Leveraging Segmentation to Improve Medical Image Registration

Published in IEEE Transactions on Biomedical and Health Informatics, 2023

We investigate the impact of segmentation onmedical image registration. We propose a novel approachcalled the “Weakly Supervised Semantic Attentive Med-ical Image Registration Network” (WSSAMNet++), whichemploys segmentation to guide the registration process.Specifically, our network utilizes the segmentation of theregions of interest to direct the attention of the registrationnetwork to anatomically-relevant features, enabling it tofocus more effectively on the relevant parts of the images.We demonstrate the effectiveness of using segmentationand WSSAMNet++ through extensive experiments on var-ious registration tasks, including single and multi-modalregistration problems, on multiple datasets. Our approachdoes not require any input from the radiologist at test timeand it improves the performance of the registration networkin all the cases tested. In conclusion, our study highlightsthe importance of leveraging semantic information to aidthe registration process and shows the effectiveness of theproposed method in achieving this goal.

Citation

Almahfouz Nasser, Sahar; Meena, Mohit; Sresth, Garweet; Sethi, Amit (2023). Leveraging Segmentation to Improve Medical Image Registration. TechRxiv

Paper Link