A Transformer-Based Network for Deformable Medical Image Registration

Deformable medical image registration plays an important role in clinical diagnosis and treatment. Recently, the deep learning (DL) based methods have been widely investigated showed excellent performance computational speed. However, these cannot provide enough accuracy because of insufficient ability representing both global local features moving fixed images. To address this issue, paper has proposed transformer method. This method uses distinctive to extract for generating deformation fields, on which registered is produced unsupervised way. Our can improve effectively by means self-attention mechanism bi-level information flow. Experimental results such brain MR datasets as LPBA40 OASIS-1 demonstrate that compared with several traditional DL methods, our provides higher terms dice values.