T. HANANIA, P. KABITZKE, M. MAZELLA, I. FILIPOV, V. ALEXANDROV, D. BRUNNER
The assessment of social behavior has become critical for the phenotyping of animals models of autism and schizophrenia. Scoring of videos is both time consuming and subjective, causing inter-observer and between-studies variability, thus emphasizing the need for automated scoring. Computer vision classification of social behavior is, on the other hand, challenging, due to the frequency of occlusions and contacts between mice. We have solved the problem of occlusion with sophisticated computer vision techniques and individual tags. Careful choice of algorithms also allows us to analyze videos in almost real-time, by minimizing CPU time. Our choice of lighting and camera sensitivity also allow us to track mice during the dark circle and to score their behavior 30 times a second, 24 hours a day, up to 6 days continuously. Assessment of social behavior in a non-stressful environment can also produce qualitatively different phenotypes, as the behavior is not confounded with anxiety or a response to manipulation. We will present data obtained with mouse models of autism, that demonstrate the utility of the system.