Research on fMRI Image Registration Method Based on Deep Learning

Functional Magnetic Resonance Imaging (fMRI) is a non-invasive, non-traumatic brain imaging technique with high tissue resolution and the ability to image in any plane. It represents a significant advancement in modern medical technology. In medical image processing, it is often necessary to compare different medical images to draw conclusion. However, due to individual differences and the impact of disease states, there are significant variations between fMRI data from different patients. To ensure data comparability, functional regions of fMRI images from different individuals need to be aligned before data analysis, which is call image registration. A deep learning network method for fMRI registration is proposed in this paper based on the U-net framework. In the network architecture, Transformer is integrated to enhance the model’s feature extraction capabilities. The method also utilizes block-based tissue-specific functional correlation.