Epileptic Seizure Detection Using Machine Learning: A Systematic Review and Meta-Analysis.

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Epileptic seizures are unpredictable, significantly impacting patient quality of life.
• Early and accurate seizure detection is critical for effective epilepsy management.
• Machine learning (ML) offers automated seizure detection capabilities using electroencephalogram (EEG) signals.

Purpose of the Study:

• To conduct a meta-analysis evaluating the performance of ML models for epileptic seizure detection.
• To identify factors influencing ML model performance, including model type, data preprocessing, and dataset characteristics.
• To provide an evidence-based foundation for developing intelligent seizure detection tools.

Main Methods:

• Systematic literature search across multiple databases up to April 2025.
• Inclusion of 60 studies and 93 datasets for meta-analysis.
• Calculation of pooled sensitivity, specificity, and AUC using Stata 17.0.
• Subgroup analyses to investigate heterogeneity and publication bias.

Main Results:

• ML models achieved high pooled performance: sensitivity 0.96, specificity 0.97, and AUC 0.99.
• Significant heterogeneity was observed, influenced by model type, data preprocessing, and dataset characteristics.
• The findings indicate robust performance of ML in EEG-based seizure detection.

Conclusions:

• ML models show substantial promise for automated, EEG-based epileptic seizure detection.
• Integrating ML into imaging devices could enhance early epilepsy diagnosis.
• Further large-scale, multicenter clinical studies are essential to validate ML algorithms for real-world application, interpretability, and safety.