tsta

فارسی

Home / دومین کنفرانس ملی عصر انفجار تکنولوژی؛ هوش مصنوعی، تحولی در صنعت، تجارت و زنجیره تامین و دومین کنفرانس ملی علم داده در کاربردهای مهندسی

Depression detection based on EEG signal analysis utilizing Inter-hemispheric Asymmetry and Correlation Dimension assessment

Authors :

Amirreza Ahmadi¹ Saeid Yarmohammdi² Ali Zeraatkar³ Reza Rostami⁴

1- دانشگاه آزاد اسلامی واحد علوم و تحقیقات تهران 2- دانشگاه آزاد اسلامی واحد تهران مرکزی 3- University of Victoria 4- دانشگاه تهران

Keywords :

EEG،Major Depressive Disorder،Machine Learning،Correlation Dimension،Inter-hemispheric Asymmetry

Abstract :

Depressive disorders represent the most significant health risk among mental illnesses. Diagnosing the disability in the first stages can improve treatment efficiency and save a patient’s life due to its curable characteristic. Questionnaire-based diagnostic criteria have been required for traditional depression diagnoses. This study suggests objective criteria and processed EEG signals of 17 MDD patients and 20 normal subjects to detect depression. The power of absolute and relative frequency bands and the inter-hemispheric asymmetry were extracted as the linear features, and the correlation dimension was considered as the non-linear feature. Five machine-learning models were used to classify the data. 91.7% of accuracy score was derived when the selected features with all the mentioned machine learning classifiers were used. In addition, the ROC-AUC score and F1 score were utilized for higher trustable results. The LR classifier demonstrated strong performance, achieving a peak F1 score of 93.3% (when using 'Absolute + Relative' features) and a peak ROC-AUC score of 97.1% (when using 'Relative' features). The results of the T-test have shown the Alpha inter-hemispheric asymmetry as not a robust biomarker. Besides, the correlation dimension was probed as an auxiliary biomarker in channels F8 and C4 to be applied with the other characteristics; the value of the T-test of other bands was insignificant. This study reveals the importance of feature selection and states that using the selected features and our suggested machine-learning models could provide a valuable tool for detecting depression.