Wireless AI-Powered IoT Sensors for Laboratory Mice Behavior Recognition

Abstract

More than 100 million animals are used in research, education, and testing per year, and 95% of them are mice and rats. We have developed wireless artificial intelligent (AI)-powered Internet of Things (IoT) sensors (AIIS) for laboratory mice motion recognition utilizing embedded microinertial measurement units (uIMUs)—a new sensing platform fills an important research gap of monitoring behaviors of many laboratory mice in parallel. We have demonstrated a wireless IoT sensor that could be attached and carried by mice (i.e., animals that typically weigh only 20 g and with body length of ~10 cm) and used the collected motion data to recognize five common mice behaviors (e.g., sleeping, walking, rearing, digging, and shaking) in cages with an accuracy of ~76%. For comparison, current commercial video-based tracking systems that track animal behaviors in real time can reach only 70% accuracy and with limited number of parallelly tracked animals. Furthermore, several machine learning algorithms were explored to solve the imbalanced sample data problem, which allowed the accuracy of mice motion recognition to improve from ~48% to ~76% (if shaking is removed from classification, an average accuracy of 86.46% could be achieved). Less frequent mice behaviors, such as rearing, digging, grooming, drinking, and scratching, could also be recognized at an average accuracy of 96.35%. We believe this work has the potential to revolutionize animal behavioral tracking methodology by offering a solution for large batches of simultaneous small animal motion tracking and AI-based behavior recognition.