Anatomically constrained and attention-guided deep feature fusion for joint segmentation and deformable medical image registration

Hee Guan Khor, Guochen Ning, Yihua Sun, Xu Lu, Xinran Zhang, Hongen Liao

Abstract: Highlights•Utilizing segmentation does foster prior knowledge in the registration process.•Registration performance improves with the proposed multi-task learning strategy.•Predictions are refined with the proposed anatomical constraint strategy.•Feature transferring by Cross-Task Attention maximizes the flow of information.•Attention mechanism can reweight attention to deformed regions.