Electroencephalography-Based Machine and Deep Learning Approaches for the Diagnosis of Dissociative Disorders: A Comprehensive Review

Dissociative disorders (DDs), including dissociative identity disorder and depersonalization disorder, are complex and often misdiagnosed psychiatric conditions due to their overlapping symptoms with other mental illnesses. Electroencephalography (EEG), a low-cost and noninvasive neuroimaging tool, is a valuable means of examining the neurophysiological signatures associated with DDs. In this review, we aim to systematically evaluate how machine learning (ML) and deep learning (DL) methods are applied to EEG data for the diagnosis and monitoring of DDs, highlighting their effectiveness, limitations, and future research directions. We reviewed and synthesized studies involving EEG-based ML and DL models applied to DD-related data. The analysis focused on EEG biomarkers, model architecture (e.g., support vector machines, convolutional neural networks [CNNs], recurrent neural networks [RNNs]), feature types (raw vs. handcrafted), performance metrics, and reported challenges. Findings indicate that DL models, especially CNN and RNN, outperform traditional ML models by learning complex spatiotemporal EEG patterns. Key EEG biomarkers identified include altered frontal EEG power, disrupted theta and alpha rhythms, and attenuated P300 components. Hybrid and raw feature-based DL approaches yielded the highest classification accuracies (up to 98.3%) in related neuropsychiatric tasks. EEG-based DL techniques offer promising advancements in diagnosing DDs. However, challenges such as data scarcity, model interpretability, and generalizability persist. Future research should focus on explainable artificial intelligence, multimodal integration, transfer learning, and personalized EEG biomarkers to bridge the gap between research and clinical application.