Classification of Psychiatric Disorders Using QEEG Data Through Convolution Neural Networks and Gated Recurrent Units

Psychiatric disorders present a significant challenge for diagnosis and treatment. Utilizing electroencephalogram (EEG) data, deep learning (DL) classifiers offer a promising avenue for discerning between healthy individuals and those with psychiatric disorders. While previous studies have explored DL applications in EEG analysis, few have focused on comprehensive classification across a range of psychiatric disorders using quantitative EEG (QEEG) data. This research aims to engineer a DL classifier capable of accurately distinguishing EEG signatures of individuals with major psychiatric disorders from healthy controls, utilizing resting-state QEEG data. A dataset comprising 850 patients with various psychiatric disorders and 95 healthy controls was analysed. QEEG parameters, including power spectrum density (PSD) and functional connectivity (FC) at different frequency bands, were used to establish classification models. Convolutional Neural Networks (CNNs) and Gated Recurrent Units (GRUs) were employed, and validation accuracies were compared. The study identified optimal feature combinations and achieved high classification accuracies: anxiety disorder (93% accuracy, high-beta FC), addictive disorder (93% accuracy, gamma FC), trauma and stress-related disorder (93% accuracy, gamma PSD), mood disorder (96% accuracy, delta FC), obsessive-compulsive disorder (90% accuracy, beta FC), and schizophrenia (93% accuracy, high-beta FC). These findings suggest the potential of DL in EEG analysis for predicting psychiatric disorders and providing an objective index for diagnosis. Such advancements may significantly contribute to improved clinical assessment and treatment strategies.