A Data Driven Application Systemfor

Laser Treatment of Cancer

David Fuentes

Institute for Computational Engineering and SciencesThe University of Texas at Austin

9th US National Congress on Computational MechanicesSan Francisco, California

July 23-26, 2007

Team

- Institute for Computational Engineering and Sciences

J. T. Oden, K. R. Diller, J. C. Browne, C. Bajaj,I. Babuska, J. Bass, L. Demkowicz, Y. Feng, A. Hawkins,S. Koshnevis, B. Kwon, S. Prudhomme, Y. Zhang

- Department of Imaging Physics, M.D. Anderson CancerCenter

J. Hazle, L. Bidaut, A. Elliott, R. J. Stafford

Acknowledgment: NSF grant CNS-0540033, FredericaDarema, Program Director

1

Outline

- Description of Problem

. Development of a Data Driven Control System For LaserTreatment of Cancer

- Cyber Work Flow Description

. Patient-Specific Calibration

- Current Results on Phantom Materials

- Work in Progress

. In Vivo Experiments, real-time laser control, hp-adaptivity studies

2

Problem Description

- Computer guided laser treatment as a minimally invasivealternative to standard treatment of cancer

- Localized hyperthermia/ablation treatment to damage anddestroy cancerous cells

. Heat source provided by diffusing interstitial laser fiber orcollimated external source

- Real-Time Thermal Imaging provides guidance to Real-Time computational prediction

- Target disease: tissues with a well-defined tumor

3

CyberInfrastructure

Animation dddas

4

Work Flow

- Data Acquisition

- Geometry Extraction

- Mesh Generation and Mesh Optimization

- Laser Parameter Optimization

- Registration

- Patient Specific Calibration- Data Filtering

- Predictions

- Visualizations

5

Data Acquisition

- Anatomical Volume Image and Thermal Image DataSets

- Images acquired at the University of Texas M.D. AndersonCancer Center in Houston, Texas

- 1.5-T MR scanner (GE)

- Avg. Data Set Size:

. Anatomical= 3.7MB256x256x30 voxel

. Thermal = .32MB256x256x5 voxel

6

Bioheat Transfer Model

The Non-Linear Pennes Model

ρcp∂u

∂t−∇ · (k(u)∇u) + ω(u)cblood(u− ua)

= Qlaser(x, t) in Ω ⊂ R3

−k(u)∇u · n = h(u− u∞) on ∂Ω

u(x, 0) = u0 in Ω

Qlaser(x) = 3Pµaexp(−µs‖x− x0‖)

4π‖x− x0‖ρ density [ kg

m3] k thermal conductivity [ Wm·K ]

ω blood perfusivity [ kgs·m3] P Power [W ]

cp specific heat [ Jkg·K ] µa,µs coeff. of absorb/scattering [ 1

m]x0 laser position [m]

7

Patient Specific Calibration

- Temperature distribution measured by in vivo MRTI

- Images transferred from Houston to Austin

. avg file size of 1 time instance ≈ .32MB

. avg bandwidth from HOU to AUS ≈ .2MB/s

- Pennes Model update of blood perfusivity and thermalconductivity

‖uh(x, t)− uMRTI(x, t)‖2L2([0,τ ];L2(Ω))

- Uses Quasi-Newton Optimization Method (TAO library)

. Gradient computed from adjoint method

8

Patient Specific Calibration

≈ 8000 Dofs

9

Patient Specific Calibration

< 30 parameters

10

Patient Specific Calibration

≈ 5000 parameters

11

Patient Specific Calibration

- Full Field Inversion too expensive for real time

- Studies show that ≈ 15% decrease of cost function in realtime is feasible

12

Verification Tests using a Phantom

- System operational and being tested on phantom materials

- Phantom material is a radiation sensitive 1% agar gel

Animation Linux Animation Windows

- Heating of phantom visible inMRTI thermal images

- External laser provides heatsource

- Fiducials used to mark laserposition

- Phantom Test provides ananimal-free means of code veri-fication of the control system

13

Verification Tests using a Phantom

14

Verification Tests using a Phantom

15

Work in Progress

- In Vivo trials

. Next Phantom run to demonstrate control of laser inHouston from HPC computers in Austin

- Registration Improvement

. Non-Rigid mesh deformation capabilities

- Utilize and Study the benifits of HP adaptivity within thecontext of this project

16

Questions

dddas.ices.utexas.edu

17