DEEP LEARNING FOR AUTOMATED EEG SIGNAL CLASSIFICATION IN EPILEPSY: CURRENT EVIDENCE AND CLINICAL IMPLEMENTATION PERSPECTIVES

Abstract

Epilepsy affects approximately 50 million people globally, with seizures representing the defining clinical manifestation in approximately 70% of patients. Electroencephalography (EEG) remains the primary diagnostic tool for epilepsy characterisation, but its interpretation requires highly specialised expertise concentrated in tertiary epilepsy centres inaccessible to the majority of affected individuals, particularly in low- and middle-income countries. Deep learning — encompassing convolutional neural networks, recurrent architectures, and transformer models — applied to EEG signal data has achieved seizure detection sensitivity of 90–98% and specificity of 92–99% in controlled research settings, approaching expert electroencephalographer performance. This review analyses 39 studies of deep learning EEG analysis in epilepsy published between 2017 and 2024, evaluating model architectures, input signal representations, performance benchmarks, and clinical deployment challenges. Particular attention is given to implementation barriers in resource-limited healthcare systems, including Uzbekistan, where epilepsy services are predominantly confined to major urban centres.