Fusing EEG Features Extracted by Microstate Analysis and Empirical Mode Decomposition for Diagnosis of Schizophrenia

Accurate early diagnosis and precise assessment of disease severity are imperative for the treatment and rehabilitation of schizophrenia patients. To achieve this, we propose a computer-aided diagnostic method for schizophrenia that utilizes fusion features derived from microstate analysis and empirical mode decomposition (EMD) based on Electroencephalography (EEG) signals. At the same time, the obtained fusion features from microstate analysis and EMD are input into the Least Absolute Shrinkage and Selection Operator (LASSO) feature selection algorithm to reduce the dimensionality of feature vectors. Finally, the reduced feature vector is fed to a Logistic Regression classifier to classify SCH and healthy EEG signals. In addition, the ability of the integrated features to distinguish the severity of schizophrenia symptoms was evaluated, and the Shapley Additive Explanations (SHAP) algorithm was used to analyze the importance of the classification features that differentiate schizophrenia symptoms. Experimental results from both public and private datasets demonstrate the efficacy of EMD features in identifying healthy controls, while microstate features excel in classifying the severity of symptoms among schizophrenia patients. The classification evaluation metrics of the fused features significantly outperform those obtained using EMD or microstate analysis features independently. The fusion feature method proposed in this study achieved accuracies of 100% and 90.7% for the classification of schizophrenia in public datasets and private datasets, respectively, and an accuracy of 93.6% for the classification of schizophrenia symptoms in private datasets.