Unsupervised Deformable Image Registration with Absent Correspondences in Pre-operative and Post-recurrence Brain Tumor MRI Scans

Registration of pre-operative and post-recurrence brain images is often needed to evaluate the effectiveness gliomas treatment. While recent deep learning-based deformable registration methods have achieved remarkable success with healthy images, most them would be unable accurately align pathologies due absent correspondences in reference image. In this paper, we propose a method that jointly estimates regions correspondence bidirectional deformation fields. A forward-backward consistency constraint used aid localization resection recurrence region from voxels absence two images. Results on 3D clinical data BraTS-Reg challenge demonstrate our can improve image alignment compared traditional approaches or without cost function masking strategy. The source code available at https://github.com/cwmok/DIRAC .