Deep Learning Approaches for EEG-Based Automatic Schizophrenia Identification: A Review

Abstract

Schizophrenia is a severe neuropsychiatric disorder characterized by disruptions in perception, cognition, and behavior, making early and accurate diagnosis crucial for effective treatment. Electroencephalography (EEG), due to its non-invasive nature and high temporal resolution, has emerged as a promising modality for identifying neural abnormalities associated with schizophrenia. Recent advances in deep learning and optimization techniques have significantly improved the performance of automated diagnostic systems. This review presents a comprehensive analysis of modern approaches that integrate dynamic functional connectivity (DFC) analysis with deep learning frameworks, particularly focusing on deep stack-augmented conditional variational autoencoders (DSA-CVAE). DFC enables the modeling of time-varying interactions between brain regions, providing richer representations of neural dynamics. Meanwhile, DSA-CVAE enhances feature learning through hierarchical latent representations and conditional constraints, improving classification accuracy and robustness. The paper systematically reviews recent studies, compares methodologies, and highlights key trends, challenges, and opportunities in this domain. Furthermore, the integration of optimization strategies such as metaheuristics and adaptive learning is discussed to enhance model performance. This review aims to provide insights into the evolving landscape of EEG-based schizophrenia detection and to guide future research toward more interpretable, scalable, and clinically applicable solutions.