RodentEpiNet: AI-Powered Recognition and Analytics for Behavioral Phenotype of Epilepsy in Rodent

Abstract

Epilepsy is a chronic neurological condition affecting over 65 million individuals worldwide, with approximately 5 million new cases annually. Characterized by spontaneous recurrent seizures (SRS), its severity ranges from subtle behavioral changes to life-threatening tonic-clonic seizures, which carry a significant risk of sudden unexpected death in epilepsy (SUDEP). Rodent models, such as mice and rats, are indispensable in preclinical epilepsy research for understanding seizure mechanisms and evaluating anti-epileptic drugs (AEDs). However, current seizure monitoring methods face significant limitations.

Video-electroencephalogram (video-EEG) monitoring, while effective, is invasive, disrupts natural behaviors, and raises ethical concerns, especially for long-term studies. Moreover, video-EEG systems are costly, resource-intensive, and require specialized setups. Manual video observation, though non-invasive, is labor-intensive, prone to human error, and inconsistent, particularly for subtle seizures (stages 1-3). These limitations highlight the need for scalable, non-invasive, and efficient tools for seizure detection in preclinical research.

To address these challenges, we developed RodentEpiNet, an automated, video-based system for detecting and classifying seizures in freely behaving rodents. Leveraging deep learning, it integrates an Action Recognition Network (ARN) to identify seizure behaviors and an Object Detection Network (ODN) for tracking animals. Advanced techniques such as transfer learning, focal loss, and data augmentation enable RodentEpiNet to achieve 98.9% sensitivity and 99.9% specificity, even with imbalanced datasets.

RodentEpiNet generalizes effectively across environments and datasets, adapting from home-caged mice to chamber-based settings with rapid high performance. Beyond seizure detection, it facilitates comprehensive behavioral analysis, revealing long-term epilepsy-related changes and potential behavioral biomarkers for screening and diagnosis. Its capabilities extend to rats, enabling automatic analysis of seizure behaviors and daily activities, uncovering species-specific differences. RodentEpiNet also supports translational research, introducing a Pig Racine Scale for precise seizure classification in epilepsy pig models.

Beyond epilepsy, RodentEpiNet has applications in broader neuroscience research, such as studying behavioral changes due to Chronic Unpredictable Mild Stress (CUMS) in mice. It highlights stress-induced alterations and tracks recovery, offering insights into stress-related disorders.

By automating long-term video analysis and enabling precise behavioral quantification, RodentEpiNet addresses critical limitations of current methods. Its non-invasive, scalable, and accurate nature significantly enhances the efficiency and ethical standards of preclinical epilepsy research. With its ability to generalize across species, environments, and datasets, RodentEpiNet is a transformative tool for advancing epilepsy research and translational neuroscience.

Date of Award	28 Jul 2025
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Jufang HE (Supervisor)