THE IMOUSE SYSTEM – VISUAL INSPECTION MEETS AI-BASED LONGITUDINAL ANALYSIS OF STANDARDIZED BEHAVIOURS

In translational research, experimental animals remain standard for assessing the efficacy and safety of potential therapeutics. However, comprehensive longitudinal datasets are often not acquired, as manual data collection must be minimized to comply with the 3R principles. Moreover, visual observations and behavioural testing can disrupt resting periods, elevate stress levels, and introduce artificial outcomes.
We aim to tackle these unmet needs by implementing the camera-based digital monitoring system (iMouse) for animal husbandry and experimental use cases by upgrading existing laboratory equipment (retrofit). Here, we examine the impact of manual handling on mice with no visible phenotype. Therefore, we recorded 6w of manual handling. We used 2w pre-experimental video material from 3 perspectives to train AI algorithms. Followed by a time series analysis of standard and unusual behaviours during the 4w experimental data set.
As a result, we showed that recorded video material in the pre-experimental phase was capable of training our existing models to detect behaviours for the specific strain with over 90% precision. Furthermore, we identified unusual phenotypic behaviours during the experimental phase was induced directly by manual handling.
In summary, our results demonstrate, that digital visual inspections of experimental animals are capable to identify a visible phenotype which have been overlooked by human based manual inspection for years. Thus, our findings lead to a better understanding of mouse models, reducing the bias through human handling. At the same time, we increase the data density and provide contact-free continuity of video surveillance.