Automatyczne wykrywanie napadów padaczki u gryzoni: wykorzystanie sztucznej inteligencji do poprawy oceny skali Racine'a

Abstract Epilepsy is a chronic neurological condition that presents episodic and unprovoked seizures, resulting in significantly impacting the quality of life. The presented research aims to revolutionize conventional seizure detection methods by leveraging the power of AI and considerably improving the accuracy and efficiency of seizure detection. Specifically, the YOLOv8 model detects and classifies alongside Deep-Sort for tracking rodents. This research used a set of videos recorded under high-confidence, controlled conditions designed for developing novel plant-based therapy directed at epilepsy. The YOLOv8 model was trained and detected rodents with an average precision of 77.1%. Moreover, Deep-Sort provided successful and efficient tracking of rodents. The outcomes of the YOLOv8 classification model highlight its accuracy in predicting seizure classification. The model performed particularly well in classifying 'stage-4' and 'normal-stage' with high precision and accuracy, and 'Hind-Limb-Extension' with strong recall. These results imply that the model effectively identified hind limb extension and the two stages with a very low misclassification rate, demonstrating the model's utility for detecting both seizure and normal conditions.