XMorpher: Full Transformer for Deformable Medical Image Registration via Cross Attention

An effective backbone network is important to deep learning-based Deformable Medical Image Registration (DMIR), because it extracts and matches the features between two images discover mutual correspondence for fine registration. However, existing networks focus on single image situation are limited in registration task which performed paired images. Therefore, we advance a novel network, XMorpher, corresponding feature representation DMIR. 1) It proposes full transformer architecture including dual parallel extraction exchange information through cross attention, thus discovering multi-level semantic while extracting respective gradually final 2) advances Cross Attention Transformer (CAT) blocks establish attention mechanism able find automatically prompts fuse efficiently network. 3) constrains computation base windows searching with different sizes, focuses local transformation of deformable enhances computing efficiency at same time. Without any bells whistles, our XMorpher gives Voxelmorph 2.8% improvement DSC, demonstrating its from We believe that has great application potential more medical Our open https://github.com/Solemoon/XMorpher