Symmetric-Constrained Irregular Structure Inpainting for Brain MRI Registration with Tumor Pathology

Deformable registration of magnetic resonance images between patients with brain tumors and healthy subjects has been an important tool to specify tumor geometry through location alignment facilitate pathological analysis. Since region does not match any ordinary tissue, it difficult deformably register a patient’s normal one. Many patient are associated irregularly distributed lesions, resulting in further distortion tissue structures complicating registration’s similarity measure. In this work, we follow multi-step context-aware image inpainting framework generate synthetic intensities the region. The coarse image-to-image translation is applied make rough inference missing parts. Then, feature-level patch-match refinement module refine details by modeling semantic relevance patch-wise features. A symmetry constraint reflecting large degree anatomical proposed achieve better structure understanding. inpainted brains, deformation field eventually used deform original data for final alignment. method was Multimodal Brain Tumor Segmentation (BraTS) 2018 challenge database compared against three existing methods. yielded results increased peak signal-to-noise ratio, structural index, inception score, reduced L1 error, leading successful patient-to-normal registration.