GPU Monte Carlo Dose Calculation Parallelization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Monte Carlo simulation-based dose calculation methods are inefficient for real-time clinical applications due to long calculation periods, despite improvements in computing speed, indicating a need for enhanced computing efficiency in radiotherapy dose calculations.
Innovation Solution
A Monte Carlo simulation-based parallel dose calculation method utilizing CUDA threads on a GPU to simulate particle transport processes in parallel using pre-generated trajectory data, where particle information and trajectory data are shared among threads, improving calculation efficiency by reducing memory access delays and enhancing parallel execution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If general-purpose Monte Carlo simulation tools use accurate transport algorithms and reaction section data to simulate particle transport processes, then calculation accuracy is improved, but calculation period becomes excessively long
Solution Approach 1:
The patent segments the calculation domain into multiple independent regions and assigns different numbers of threads to simulate particle transport in each region simultaneously. This parallelization strategy maintains accurate transport algorithms while reducing total calculation time by dividing the computational workload across multiple processing units.
Solution Approach 2:
The patent transitions from sequential single-thread simulation to multi-dimensional parallel processing by utilizing GPU architecture with thousands of threads working simultaneously. This dimensional expansion from 1D sequential execution to 3D parallel execution space dramatically reduces calculation period while preserving accuracy through consistent application of transport algorithms across all threads.
2Productivity
If fast Monte Carlo computing models are proposed to improve computing efficiency, then computing speed is improved, but there is still insufficient speed for near-real-time computing requirements
Solution Approach 1:
The patent performs preliminary actions by pre-calculating and storing transport parameters and reaction section data in global memory before simulation begins. This pre-positioning of data eliminates runtime memory access delays and allows computing threads to execute transport algorithms at maximum speed, pushing computing performance toward near-real-time requirements.
Solution Approach 2:
The patent introduces shared memory as an intermediary layer between global memory and computing threads. This intermediary buffer allows frequent access to particle state information and trajectory data without costly global memory accesses, significantly boosting computing speed while maintaining data accuracy for dose calculation.
3Productivity
If multiple threads access trajectory data from global memory simultaneously, then parallel processing capability is improved, but memory access delays increase
Solution Approach 1:
The patent segments trajectory data access by allocating specific data portions to specific thread blocks, reducing contention for global memory bandwidth. Each thread block accesses distinct regions of trajectory data, allowing simultaneous memory accesses without conflicts and maintaining high parallel processing capability while minimizing access delays.
Solution Approach 2:
The patent introduces shared memory as an intermediary cache layer between global memory and computing threads. Frequently accessed trajectory data is loaded into shared memory once and then rapidly accessed by multiple threads without repeated global memory accesses, dramatically reducing memory access delays while preserving parallel processing throughput.
Data Source
AI summary
Provided is a Monte Carlo simulation-based parallel dose calculation method, a computer device, and a storage medium. The method includes: acquiring sampled particles by sampling according to a radiation source model of a device, wherein information of the sampled particles includes positions, movement directions, energies, and weights of the sampled particles; stochastically generating a plurality of new positions in a region of interest along the movement directions of the sampled particles; and simulating transport processes of particles in parallel at the plurality of new positions using same particle information and same trajectory data, wherein the particle information used at the plurality of new positions is the information of the sampled particles.