Parallel generation of digitally reconstructed radiographs on heterogeneous multi-GPU workstations

Area of Science:

• Medical Physics
• Computational Imaging
• Radiotherapy Technology

Background:

• Three-dimensional radiotherapy relies on computational workflows to simulate X-ray images (DRRs) from CT data for patient alignment.
• Current DRR rendering pipelines struggle with the large datasets from modern CT scanners.

Purpose of the Study:

• To develop a high-performance rendering pipeline for generating high-quality DRRs from large-scale CT volumes.
• To address the limitations of existing DRR pipelines in handling extensive imaging data.

Main Methods:

• Implemented a high-performance X-ray rendering pipeline utilizing OpenCL for heterogeneous computing platforms.
• Incorporated load-balancing optimization to distribute rendering tasks evenly across computing resources.
• Leveraged multi-GPU workstations for accelerated DRR generation.

Main Results:

• The pipeline successfully generates high-quality DRRs from large CT volumes at interactive frame rates.
• Demonstrated efficient rendering, achieving 85 frames per second for a 512^2 DRR from a 1024 × 2048 × 2048 CT volume.
• Validated the pipeline's performance on cost-effective multi-GPU workstations.

Conclusions:

• The developed pipeline offers a significant improvement in DRR generation speed and quality for large CT datasets.
• This advancement supports enhanced image-guided radiotherapy through faster and more reliable patient alignment verification.
• The use of OpenCL and load balancing enables efficient utilization of heterogeneous computing resources.