Rodent Tracking and Abnormal Behavior Classification in Live Video using Deep Neural Networks

This study is focused on developing tools to assess behavioral activity in neonatal mice undergoing opioid withdrawal. Opioids are used to manage pain in both young and old, but debilitating withdrawal symptoms can occur when opioid medications are stopped. We propose to apply various pre-trained neural network models such as ResNet, MobileNet, and EfficientNet to track mice behavior in live video using the DeepLabCut package. We track the animal by detecting movement and positioning of the “nose”, “right ear”, “left ear” and “tail base” in each video frame. Tracking is a prerequisite to detecting the abnormal behavior observed during precipitated withdrawal. We compare the performance of the different pretrained models in terms of accuracy. The highest accuracy that we achieved under abnormal conditions-opioid withdrawal using ResNet_50 was 95.82%.