Computational approaches to deep learning-driven EEG analysis for neurological disease diagnosis: a systematic review

Electroencephalography (EEG) is a widely used non-invasive neurophysiological technique for analyzing brain dynamics and diagnosing neurological diseases. However, the non-stationary nature, low signal-to-noise ratio, and inter-subject variability of EEG signals cause significant challenges to traditional analytical approaches. In recent years, deep learning has emerged as a powerful computational tool for EEG analysis, which enables automated feature learning and improved diagnostic accuracy with minimal human-driven feature extraction. This review systematically explains computational approaches in deep learning-driven EEG analysis for the diagnosis of neurological diseases. It includes recent advances in deep learning architectures, such as convolutional neural networks (CNN), recurrent neural networks (RNN), hybrid configurations, Generative Adversarial Networks (GAN), Large Language Models (LLM), autoencoders, and transformer-based models, with a focus on their capacity to model spatial, temporal, and spatiotemporal dynamics in EEG signals. This is applicable in diagnosing major neurological disorders such as Epilepsy, Alzheimer's disease (AD), Encephalopathy, Parkinson's disease (PD), Mild Cognitive Impairment (MCI), sleep disorders, and developmental brain disorders. Furthermore, we discuss essential preprocessing strategies and denoising techniques, showing how removing biological artifacts prevents diagnostic models from confusing noise with actual signs of disease. Key challenges such as limited data availability, lack of model interpretability, inter-subject variability, and difficulty in clinical translation are identified and critically discussed. Finally, key emerging trends and future research directions are identified to support the development of clinical applications of EEG-based disease diagnostic systems.