Normalization for connectivity of brain regions in schizophrenia classification

It is a popular method to use resting-state functional magnetic resonance imaging (rs-fMRI) to obtain brain connectivity, which can study whether brain function is abnormal, and can be constructed by techniques such as correlation and entropy. Usually, the techniques need to normalize the signal first, and how to normalize has an important impact on the obtained results. Unfortunately, there is little discussion about it in existing methods. For this problem, this paper focuses on the normalization, analyzes its impact on the correlation and entropy of the brain region connectivity, and proposes a normalization method for the region of interest (ROI) signals of a brain, which can be used to improve the classification accuracy of schizophrenia (SCZ). In this experiment, a public dataset is used, and the experimental results show that the classification accuracy of the method proposed in this paper is about 10% higher than that of the traditional normalization method. After the two-sample T-test, the significant regions obtained by the proposed method in the parietal lobe (PAR), cortex (SUB), and temporal lobe (TEM) are different from those of the traditional method, which indicates that the proposed method have some information that the traditional one does not have.