. These included 18 MV exit beamsshifted in the anterior and combined directions, and 6 MVSBRT entrance beams shifted in the anterior and combineddirections. The lowest detectability threshold observed for

6 MV standard fractionation beams was 3 mm in the anteriorand combined directions for an exiting field.

It is important to note that this was the detectability limitachieved from one fixed camera position. It could be postu-lated that the use of additional cameras at other positionscould improve the detectability threshold for a wide variety ofshifts. This could involve a setup in which the camera systemis mounted to the gantry to maintain a fixed geometry relativeto the radiation beam. In addition, higher resolution camerasystems are available that could improve on this detectabilitythreshold. A Cherenkov system designed such that the cam-eras were optimized to record high resolution, higher framerate images of a patient surface, and have constant fixed posi-tion relative to the linac gantry could introduce largeimprovements to the utility of Cherenkov imaging in externalbeam treatment verification. While these studies utilized ananthropomorphic phantom, only three materials were presentin the phantom: lung, spinal cord, and water equivalent “tis-sue.” This is far more homogeneous than a live patient. Cher-enkov imaging of live patients by other groups27 has revealedfine anatomical details that are simply not present in ananthropomorphic phantom. This includes the resolution ofblood vessel networks, which can dramatically aid in thedetection of positioning shifts. While most of this work hasbeen limited to breast irradiations, the technique we devel-oped in this study could be expanded to nearly any site on thebody, as long as the skin surface is visible to the Cherenkovcamera system. Given these factors, we would anticipateimproved shift detectability due to the larger number of vari-ances in a live patient that would affect the emission of Cher-enkov light in a clinical situation.

Our focus on IMRT fields delivered to a 3D anthropomor-phic phantom has revealed a number of properties importantto the development of Cherenkov imaging as a treatment veri-fication technique. First, higher energy beams have a highersignal to noise and are therefore more sensitive to positioningshift detection. This is expected due to the higher Cherenkovlight yield with higher energy beams observed in other stud-ies.49 Second, beams exiting a surface have a higher SNR ofCherenkov emission when compared to a beam entering asurface. This is likely to be strongly influenced by the direc-tionality of scattered electrons generated in the material beingirradiated. This phenomenon becomes more apparent forhigher energy beams, further supporting this explanation.Finally, IMRT fields delivered with higher MU produce astronger Cherenkov signal. The reason for this is obvious, asit results in higher electron fluence at the surface and subse-quently more Cherenkov light. Considering all of these fac-tors, we determined the beam parameters ideal for IMRTfield detection and measurement using Cherenkov imaging.As such, image quality and the resultant shift detection wouldimprove with increasing beam energy, increasing MU, andobservation of a beam exiting a surface as opposed to enter-ing it. All of these findings are true for the phantom materialthat we tested in these series of experiments. It should benoted that work on other biological or nonbiological materi-als could show some variance in these effects, but the general



properties that we have outlined here should persist. Thisinformation will be valuable in the further development ofthis technique for clinical purposes, and can be incorporatedinto future studies that explore the full utility of this techniquein therapeutic radiation oncology.

5. CONCLUSION

This study builds on the current work in the fieldfocused on the clinical applications of Cherenkov imaging.We determined detectability limits for two clinically rele-vant situations: field matching and patient positioning shiftduring IMRT field delivery. This study indicated that fieldmatching gaps and overlaps were quantifiable for all fieldpositions tested, and that AP beams agreed within 2 mmof expected values. Larger discrepancies were observed foroblique beams, with beam energy and beam angleobserved to have a direct effect on the size of this dis-crepancy in some cases. For all cases where significantdiscrepancies existed, they were systematic, in which Cher-enkov images appeared to have fields placed closertogether than their known spacing. The heterogeneityobserved in Cherenkov light intensity for electrons waslikely a significant contributing factor to these discrepan-cies and was determined to be consistent with the dosi-metric behavior of oblique electron beams at the materialsurface. Further work will be required in developingrobust correction factors for improving on the quantitationof surface distances using Cherenkov light. We were ableto determine that gaps and overlaps as small as 2 mmwere quantifiable using this system for AP beams. Withthe development of robust correction factors in futurestudies, this technique could be used to detect field junc-tion errors in real time, allowing for the possible quantita-tion and correction of dosimetric errors in later treatmentfractions.

Our patient positioning shift experiments while deliveringIMRT fields determined that 2 mm positioning errors weredetectable in certain situations using our fixed camera setup.We propose that this detectability limit would be achievablein all directions through modifications to the camera systemand experimental setup. Our observations on factors mostaffecting the detectability limit, such as entrance vs exitgeometry, beam energy, and total MU are supported by previ-ously published work in the field of Cherenkov imaging.After further development, Cherenkov imaging could beincorporated into the treatment verification tools of a linearaccelerator to further improve on treatment delivery andaccuracy in clinical situations.

CONFLICTS OF INTEREST

The authors report no conflicts of interest.

a)Author to whom correspondence should be addressed. Electronic mail:[email protected]; Telephone: 585-613-8585.

