Artificial intelligence driven neuropsychiatry: a systematic review of electroencephalography-based computational...

Major Depressive Disorder is the most prominent global mental health issue impacting millions of individuals worldwide. Electroencephalogram signals capturing intricate brain dynamics have emerged as a promising modality for predicting depression. This systematic review synthesizes research on computational techniques for depression prediction using neural signals, addressing the lack of standardized computational frameworks. The review identifies the significant variations in preprocessing pipelines such as bandpass filtering, artifact removal, and feature extraction techniques to assess their impact on predictive performance. A comparative analysis revealed significant variability in methodological approaches with deep learning architectures particularly convolutional neural networks and hybrid convolutional neural networks- Long Short-Term Memory models outperform traditional machine learning methods. Support Vector Machine also demonstrated competitive performance in several studies in terms of accuracy. Single channel and few-electrode configurations also show potential for cost-effective and portable diagnostic tools. Despite some studies reporting classification accuracies exceeding 90%, a lack of standardized evaluation protocols hinders the comparability of findings, representing a translational barrier for clinical adoption. This reflects a broader critical trend where methodological inconsistencies, data heterogeneity, and limited model interpretability hinder the generalizability and reliability of model. Future research should prioritize standardized protocols, diverse datasets, multimodal data integration, and explainable Artificial Intelligence to enhance clinical applicability. These insights provide a framework for artificial intelligence-driven electroencephalogram analysis to revolutionize depression diagnostics by guiding researchers to advance precision psychiatry through innovative and scalable solutions.