An Improved Convolutional Neural Network for 3D Unsupervised Medical Image Registration
Gangcheng Cai, Jiajun Wang, Zheru Chi · 2020
In recent years, convolutional neural networks have been widely used in medical image registration. VTN network has been the most popular architecture in medical registration community. Although the structure has outstanding overall performance in medical image registration, the VTN architecture seems to own drawbacks in focusing on the areas of interest and extracting features efficiently. Therefore, we propose an improved version called CIRVTN based on the traditional VTN model where global Residual paths, the Inception module and the CBAM attention module are introduced. The introduction of global Residual path can not only alleviate the problem of gradient disappearance, but also improve the reusability of medical image features. Upon introducing the Inception and CBAM attention modules, the adaptability of the network to the diversities in shapes and locations of tissue contents in medical images are improved. Extensive experiments on three professional medical liver datasets showed that the network proposed in this paper outperforms the traditional VTN both in Dice score and computing efficiency.