Exploiting Heterogeneous Architectures for Rigid Image Registration

Current medical practice increasingly relies on multiple imaging techniques to deliver faster and more accurate diagnosis as well as live medical imaging integration to guide critical surgery procedures. An essential stage of these processes is Image Registration, which is responsible for geometrically overlaying floating images on a reference. However, most of the available tools are proprietary, cumbersome to use/customize, or do not deliver real-time performance. Therefore, we propose an easy-to-use open-source framework for rigid multi-modal image registration that exploits heterogeneous architectures and algorithmic optimizations. Compared to literature software tools, our solution shows outstanding speedups from a 3.8x to a 16x and an improved Intersection over Union of 0.97.