Diagnostic Support for Major Depressive Disorder: A Multimodal Analysis of EEG and Audio Biomarkers
Depression is a prevalent psychological disorder that affects millions worldwide and poses a diagnostic challenge. Existing criteria, such as the international questionnaires Diagnostic and Statistical Manual of Mental Disorders (DSM-5) and Hamilton Depression Rating Scale (HAMD), while widely used, may not consistently and accurately diagnose depression. In response, recent research efforts have tried to explore promising biomarkers for enhancing depression diagnosis by leveraging electroencephalogram (EEG) and audio data. Using the multimodal MODMA dataset, the proposed framework encompasses EEG and audio data preprocessing, feature extraction, feature selection, and the implementation of a Support Vector Machine (SVM) algorithm for classification. Considering the use of EEG data and mRMR feature selection, an accuracy of 76% was obtained considering all available patients, while a performance of 49,82% was obtained using the common patients for which EEG and audio data are available. As for the audio processing, an accuracy of 80,9% was achieved using all data available and 78,03% when common patients were considered. A multimodal analysis integrating EEG and audio features showed improved performance, achieving an accuracy of 80.32%, highlighting the value of combining complementary biomarkers. These results reinforce the potential of multimodal approaches to enhance the accuracy and reliability of depression diagnosis.