Rigid Image Registration for Head MRI Based on 3DCNN Incorporated Global Information

Since Alzheimer's disease is incurable nowadays, early detection and treatment of Alzheimer is extremely important. One of the diagnostic methods is to evaluate the brain atrophy with the progression of Alzheimer's disease according to MRI image. Computer-Aided Diagnosis (CAD) system with medical image processing and artificial intelligence provides an automatic way to analyze the degree of brain atrophy to physicians. As one of the critical steps, three-dimensional (3D) image registration method is utilized to analyze temporal variation of the brain volume. In this paper, a 3D convolutional neural network (3DCNN) based method with the blocks of 3D Grid Loss, Stem Block, Vision Transformer, and s decoupled fully connected (DFC) Attention is proposed. We applied our method to the ADNI dataset, which includes 328 head MRI cases. The results showed that the Normalized Cross-Correlation (NCC) was 0.952, with more than 0.13 improvement comparing the up-to-date methods.