Band-sensitive seizure onset detection via CSP-enhanced EEG features

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Epileptic seizure onset detection from scalp electroencephalography (EEG) is crucial for patient care and research.
• Existing methods face challenges in accurately discriminating between seizure and non-seizure states, impacting sensitivity and detection latency.

Purpose of the Study:

• To develop and evaluate two novel epileptic seizure onset detectors based on common spatial pattern (CSP) feature enhancement.
• To improve the discrimination between seizure and non-seizure EEG signals for more accurate and timely detection.

Main Methods:

• Implemented a feature enhancement stage using CSP to increase variance between seizure and non-seizure EEG data.
• Developed two detectors: one using a Support Vector Machine (SVM) with energy features from subbands, and another pooling SVM detections across spectral bands using logical operators.

Main Results:

• The first detector achieved 95.2% sensitivity, 6.43s latency, and 0.59 false alarms/hour.
• The second detector (with MAJORITY fusion) achieved 100% sensitivity, 7.28s latency, and 1.2 false alarms/hour.
• Both detectors demonstrated improved performance over state-of-the-art methods in sensitivity and detection latency.

Conclusions:

• The proposed CSP-based feature enhancement effectively improves seizure onset detection from EEG.
• The two novel detectors offer promising advancements in epilepsy monitoring, providing high sensitivity and reduced detection latency.