SC19 Proceedings

The International Conference for High Performance Computing, Networking, Storage, and Analysis

Poster 61: Fast 3D Diffeomorphic Image Registration on GPUs

Authors: Malte Brunn (University of Stuttgart), Naveen Himthani (University of Texas), George Biros (University of Texas), Miriam Mehl (University of Stuttgart), Andreas Mang (University of Houston)

Abstract: 3D image registration is one of the most fundamental and computationally expensive operations in medical image analysis. Here, we present a mixed-precision, Gauss-Newton-Krylov solver for diffeomorphic registration. Our work extends the publicly available CLAIRE library to GPU architectures. Despite the importance of image registration, only a few implementations of large deformation diffeomorphic registration packages support GPUs. Our contributions are new algorithms and dedicated computational kernels to significantly reduce the runtime of the main computational kernels in CLAIRE: derivatives and interpolation. We deploy (i) highly-optimized, mixed-precision GPU-kernels for the evaluation of scattered-data interpolation, (ii) replace FFT-based first-order derivatives with optimized 8th-order finite differences, and (iii) compare with state-of-the-art CPU and GPU implementations. As a highlight, we demonstrate that we can register 256^3 clinical images in less than 6 seconds on a single NVIDIA Tesla V100. This amounts to over 20x speed-up over CLAIRE and over 30x speed-up over existing GPU implementations.

Best Poster Finalist (BP): no

Poster: PDF
Poster summary: PDF

Back to Poster Archive Listing