REFERENCES

1. Mackie TR, Kapatoes J, Ruchala K, et al. Image guidance for preciseconformal radiotherapy. Int J Radiat Oncol Biol Phys. 2003;56:89–105.

2. Bushberg JT. The Essential Physics of Medical Imaging. Philadelphia,PA: Lippincott Williams & Wilkins; 2002.

3. Song PY, Washington M, Vaida F, et al. A comparison of four patientimmobilization devices in the treatment of prostate cancer patients withthree dimensional conformal radiotherapy. Int J Radiat Oncol Biol Phys.1996;34:213–219.

4. Negoro Y, Nagata Y, Aoki T, et al. The effectiveness of an immobiliza-tion device in conformal radiotherapy for lung tumor: reduction of respi-ratory tumor movement and evaluation of the daily setup accuracy. Int JRadiat Oncol Biol Phys. 2001;50:889–898.

5. Klein EE, Hanley J, Bayouth J, et al. Task Group 142 report: qualityassurance of medical accelerators. Med Phys. 2009;36:4197–4212.

6. Li G, Citrin D, Camphausen K, et al. Advances in 4D medicalimaging and 4D radiation therapy. Technol Cancer Res Treat.2008;7:67–81.

7. Berson AM, Emery R, Rodriguez L, et al. Clinical experience using res-piratory gated radiation therapy: comparison of free-breathing andbreath-hold techniques. Int J Radiat Oncol Biol Phys. 2004;60:419–426.

8. Ramsey CR, Scaperoth D, Arwood D, Oliver AL. Clinical efficacy ofrespiratory gated conformal radiation therapy. Med Dosim.1999;24:115–119.

9. Keall PJ, Joshi S, Vedam SS, Siebers JV, Kini VR, Mohan R. Four-dimensional radiotherapy planning for DMLC-based respiratory motiontracking.Med Phys. 2005;32:942–951.

10. Seppenwoolde Y, Shirato H, Kitamura K, et al. Precise and real-timemeasurement of 3D tumor motion in lung due to breathing and heart-beat, measured during radiotherapy. Int J Radiat Oncol Biol Phys.2002;53:822–834.

11. Betgen A, Alderliesten T, Sonke J-J, van Vliet-Vroegindeweij C, Barte-link H, Remeijer P. Assessment of set-up variability during deep inspira-tion breath hold radiotherapy for breast cancer patients by 3D-surfaceimaging. Radiother Oncol. 2013;106:225–230.

12. Mans A, Wendling M, McDermott L, et al. Catching errors with in vivoEPID dosimetry. Med Phys. 2010;37:2638–2644.

13. Sihono DSK, Vogel L, Weiß C, et al. A 4D ultrasound real-time trackingsystem for external beam radiotherapy of upper abdominal lesions underbreath-hold. Strahlenther Onkol. 2017;193:213–220.

14. Raaymakers B, J€urgenliemk-Schulz I, Bol G, et al. First patients treatedwith a 1.5 T MRI-Linac: clinical proof of concept of a high-precision,high-field MRI guided radiotherapy treatment. Phys Med Biol. 2017;62:L41.

15. Jarvis L, Zhang R, Andreozzi J, et al. Cherenkoscopy is a novel imagingtechnique with the potential to improve accuracy and detect radiationtherapy misadministrations during whole breast radiation therapy. Int JRadiat Oncol Biol Phys. 2014;90:S249.

16. �Cerenkov P. Visible radiation produced by electrons moving in a med-ium with velocities exceeding that of light. Phys Rev. 1937;52:378.

17. Cherenkov P. The spectrum of visible radiation produced by fast elec-trons. CR Acad Sci URSS. 1938;20:651–655.

18. Tanha K, Pashazadeh AM, Pogue BW. Review of biomedical �Cerenkovluminescence imaging applications. Biomed Opt Express. 2015;6:3053–3065.

19. Glaser AK, Zhang R, Davis SC, Gladstone DJ, Pogue BW. Time-gatedCherenkov emission spectroscopy from linear accelerator irradiation oftissue phantoms. Opt Lett. 2012;37:1193–1195.

20. Glaser AK, Zhang R, Gladstone DJ, Pogue BW. Optical dosimetry ofradiotherapy beams using Cherenkov radiation: the relationship betweenlight emission and dose. Phys Med Biol. 2014;59:3789.

21. Glaser AK, Voigt WH, Davis SC, Zhang R, Gladstone DJ, Pogue BW.Three-dimensional �Cerenkov tomography of energy deposition from ion-izing radiation beams. Opt Lett. 2013;38:634–636.

22. Zhang R, Fox CJ, Glaser AK, Gladstone DJ, Pogue BW. Superficialdosimetry imaging of �Cerenkov emission in electron beam radiotherapyof phantoms. Phys Med Biol. 2013;58:5477.

23. Zhang R, Gladstone DJ, Jarvis LA, et al. Real-time in vivo Cherenko-scopy imaging during external beam radiation therapy. J Biomed Optics.2013;18:110504–110504.



mailto:

24. Zhang R, Jarvis L, Gladstone D, et al. In Vivo Cherenkov Video Imag-ing during External Beam Radiation Therapy. Paper presented at:Biomedical Optics 2014.

