Trainable, vision-based automated home cage behavioral phenotyping

Abstract

We describe a fully trainable computer vision system enabling the automated analysis of complex mouse behaviors. Our system computes a sequence of feature descriptors for each video sequence and a classifier is used to learn a mapping from these features to behaviors of interest. We collected a very large manually annotated video database of mouse behaviors for training and testing the system. Our system performs on par with human scoring, as measured from the ground-truth manual annotations of thousands of clips of freely behaving mice. As a validation of the system, we characterized the home cage behaviors of two standard inbred and two nonstandard mouse strains. From this data, we were able to predict the strain identity of individual mice with high accuracy.