Development and Application of Machine Learning-Based Digital Biomarkers for Monitoring Spontaneous Seizures in Preclinical Epilepsy Models

For this work, we ran a natural history study utilizing home-cage video data from two mouse models of Dravet Syndrome, a severe epileptic encephalopathy. Dravet mice and wildtype littermates were weaned into video-integrated cages, where monitoring occurred from postnatal day 21 to postnatal day 50. To identify spontaneous seizures, machine-based algorithms were trained to detect “taggable” features of tonic-clonic seizures, such as loss of righting reflex. Loss of righting reflex, a reliable feature of loss of consciousness in mice, occurs consistently during spontaneous seizure.

Here we describe the development of the machine learning technology, and the application in spontaneous seizure monitoring and multiplexing other phenotypic readouts (activity, sleep wake cycle, etc.).