TST-network: A two-stage mutually reinforcing deep learning network for brain MR registration

Abstract

Abstract: Image registration is a critical component of various medical image analysis applications. In recent years, medical image registration models based on deep learning have developed rapidly. The realization of the deep neural network provides opportunities for some medical applications, such as image registration with high accuracy in a shorter time, which plays a crucial role in anti-tumor during surgery. To improve registration accuracy, we propose a new two-stage image registration framework based on deep learning, which can make two identical VoxelMorph convolution neural networks cascade by fighting against the loss function so that the registration effect can promote each other. We demonstrated our method in the brain magnetic resonance (MR) image registration task and showed that the new model could achieve better registration results.