CLAIRE-ROP: Rapid Partitioning-based Deformable Image Registration on Multi-GPU Accelerator

Deformable image registration (DIR) is an important tool for clinical applications, especially in medical imaging and radiotherapy, as it ensures accurate image alignment and analysis. Given the high computational demands, acceleration of DIR is essential to ensure efficient and timely medical procedures. The use of a multi-GPU implementation holds the potential for significant DIR acceleration. However, the challenges posed by high communication times and increasing complexity have made using multi-GPU configurations impractical in the clinical setting. In this paper, we propose a novel solution, called CLAIRE-ROP (Rapid Overlapped Partitioning-based) to address these challenges to develop an efficient multi-GPU implementation. Our approach incorporates an efficient partitioning scheme that divides lung images into multiple partitions and allows individual registration for each partition without compromising accuracy. Our method successfully registers images from the largest openly available lung dataset in less than 0.5 seconds with a Dice score of 0.991.