Machine learning based seizure classification and digital biosignal analysis of ECT seizures

Area of Science:

• Neuroscience
• Medical Technology
• Artificial Intelligence

Background:

• Electroconvulsive therapy (ECT) applications in medicine are expanding, yet AI integration remains limited.
• Digital seizure collection systems necessitate advanced digital seizure analysis for improved ECT.
• AI offers enhanced analytical capabilities for precision in ECT procedures.

Purpose of the Study:

• To develop the first machine learning (ML) framework for classifying ictal and non-ictal EEG segments in ECT.
• To accurately identify seizure endpoints for deriving seizure quality parameters.
• To compute ECT seizure quality metrics using ML with high reliability.

Main Methods:

• Developed an ML framework to classify electroencephalogram (EEG) segments during ECT.
• The model was trained to discriminate between seizure (ictal) and non-seizure (non-ictal) EEG activity.
• Evaluated the ML model's accuracy, precision, and sensitivity in identifying seizure endpoints.

Main Results:

• The ML model achieved 89% accuracy, precision, and sensitivity in discriminating ictal from non-ictal EEG segments.
• Reproducible ECT quality parameters derived from ML showed high correlations (up to 0.99) with device-calculated values.
• Mean seizure duration differences were minimal compared to expert raters and stimulation devices.

Conclusions:

• Integrating ML into ECT significantly enhances the precision of seizure detection and quality metric calculation.
• Accurate seizure detection is crucial for reliable seizure duration determination and quality index derivation.
• This approach paves the way for individualized ECT treatment strategies and novel seizure quality assessment methods.