Ebook Hardware Acceleration of a Monte Carlo Simulation for Photodynamic Therapy Treatment Planning
Photodynamic therapy (PDT) is an emerging, minimally invasive treatment modality in oncology and other fields. For treating oncologic conditions, the key steps in PDT include the uptake of a light-sensitive drug in the patient’s tumour and the local activation of the drug by delivering a sufficient light dose selectively to the region. To effectively target the therapy at the tumour, while sparing the healthy tissue nearby, accurate light dosimetry is critical during treatment planning.
Among other techniques for computing light dose, the Monte Carlo (MC) method is considered the gold standard approach in terms of accuracy and flexibility in modelling complex 3-D geometries. However, the use of MC-based models for solving iterative, inverse optimization problems such as PDT treatment planning is currently hindered by its long computation time. Accelerating MC simulations would enable their use for solving such computationally intensive inverse problems. Specifically, this thesis explores two different hardware-based approaches to accelerate an MC simulation for computing light dose in multi-layered biological tissue.
The following sections review the progress in PDT and the clinical dosimetric concepts unique to PDT. In particular, the notion of light dosimetry is introduced to explain how the MC method can be used for PDT treatment planning.
Contents
List of Tables
List of Figures
List of Abbreviations and Symbols
1 Introduction
1.1 Progress in Photodynamic Therapy
1.2 Clinical Dosimetry for PDT Treatment Planning
1.3 Light Dosimetry Models
1.4 MC-based Light Dosimetry
1.5 Organization of Dissertation
2 The MCML Light Dose Computation Method
2.1 The Monte Carlo Method
2.2 The MCML Algorithm
- 2.2.1 Photon Initialization
2.2.2 Position Update
2.2.3 Direction Update
2.2.4 Fluence Update
2.2.5 Photon Termination
2.3 Summary
3 FPGA-based Acceleration of the MCML Code
3.1 Field-Programmable Gate Arrays
3.2 Related Work
3.3 Hardware Design Method
- 3.3.1 Overview
3.3.2 Hardware Acceleration Techniques
3.4 FPGA-based Hardware Implementation
- 3.4.1 Modifications to the MCML code
3.4.2 System Overview
3.4.3 Overview of Hardware Design
3.4.4 Pipeline Stages in the Fluence Update Core
3.4.5 Design Challenges for the Direction Update Engine
3.4.6 Importance of Managing Resource Usage
3.4.7 Trade-offs
3.5 Validation
- 3.5.1 FPGA System-Level Validation Procedures
3.5.2 Results
3.6 Performance
- 3.6.1 Multi-FPGA Platform: TM-4
3.6.2 Modern FPGA Platform: DE3 Board
3.7 Resource Utilization
3.8 Power
3.9 Summary
4 GPU-based Acceleration of the MCML Code
4.1 Graphics Processing Units
4.2 Related Work
4.3 CUDA-based GPU Programming
- 4.3.1 GPU Hardware Architecture
4.3.2 Programming with CUDA
4.3.3 CUDA-specific Acceleration Techniques
4.4 GPU-accelerated MCML Code
- 4.4.1 Parallelization Scheme
4.4.2 Key Performance Bottleneck
4.4.3 Solution to Performance Issue
4.4.4 Other Key Optimizations
4.4.5 Scaling to Multiple GPUs
4.5 Performance
- 4.5.1 GPU and CPU Platforms
4.5.2 Speedup
4.5.3 Effect of Optimizations
4.5.4 Effect of Grid Geometry
4.6 Validation
- 4.6.1 Test Cases
4.6.2 Error Distribution
4.6.3 Light Dose Contours
4.7 Summary
5 Conclusions
5.1 Summary of Contributions .
5.2 Future Work
- 5.2.1 Extension to 3-D and Support for Multiple Sources
5.2.2 Sources of Uncertainties
5.2.3 PDT Treatment Planning using FPGA or GPU Clusters
A Source Code for the Hardware
B Source Code for the CUDA program
Bibliography
Posted in :
