Accurate detection of spontaneous seizures using a generalized linear model with external validation

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Electroencephalography (EEG) analysis is crucial for neurocritical care, epilepsy management, and pre-clinical research.
• Existing seizure detection methods lack widespread adoption due to limited validation.
• There is a need for robust, high-throughput seizure detection systems in both clinical and research settings.

Purpose of the Study:

• To develop and validate a high-performance seizure detection approach using EEG signal features.
• To create a generalized linear model for accurate seizure classification.
• To provide a reliable tool for automated EEG analysis and interventional approaches.

Main Methods:

• A generalized linear model was trained on 141 EEG signal features from a Focal Epilepsy dataset (16 rats, 1012 seizures).
• The model incorporated features from time, frequency, univariate, and multivariate domains.
• Performance was validated on independent Focal Epilepsy and Multifocal Epilepsy datasets (96 rats, 2883 seizures).

Main Results:

• The pooled classifier achieved an Area Under the Receiver Operating Characteristic (AUROC) of 0.995 on the Focal Epilepsy dataset.
• Validation on the Multifocal Epilepsy dataset yielded a pooled AUROC of 0.963.
• Detection latency was under 5 seconds for over 80% of seizures.

Conclusions:

• The developed seizure detection method demonstrates superior performance across multiple independent datasets.
• This approach is suitable for automated EEG analysis pipelines and closed-loop interventions.
• It offers significant utility for high-throughput, standardized seizure analysis in animal research.