Machine Learning Approaches for MDD Detection and Emotion Decoding Using EEG Signals

• Published in: Frontiers in human neuroscience Vol. 14; p. 284
• Main Authors: Duan, Lijuan, Duan, Huifeng, Qiao, Yuanhua, Sha, Sha, Qi, Shunai, Zhang, Xiaolong, Huang, Juan, Huang, Xiaohan, Wang, Changming
• Format: Journal Article
• Language: English
• Published: Lausanne Frontiers Research Foundation 23.09.2020; Frontiers Media S.A
• Subjects: Accuracy; Artificial intelligence; Asymmetry; Brain; Cerebral hemispheres; Classification; cross correlation; EEG; Electroconvulsive therapy; Electrodes; Electroencephalography; feature; Human Neuroscience; interhemispheric asymmetry; Learning algorithms; Machine learning; major depressive disorder (MDD); Medical research; Mental depression; Mental disorders; Neural networks; Time series; Wavelet transforms; Beijing China
• ISSN: 1662-5161; 1662-5161
• DOI: 10.3389/fnhum.2020.00284

Emotional decoding and automatic identification of major depressive disorder (MDD) is helpful to doctors in diagnosis of the disease on time, and electroencephalogram (EEG) is sensitive to the changes of functional state of human brain, showing its potential to help to diagnose MDD. In this paper, an approach for identifying MDD by fusing interhemispheric asymmetry and cross correlation with EEG signals is proposed and tested on 32 subjects (16 MDD and 16 healthy controls (HC)). First, the structure feature and connectivity feature of theta, alpha and beta band are extracted on the preprocessed and segmented EEG. Second, the structure feature matrix of theta, alpha and beta are added to and subtracted the connectivity feature matrix respectively to obtain the mixed features. Finally, the structure feature, connectivity feature and the mixed features are fed to six classifiers respectively to select the suitable features for the classification, and it is found that we have the best classification results using the mixed features. The results are also compared with those from some of the state-of-the-art methods, and we achieve accuracy of 94.13%, sensitivity of 95.74%, specificity of 93.52% and f1_score of 95.62% on the data from the Beijing Anding Hospital, Capital Medical University. The study could be generalized to develop a Brain–computer interfacing(BCI) system that may help for clinical purposes.