Detecting temporal lobe seizures in ultra long-term subcutaneous EEG using algorithm-based data reduction.

Area of Science:

• Neurology
• Biomedical Engineering
• Medical Technology

Background:

• Epilepsy management relies on accurate seizure tracking, often using subjective patient diaries.
• Subcutaneous electroencephalography (sqEEG) offers objective, long-term outpatient seizure monitoring.
• Automated seizure detection algorithms are crucial for efficient analysis of sqEEG data.

Purpose of the Study:

• To evaluate the performance of a deep-neural-network algorithm for automatic seizure detection using sqEEG data.
• To assess the algorithm's sensitivity and specificity in identifying electrographic seizures.

Main Methods:

• Analysis of multicenter sqEEG recordings from 9 people with epilepsy (PWE) and 12 healthy subjects (965 total days).
• Comparison of algorithm-detected seizures against annotations from three human experts.
• Calculation of data reduction ratios, sensitivity, and false detection rates.

Main Results:

• High data reduction ratios achieved: 99.6% in PWE and 99.9% in controls.
• Cross-PWE sensitivity was 86% (median 80%), with a median false detection rate of 2.4 per 24 hours.
• The algorithm demonstrated clinically applicable specificity.

Conclusions:

• The sqEEG-based automatic seizure detection algorithm performs with high sensitivity and clinically applicable specificity.
• This automated step is effective for semi-automatic seizure detection and review processes.
• Ultra long-term sqEEG monitoring holds significant potential for improving objective seizure quantification in epilepsy.