Benchmark of EEG-based seizure detection algorithms with SzCORE

Area of Science:

• Neuroscience
• Biomedical Engineering
• Medical Informatics

Background:

• Epilepsy diagnosis and treatment rely on EEG monitoring, necessitating reliable automated seizure detection.
• Existing reviews face challenges due to a lack of standardized evaluation methodologies for seizure detection algorithms.
• Direct comparison of seizure detection algorithms is difficult, hindering progress in the field.

Purpose of the Study:

• To provide a fair and transparent comparison of state-of-the-art EEG-based seizure detection algorithms.
• To address the lack of standardization in algorithm evaluation using a novel framework, SzCORE.
• To highlight discrepancies between reported and standardized performance metrics for seizure detection algorithms.

Main Methods:

• Literature review of patient-independent EEG-based seizure detection algorithms trained on public datasets.
• Re-implementation and evaluation of selected algorithms using the standardized SzCORE framework.
• Analysis of performance metrics, including sensitivity, precision, and variability across datasets.

Main Results:

• Nineteen relevant papers were identified; three algorithms were re-implemented and evaluated.
• Significant discrepancies were found between reported and standardized algorithm performances.
• Algorithms demonstrated high sensitivity (>90%) but low precision (10-40%), indicating high false-positive rates.
• Performance variability was observed, potentially linked to seizure frequency in evaluation datasets.

Conclusions:

• Standardized evaluation methodologies, like SzCORE, are crucial for transparent benchmarking of EEG-based seizure detection algorithms.
• Current state-of-the-art algorithms require improvement, particularly in reducing false positives.
• Further research is needed to enhance algorithm accuracy and reliability for clinical application in epilepsy management.