Application Of Machine And Deep Learning In Epileptic Seizure Detection a Systematic Review

Abstract

Epilepsy, the fourth most common neurological disorder, affects 1% of the global population. Manual EEG-based seizure detection is error-prone, impacting diagnostic and prognostic accuracy. To evaluate the performance of machine and deep learning methods in improving detection of epileptic seizures, define challenges and map out future opportunities for clinical translation. This review collected and analyzed studies published between 2015 and 2023 indexed in PubMed, IEEE Xplore, and ScienceDirect. This review selected studies with an approach to seizure detection based on ML and DL for electroencephalogram  (EEG) data. Articles were excluded if they were non-English, were reviews, or did not present empirical results. Text, metrics and challenges data related to algorithms were captured and processed. The review included 50 studies which used over 10 public EEG datasets including CHB-MIT and Bonn. The traditional ML algorithms like SVM resulted in an accuracy of 90%, and modern DL models like the CNNs had an accuracy of 95% or more. Hybrid CNN-LSTM models achieved the best rates compared to other techniques, as they used spatial and temporal EEG data jointly. Disadvantages are focused on data, interpretability, computational complexity, and a need for proprietary hardware. This study highlights the potential of machine and deep learning for high-accuracy automated seizure detection. Addressing data quality, interpretability, and computational demands is crucial for clinical implementation. Future AI models should prioritize personalization and explain ability

Keywords

Epilepsy detection, EEG analysis, Deep learning, Machine learning, seizure