25. Holt RW, Zhang R, Esipova TV, et al. Cherenkov excited phosphores-cence-based pO2 estimation during multi-beam radiation therapy: phan-tom and simulation studies. Phys Med Biol. 2014;59:5317.

26. Zhang R, Glaser A, Esipova TV, et al. �Cerenkov radiation emission andexcited luminescence (CREL) sensitivity during external beam radiationtherapy: Monte Carlo and tissue oxygenation phantom studies. BiomedOpt Express. 2012;3:2381–2394.

27. Zhang R, Andreozzi JM, Gladstone DJ, et al. Cherenkoscopy basedpatient positioning validation and movement tracking during post-lum-pectomy whole breast radiation therapy. Phys Med Biol. 2014;60:L1.

28. Andreozzi JM, Zhang R, Gladstone DJ, et al. Cherenkov imagingmethod for rapid optimization of clinical treatment geometry in totalskin electron beam therapy. Med Phys. 2016;43:993–1002.

29. Glaser AK, Andreozzi JM, Zhang R, Pogue BW, Gladstone DJ. Opticalcone beam tomography of Cherenkov-mediated signals for fast 3Ddosimetry of x-ray photon beams in water. Med Phys. 2015;42:4127–4136.

30. Elrick R, Parker R. The use of Cerenkov radiation in the measurementof b-emitting radionuclides. Int J Appl Radiat Isotopes. 1968;19:263–271.

31. Robertson R, Germanos MS, Li C, Mitchell GS, Cherry SR, Silva MD.Optical imaging of Cerenkov light generation from positron-emittingradiotracers. Phys Med Biol. 2009;54:N355.

32. Glaser AK, Zhang R, Andreozzi JM, Gladstone DJ, Pogue BW. Cheren-kov radiation fluence estimates in tissue for molecular imaging and ther-apy applications. Phys Med Biol. 2015;60:6701.

33. Lewis MA, Kodibagkar VD, €Oz OK, Mason RP. On the potential formolecular imaging with Cerenkov luminescence. Opt Lett.2010;35:3889–3891.

34. Glaser AK, Andreozzi JM, Davis SC, et al. Video-rate optical dosimetryand dynamic visualization of IMRT and VMAT treatment plans in waterusing Cherenkov radiation.Med Phys. 2014;41:062102.

35. Andreozzi JM, Zhang R, Glaser AK, Jarvis LA, Pogue BW, Glad-stone DJ. Camera selection for real-time in vivo radiation treatment

verification systems using Cherenkov imaging. Med Phys.2015;42:994–1004.

36. Whelan TJ, Olivotto IA, Parulekar WR, et al. Regional nodal irradiationin early-stage breast cancer. N Engl J Med. 2015;373:307–316.

37. Yoon M, Shin DH, Kim J, et al. Craniospinal irradiation techniques:a dosimetric comparison of proton beams with standard andadvanced photon radiotherapy. Int J Radiat Oncol Biol Phys.2011;81:637–646.

38. Webb S. Intensity-Modulated Radiation Therapy. Boca Raton, FL: CRCPress; 2015.

39. Agazaryan N, Solberg TD, DeMarco JJ. Patient specific quality assur-ance for the delivery of intensity modulated radiotherapy. J Appl ClinMed Phys. 2003;4:40–50.

40. Glaser AK, Davis SC, McClatchy DM, Zhang R, Pogue BW, GladstoneDJ. Projection imaging of photon beams by the �Cerenkov effect. MedPhys. 2013;40:012101.

41. Collins TJ. ImageJ for microscopy. Biotechniques. 2007;43:S25–S30.42. Schindelin J, Arganda-Carreras I, Frise E, et al. Fiji: an open-

source platform for biological-image analysis. Nat Methods.2012;9:676.

43. Antcheva I, Ballintijn M, Bellenot B, et al. ROOT—A C++ frameworkfor petabyte data storage, statistical analysis and visualization. ComputPhys Commun. 2009;180:2499–2512.

44. Wang H, Fallone BG. A mathematical model of radiation field edgelocalization. Med Phys. 1995;22:1107–1110.

45. Nalwa VS, Binford TO. On detecting edges. IEEE Trans Pattern AnalMach Intell. 1986;6:699–714.

46. Zhang R, Glaser AK, Andreozzi J, et al. Beam and tissue factors affect-ing Cherenkov image intensity for quantitative entrance and exit dosime-try on human tissue. J Biophotonics. 2017;10:645–656.

47. Attix FH. Introduction to Radiological Physics and Radiation Dosime-try. Hoboken, NJ: John Wiley & Sons; 2008.

48. Khan FM. The Physics of Radiation Therapy. Pheladelphia, PA: Lippin-cott Williams & Wilkins; 2010.

49. Axelsson J, Davis SC, Gladstone DJ, Pogue BW. Cerenkov emissioninduced by external beam radiation stimulates molecular fluorescence.Med Phys. 2011;38:4127–4132.



an investigation of clinical treatment field delivery …...imrt field delivery, vertical and...

Documents