Epileptic seizure detection using CHB-MIT dataset: The overlooked perspectives.

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Epilepsy is a serious neurological disorder requiring accurate seizure detection.
• Manual detection of epileptic seizures (ES) from electroencephalography (EEG) is time-consuming and impractical.
• Existing automated seizure detection systems often neglect critical real-world factors like data imbalance and inter-subject variability.

Purpose of the Study:

• To develop and evaluate a generalized, cross-subject seizure event detection system for continuous EEG signals.
• To address key challenges in real-world seizure detection: class imbalance, subject variability, and event-based detection.

Main Methods:

• Utilized the CHB-MIT continuous EEG dataset.
• Extracted 92 features from 5-second non-overlapping windows, selecting the top 32 significant features per channel.
• Employed a Random Forest (RF) classifier for segment classification and a post-processing step for event detection.

Main Results:

• Achieved 72.63% sensitivity in subject-wise 5-fold cross-validation.
• Achieved 75.34% sensitivity in leave-one-out cross-validation.
• Demonstrated a practical approach to seizure event detection in continuous EEG data.

Conclusions:

• The proposed system effectively addresses critical real-world factors for automated seizure detection.
• This study advances the understanding and applicability of EEG-based seizure event detection systems.
• The findings contribute to developing more robust and reliable tools for epilepsy management.