Abstract: At present, brain network classification has become a research focus. Researchers use different methods to extract and select features from label data, in order to realize automatic classification of data. However, it is timeconsuming to extract and select the optimal features from a large number of label data. In order to solve the above problems, a similarity calculation method of brain network is proposed and a clustering framework based on brain network is constructed. The cosine similarity and the subnetwork kernel are used to measure the attribute similarity and structural similarity of the brain network, and the structural similarity and attribute similarity are integrated into a similarity matrix. Finally, the spectral clustering is used to realize the clustering of the brain network. A cluster test is carried out on 50 patients with schizophrenia and 49 normal controls in open fMRI database. The results show that the Rand index is 0.91, the accuracy rate is 0.86, the recall rate is 0.98, and the F1 value is 0.92. Therefore, the proposed method can accurately calculate the similarity of brain network and show high clustering performance.

Key words: brain network, fMRI data, attribute similarity, structure similarity, clustering, schizophrenia