Classification Model of Suicide Attempt based on Psychological Pain and Rest-state EEG Signals in Patients with Major Depressive Disorder
Objective:To examine the efficiency of suicide attempt classification model and the ranks of pain avoidance and the rest-state EEG signals in the optimal feature set by using machine learning technique.Methods:Seventy-three pa-tients with major depressive disorder and 36 healthy controls were selected by convenience sampling for clinical evaluation and rest-state EEG data collection.Then through fusion of demographic feature,clinical scores and rest-state EEG feature,and based on the support vector machine algorithm for machine learning,the suicide classification model was constructed.Results:①The accuracy of the multimodal classification model for suicide attempts was 85.01％with an AUC of 0.93;the accuracy of the unimodal classification model for suicide attempts constructed using only rest-state EEG indicators was 72.10％with an AUC of 0.59;②The important feature set of multimodal classification models of suicide attempts were:dis-tress avoidance,age,low-frequency gamma of the oz channel and depression;③The low-frequency and high-frequency gamma were common important features in both the pain avoidance classification model and the multimodal classification model of suicide attempts.Conclusion:The efficacy of suicide attempts multimodal classification model is excellent.Pain avoidance is the best behavioral feature distinguishing depressed patients with suicide attempts and it's better than depres-sion.As an EEG feature closely related to pain avoidance,the abnormal of gamma may be the neural basis of suicide at-tempts.
万方数据
维普
