TransMorph: Transformer for unsupervised medical image registration

Junyu Chen, Eric C. Frey, Yufan He, William Paul Segars, Ye Li, Yong Du

Abstract: Highlights•A novel hybrid Transformer-ConvNet model was designed for 3D medical image registration.•The self-attention mechanism and the large effective receptive field of the Transformer improve registration performance.•The diffeomorphic variations of the proposed model were presented to ensure topology-preserving deformations.•The Bayesian variant of the proposed model was introduced, which produces a well-calibrated uncertainty estimate.•The source code, the pre-trained models, and the preprocessed dataset are publicly available at https://bit.ly/37eJS6N.