Epileptic seizure detection with Convolutional Neural Networks and the Continuous Wavelet Transform

Abstract

The study of epileptic seizure often involves animal models to simulate the human behavior. Such models demand monitoring the evolution of the animal behavior continuously. Detecting seizure in this setup remains a challenge, because it typically requires trained personnel to annotate video sequences looking for the timestamps of seizure events. Deep Learning methods can help to solve this task in a more automatic and efficient manner due to their capacity of retrieving patterns from data. In this work, we conducted a pilot study to detect epileptic seizure from the images of small rodents using Convolutional Neural Networks (CNN) and the Continuous Wavelet Transform (CWT). We used the Social LEAP Estimates Animal Poses (SLEAP) framework for animal recognition to extract the morphological skeleton. Then, our CWT-CNN method used information of the frequency, magnitude and temporal evolution of head and thorax displacements to classify the animal behavior. The results showed a mean accuracy of 82.7%in the classification of epileptic seizure events.