Detection of Postictal Generalized Electroencephalogram Suppression: Random Forest Approach

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Sudden unexpected death in epilepsy (SUDEP) is a significant cause of premature death in epilepsy patients.
• Postictal generalized electroencephalogram (EEG) suppression (PGES) following seizures is a key risk factor for SUDEP.
• Automated PGES detection is crucial for consistent and efficient SUDEP risk assessment, overcoming limitations of manual analysis.

Purpose of the Study:

• To develop and present a random forest-based method for automatic PGES detection.
• To utilize multichannel human EEG recordings from epilepsy monitoring units for PGES detection.
• To evaluate the performance of the developed algorithm using a novel time-distance-based metric.

Main Methods:

• Feature extraction from EEG signals, including temporal, frequency, wavelet, and interchannel correlation features.
• Training a random forest classifier using the extracted features for PGES detection.
• Implementation of confidence-based correction rules and a time-distance-based evaluation method.

Main Results:

• The random forest approach achieved a 0.95 positive prediction rate with a 5-second tolerance on artifact-free EEG signals.
• Performance varied between 0.68 and 0.81 for signals with varying levels of artifacts.
• The time-distance-based evaluation demonstrated improved performance with reduced signal artifacts.

Conclusions:

• A novel feature-based random forest method for automatic PGES detection in multichannel EEG was successfully developed.
• The algorithm's performance is sensitive to signal artifact levels, highlighting the need for robust artifact handling.
• Further research is required to enhance PGES detection algorithm performance across diverse artifact conditions for clinical utility.