Abstract
The reproducibility crisis and translational gap in preclinical research underscore the need for more accurate and reliable methods of health monitoring in animal models. Manual testing is labor-intensive, low-throughput, prone to human bias, and often stressful for animals. Although many smart cages have been introduced, they have seen limited adoption due to either low throughput (being limited to single animals), low data density (a few metrics only), high costs, a need for new space or infrastructure in the vivarium, high complexity use, or a combination of the above. Although technologies for video-based single-animal tracking have matured, no existing technology enables robust and accurate multi-animal tracking in standard home cages. To solve these problems, we built a new type of assay device: the Smart Lid. Smart Lids mount to existing racks, above standard home cages and stream video and audio data, turning regular racks into high-throughput monitoring platforms. To solve the multi-animal tracking problem, we developed a new computer vision pipeline (MOT - Multi-Organism Tracker) along with a new ear tag purpose-designed for computer vision tracking. MOT achieves over 97% accuracy in multi-animal tracking while maintaining an affordable runtime cost (less than $100 per month). The pipeline returns 21 health-related metrics, covering activity, feeding, drinking, rearing, climbing, fighting, cage positioning, social interactions and sleeping, with additional metrics under development.