Detection of Major Depressive Disorder Using EEG Multifeature Fusion

Major depressive disorder (MDD) is a largely prevailing psychological disorder severely affecting a person’s mental health and may lead to suicidal attempts in extreme conditions. Proper diagnosis of MDD can greatly help in the treatment of MDD, thus reducing its impact on mental well-being. However, the prevalent diagnosis of MDD through questionnaires can introduce subjectivity and human error in the diagnosis. So, in recent years, significant efforts have been made to diagnose MDD using easily measurable neurological signals, specifically the electroencephalogram (EEG) signal. Despite that, the existing approaches for MDD detection using EEG often fail to exploit the full potential of EEG in detecting MDD. Specifically, none of the existing MDD detection methods exploits the multivariate temporal, time-varying spatial, and connectivity information of the EEG data simultaneously. So, this article proposed three novel MDD detection models, viz., multirandom convolutional kernel transform (MultiRocket)-based MTSC model, graph-based model, and connectivity-based model, that effectively exploited the multidimensional temporal, time-varying spatial, and connectivity information, respectively. Eventually, the three suggested models were combined to construct a unified MDD detection model that exploited the distinctive information from each model concurrently to achieve a cutting-edge classification accuracy of ~95%.