Abnormal Brain Function Network Analysis Based on EEG and Machine Learning

Neurosis is one of the chronic diseases, which is difficult to diagnose based on whether the brain tissue is damaged or not. Effectively distinguishing different types of neuroses is a challenging task. In this paper, neurosis is divided into somatic symptom dimension and emotional symptom dimension, and a feature extraction algorithm for neurosis is proposed to extract effective features and accurately classify neurosis patients in different dimension groups. Electroencephalogram (EEG) is used to extract the feature data of brain function network to locate the abnormal connections in the brain. The EEG data were obtained from 96 screened neurosis patients. Each data was carefully collected and preprocessed. The brain function network is constructed by using threshold selection method of network sparsity. Four topological properties, i.e. clustering coefficient, average shortest path, local efficiency and global efficiency are extracted to analyze the differences between different dimension groups. According to the characteristics of significant differences, the abnormal connections of brain functional networks were explored. For the somatic dimension group, the connections between various brain regions and the parietal lobe, temporal and parietal lobe and occipital lobe were abnormally enhanced, while for the emotional dimension group, the connections between frontal lobe, prefrontal lobe and central parietal lobe were abnormal enhanced. Different machine learning algorithms are used to classify neurosis patients based on the brain network features extracted from the three indicators, i.e. Phase Lag Index (PLI), Weighted Phase Lag Index (WPLI) and Coherence (COH), and the classification accuracy can reach more than 70%. In particular it was more than 90% by using the feature combination of PLI and WPLI, which objectively proves that the extracted differential features can effectively distinguish between somatic dimension neurosis and emotional dimension neurosis.