Fast convolution-superposition dose calculation on graphics hardware.

Area of Science:

• Medical Physics
• Computational Science
• Radiation Oncology

Background:

• Accurate and fast dose calculation is critical for radiation therapy treatment planning and optimization.
• Current dose calculation algorithms often face speed limitations due to the complexity of patient-specific data and beam modeling.
• Massively parallel hardware architectures offer potential solutions for accelerating computationally intensive tasks.

Purpose of the Study:

• To implement and evaluate a modern convolution-superposition photon beam dose calculation algorithm on a graphics processing unit (GPU).
• To investigate the impact of different GPU implementation strategies on execution speed and numerical accuracy.
• To assess the potential of GPU acceleration for improving radiation therapy dose calculations.

Main Methods:

• Implementation of a convolution-superposition dose calculation algorithm on a commodity GPU.
• Comparison of GPU implementation performance against traditional central processing unit (CPU) implementations.
• Investigation of simple porting and more complex GPU optimization techniques.

Main Results:

• GPU implementations achieved speed improvements ranging from 10 to 20 times compared to CPU implementations.
• Higher speed-ups were observed with larger kernel and calculation grid sizes.
• Numerical accuracy was preserved in GPU calculations relative to CPU calculations.

Conclusions:

• Graphics processing units (GPUs) can significantly accelerate dose calculation algorithms in radiation therapy.
• GPU acceleration holds promise for numerically intensive processes like treatment optimization, especially for complex techniques like intensity-modulated radiation therapy (IMRT).
• Streaming architectures offer a viable path to enhance the efficiency of radiation therapy planning systems.