An automated system for post-surgical health and environmental monitoring with real-time alerts for laboratory rodents using scalable hardware and deep learning

Abstract This proposal responds directly to the PAR-21-225 Novel Tools and Devices for Animal Research Facilities and to Support Care of Animal Models. While it is standard practice to continuously monitor the health of human patients following a major surgery, this is typically not the case for monitoring recovery in rodent pre- clinical research. For example, it is common practice that rats and mice are provided an hour of post-surgical supervision and then returned to their home cage and colony room with no follow-up until the following day. Thus, animals can go un-monitored for more than 10 to 12 hours following a major surgery. Unmonitored animals can suffer from post-surgical complications which may result in undue animal suffering, add unnecessary variability to ongoing experiments, or reduce reproducibility within and among research groups. Studies involving new procedures, surgical techniques or new personnel have seen very high mortality rates (50% - 90%). While the monetary cost of the loss of a rat or mouse is low, the time invested to train the animal or to build the implant (e.g., electrode array) can be very high. We propose a wireless and automated in-cage monitoring and alerting system that incorporates multiple environmental sensors, a thermal camera, and advanced machine learning (ML) software to identify aberrant post-surgical behavior (e.g., hunched posture), fluctuations in body temperature, and unforeseen environmental conditions (e.g., changes in within-cage temperature, humidity, light, and noise) that could affect animal welfare and survival. By identifying typically unobserved changes in the animal and the in-cage environment, this device could improve reproducibility of research by alerting researchers via email or text notifications following unplanned variations in conditions. The proposed final system will be low-power, wireless and can be mounted on any standard rodent home cage. It utilizes a thermal-imaging camera combined with ML based anomaly detection to identify changes in behavior (e.g., changes in movement, body temperature, and posture) and environmental conditions. This system will be developed through close collaboration with veterinary professionals and researchers and the machine learning models trained on thermal imaging video data from cohorts of rats and mice under various conditions so that it can identify behaviors associated with acute pain, distress, and health complications. This system has the potential to improve the wellbeing and survival of laboratory rodents as well as providing researchers with important behavioral and environmental data that can be monitored and stabilized to improve the reproducibility of ongoing experiments.