From Data to Diagnosis: A Systematic Review of EEG Databases for Epilepsy
摘要
Epilepsy remains a high-burden neurological disorder across the lifespan, with a disproportionate impact in low- and middle-income regions where specialist care and timely diagnosis are often constrained. Electroencephalography (EEG) is the most widely available, noninvasive signal source for seizure evaluation, yet the reliability of automated detection and prediction depends less on model novelty than on the quality, structure, and accessibility of the underlying data. In this systematic review, we identified 26 EEG databases explicitly used for epilepsy research, comprising eight open access resources and 18 restricted repositories that require institutional approval, ethics clearance, or data-use agreements. Following a PRISMA-inspired selection workflow, we compared data sets in terms of cohort characteristics, electrode configurations, acquisition procedures, sampling rates, recording duration, and practical access routes to make the landscape auditable and actionable. Across repositories, documentation depth, data organization, and access mechanisms vary sharply, which weakens reproducibility and complicates cross-study benchmarking even when reported performance is high under controlled or subject-dependent protocols. To keep comparisons clinically and methodologically defensible, we propose a hierarchical taxonomy that organizes databases by access regime, recording modality, population focus, and repository scope, clarifying which resources support fair benchmarking and which primarily enable clinically realistic validation, biomarker-driven analyses, prediction, or long-term implantable monitoring.