Visual detection of seizures in mice using supervised machine learning.

Area of Science:

• Neuroscience
• Computational Biology

Background:

• Seizures manifest as abnormal brain activity, traditionally scored visually using scales like the Racine scale.
• Visual seizure scoring is time-consuming, subjective, and limits high-throughput research.
• Quantitative, scalable methods are needed for accurate seizure assessment in preclinical models.

Purpose of the Study:

• To develop automated machine learning classifiers for predicting seizure severity from noninvasive video data.
• To enable high-throughput, objective, and standardized seizure scoring in preclinical epilepsy research.
• To facilitate downstream applications in neurogenetics and therapeutic discovery.

Main Methods:

• Supervised machine learning approaches were employed.
• Video-only classifiers were trained using the pentylenetetrazol (PTZ)-induced seizure model in mice.
• Classifiers predicted ictal events, univariate seizure intensity, and time-varying seizure intensity scores.

Main Results:

• Automated classifiers accurately predicted seizure events and intensity directly from overhead video recordings.
• The study demonstrates the first rigorous quantification of seizure events and intensity using supervised methods from video data.
• The developed approach enables objective, high-throughput seizure scoring.

Conclusions:

• Supervised machine learning applied to noninvasive video data provides a robust method for quantifying seizure severity.
• This automated approach overcomes limitations of traditional visual scoring, offering scalability and objectivity.
• The findings support advanced applications in epilepsy research, including neurogenetics and drug discovery.