A two-stage algorithm to detect electrographically focal seizures using a wearable single-channel EEG sensor.

Area of Science:

• Medical Technology
• Machine Learning
• Neurology

Background:

• Epilepsy monitoring often relies on complex EEG setups.
• Wearable single-channel EEG sensors offer a more accessible approach to continuous monitoring.
• Accurate and reliable seizure detection algorithms are crucial for clinical decision-making and patient care.

Purpose of the Study:

• To develop and evaluate a two-stage machine learning model for electrographic seizure detection.
• To assess the model's performance using wearable single-channel scalp EEG data.
• To improve seizure detection sensitivity and reduce false alert rates compared to single-stage methods.

Main Methods:

• A two-stage machine learning algorithm was designed for seizure detection.
• Stage I detects potential seizures in short segments; Stage II refines these detections to minimize false alerts.
• A post-processing framework was applied to segment-level results for event-level decisions.

Main Results:

• The two-stage system demonstrated statistically significant improvements in detecting electrographically focal seizures.
• Sensitivity increased from 61% to 75% with a reduction in false alert rate from 3.3/hr to 2.4/hr.
• Enhancements to Stage I, including memory and iterative learning, further improved performance.

Conclusions:

• The two-stage algorithm offers superior performance for focal seizure detection compared to single-stage approaches.
• This technology holds potential for enhancing support systems for epileptologists and enabling long-term seizure monitoring.
• The developed system represents a step towards practical, long-duration seizure monitoring using wearable EEG devices during daily activities.