Multi-GPU configuration of 4D intensity modulated radiation therapy inverse planning using global optimization

Area of Science:

• Medical Physics
• Computational Science
• Radiotherapy

Background:

• Radiotherapy treatment planning requires complex, computationally intensive optimization.
• Higher-order optimization techniques can improve treatment plan quality but demand significant computational resources.
• Existing platforms may face limitations in handling large datasets and complex calculations for advanced techniques like 4D-IMRT.

Purpose of the Study:

• To design, implement, and characterize a multi-graphic processing unit (GPU) computational platform for higher-order optimization in radiotherapy treatment planning.
• To develop and evaluate a parallelized particle swarm optimization (PSO) technique for four-dimensional (4D) intensity modulated radiation therapy (IMRT).
• To address challenges in data management and non-uniform memory access (NUMA) for large-scale radiotherapy optimization.

Main Methods:

• Configured a research prototype GPU-enabled workstation with dual Xeon processors, 256 GB RAM, and four NVIDIA Tesla K80 GPUs.
• Developed a parallelized particle swarm optimization (PSO) engine coupled to the Eclipse treatment planning system via a scripting interface.
• Managed large datasets (approx. 300 GB) by optimizing data transfer and parallelization strategies to mitigate NUMA effects.

Main Results:

• The GPU platform demonstrated utility for large radiotherapy optimization problems, specifically for 4D-IMRT.
• The optimized 4D-IMRT plan achieved an average reduction of [Formula: see text] in maximum dose to organs at risk compared to the clinical plan.
• Computation speed did not monotonically increase with the number of GPUs; optimal performance was achieved with 5 GPUs in this configuration, completing optimization in 35 minutes.

Conclusions:

• A multi-GPU computational platform can significantly enhance radiotherapy treatment planning optimization.
• The developed parallelized PSO technique effectively improves plan quality by enhancing organ-at-risk sparing.
• Hardware specifications and data handling strategies are critical for achieving optimal performance in GPU-accelerated radiotherapy planning.