Virtual Particle Monte Carlo Dose Calculation for GPU Proton Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Monte Carlo (MC) codes for dose calculation in radiation therapy, both general-purpose and fast MC codes, are too slow for advanced optimizations in spot scanning proton therapy, such as real-time on-the-bed adaptive proton therapy and 4D robust optimization, due to thread divergence and racing conditions inherent to GPU hardware architecture, which hinder efficient use of GPU computing power.
Innovation Solution
A treatment planning system using a GPU with pre-calculated models for soft and hard particle interactions, simulating virtual particles that start at the same location as primary charged particles, minimizing thread divergence and racing conditions, and calculating dose distribution based on these virtual particles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If fast Monte Carlo codes use GPU acceleration with one particle per thread technique, then computing speed is improved, but thread divergence occurs causing loss of computing performance
Solution Approach 1:
The patent segments the particle simulation into multiple independent virtual particles, each handled by a separate GPU thread. This segmentation allows parallel processing while maintaining thread independence, reducing thread divergence issues inherent in shared-memory GPU architectures when simulating complex particle interactions.
Solution Approach 2:
The patent introduces an intermediary data structure that mediates between the particle simulation logic and GPU shared memory. This intermediary layer manages memory access patterns to minimize thread divergence and racing conditions, allowing efficient GPU parallelization without sacrificing simulation accuracy.
2Productivity
If fast Monte Carlo codes use large-scale shared memory parallel computation, then processing capacity is improved, but racing conditions occur causing memory conflicts
Solution Approach 1:
The patent performs preliminary actions by pre-defining particle trajectories and interaction points before GPU execution. This pre-computation reduces the need for dynamic shared memory access during simulation, minimizing racing conditions while maintaining parallel processing capacity.
Solution Approach 2:
The patent uses copying by creating independent copies of particle data structures for each GPU thread. This eliminates the need for threads to share and synchronize access to the same memory locations, preventing racing conditions while enabling efficient parallel computation.
3Measurement precision
If general purpose Monte Carlo codes are used for accurate dose calculation, then calculation accuracy is improved, but calculation time increases to several hours or days
Solution Approach 1:
The patent changes key simulation parameters including particle energy thresholds, interaction cross-section approximations, and step-size controls. These parameter optimizations maintain clinical dose calculation accuracy while reducing computation time from hours to minutes or seconds through GPU acceleration.
Solution Approach 2:
The patent applies partial action by simulating only the most significant particle interactions that contribute to dose deposition. Less significant interactions are approximated or omitted, maintaining sufficient accuracy for clinical decision-making while dramatically reducing calculation time.
Data Source
AI summary
Radiation treatment planning for charged particle therapy systems, such as proton therapy systems, includes calculating dose distribution(s) using a virtual particle Monte Carlo (“VPMC”) dose calculation engine that utilizes virtual particles. Virtual particles inherit the physical properties from realistic particles, but are conceptually designed for parallel computing in graphics processing units (“GPUs”) by avoiding the simulation of secondary particles. Simulation of virtual particles instead of realistic particles takes full advantage of the GPU hardware architecture (e.g., by avoiding thread divergence and racing conditions).