Large Deformation Features Guided Network for Medical Image Registration

Abstract

Aiming at the problem of the loss of detail information in medical image registration, which leads to poor registration results in large deformation regions, a large deformation features guided network(LDGNet) for medical image registration is proposed. LDGNet mainly includes two contributions: first, a large deformation feature enhancement module is designed at the encoding and decoding connection to enable the network to enhance the extraction of large deformation features. Secondly, a large deformation feature guidance module is designed at the skip connection, which can help fully fuse the large deformation features from the encoded feature map, and effectively improve the registration accuracy of the network in large deformation regions. Registration experiments on the brain dataset IXI show that LDGNet achieves higher registration accuracy compared with current popular medical image registration methods.