Automated Detection of Seizures in Rodents: Use of Artificial Intelligence (AI) to Improve Racine Score Scoring

Epilepsy is a condition where the brain's electrical signals misfire, causing sudden, uncontrolled movements or changes in awareness. To understand how to treat this in people, scientists often study it in small animals like mice, which share enough genetic similarity to serve as reliable models. In these experiments, researchers watch the animals closely to grade the severity of their seizures. They use a standard system called the Racine scale, which assigns a number to different types of behavior, from a twitching nose to a full-body convulsion. For decades, this grading has been done by human observers watching video recordings. While necessary, this manual process is slow, tiring, and prone to human error, as different people might interpret the same movement differently. This inconsistency can make it difficult to know if a new medicine is truly working or if the change in behavior was just a matter of who was watching.

A team of researchers from the University of the Punjab and Minhaj University in Pakistan has begun to change how this work is done by teaching computers to watch the videos instead. They developed a system that uses artificial intelligence to automatically spot mice in a video, follow their movements, and classify the severity of their seizures without human help. The researchers started with video footage originally recorded to test a new plant-based treatment for epilepsy. They fed these videos into a computer program designed to recognize objects, specifically training it to find the mice. Once the computer could reliably find the animals, they added a second layer of software to keep track of each individual mouse as it moved around the cage, ensuring the system knew which mouse was doing what, even if they crossed paths.

The computer system proved quite capable at its task. It successfully identified the presence of a mouse in the video frames with a high degree of confidence. More importantly, it learned to distinguish between a mouse that was behaving normally and one that was having a seizure. The system was particularly good at spotting two specific things: a normal, calm state and a severe seizure stage where the mouse rears up on its hind legs and extends them stiffly. It also did well at identifying a stage where the mouse's hind legs are fully extended. While the system was less certain about the very early, subtle stages of a seizure, it handled the clear, dramatic movements with strong accuracy. The researchers found that the computer could correctly identify these specific behaviors about 77 percent of the time on average, a promising result for a machine learning model working with complex biological movement.

The study also highlighted the practical limits of this new approach. The computer struggled a bit with the earliest signs of a seizure because those movements are very small and hard to see, and it rarely saw the most extreme seizure stages in the available footage. The researchers noted that the system took a significant amount of time to process the videos, requiring over three seconds to analyze a single frame on standard equipment. This means that watching an hour-long experiment would take much longer than the experiment itself. However, the team believes that with faster computers and better video recording setups, these hurdles can be overcome. By automating the scoring, the researchers aim to remove the guesswork and fatigue from the process, providing a consistent, objective way to measure how well epilepsy treatments work. This shift from human eyes to digital analysis could lead to more reliable results in drug development, helping scientists find better ways to manage seizures for both animals and people.