False alarms reduction in non-convulsive status epilepticus detection via continuous EEG analysis

Area of Science:

• Neurology
• Biomedical Engineering
• Signal Processing

Background:

• Computer-assisted monitoring systems for non-convulsive status epilepticus (NCSE) generate frequent false alarms.
• These false alarms can compromise patient safety during continuous electroencephalography (EEG) monitoring.
• Common causes include misinterpreting abnormal background activity, short ictal discharges, and continuous interictal discharges as ictal events.

Purpose of the Study:

• To reduce false alarms in NCSE detection.
• To improve the accuracy of automated seizure detection systems.
• To enhance patient safety by minimizing unnecessary alerts.

Main Methods:

• Analysis of 127-hour EEG recordings from 10 participants with 310 ictal discharges.
• Integration of morphological features (visibility graph) with time-frequency features to address abnormal background activity.
• Development of a 4-class classifier using synthetic data ('Non-ictal', 'Ictal', 'Suspected Non-ictal', 'Suspected Ictal') to differentiate subtle discharge types.
• Comparison of the 4-class model against a standard 2-class model using precision-recall curves and leave-one-out cross-validation.

Main Results:

• The 4-class classification model significantly improved performance over the standard 2-class model.
• With time-frequency features alone, the 4-class model increased precision by 15% at 80% sensitivity.
• Incorporating morphological features, the 4-class model achieved a group-level sensitivity of 93% ± 12% and precision of 55% ± 30%.
• 100% accuracy was achieved on a subset of 4.3-hour recording with 5 ictal discharges.

Conclusions:

• The proposed 4-class classification model effectively reduces false alarms in NCSE detection.
• The combination of morphological and time-frequency features enhances the reliability of EEG-based seizure detection.
• This approach holds significant potential for improving patient safety in clinical settings.