Accelerating DRR generation using Fourier slice theorem on the GPU.

Area of Science:

• Medical Imaging
• Radiotherapy
• Computational Science

Background:

• Digitally Reconstructed Radiographs (DRRs) are crucial for monitoring patient positioning in image-guided therapies.
• Conventional DRR generation methods, like ray casting, suffer from high computational complexity (O(N(3))).
• The Fourier slice theorem offers a more efficient k-space approach for DRR generation.

Purpose of the Study:

• To develop a high-performance, scalable, and optimized DRR generation pipeline.
• To leverage Graphics Processing Unit (GPU) acceleration for faster DRR computation.
• To demonstrate the pipeline's efficiency on contemporary GPUs.

Main Methods:

• Implementation of a DRR generation pipeline utilizing GPU acceleration.
• Exploration of k-space methods based on the Fourier slice theorem.
• Performance evaluation focusing on strong scaling using two distinct GPUs.

Main Results:

• The developed pipeline achieves high performance and scalability on GPUs.
• DRRs for 512^3 volumes can be generated in under one millisecond.
• The strong scaling performance of the pipeline is validated on modern GPU hardware.

Conclusions:

• The GPU-accelerated pipeline offers a significant speedup for DRR generation.
• This advancement can enhance the efficiency of radiotherapy and medical imaging procedures.
• The optimized pipeline demonstrates the potential of GPUs for complex image reconstruction tasks.