Refining animal care through technology: Addressing alopecia in Jaculus jaculus with validated computer vision analysis

Abstract

We validated the use of an open-source computer vision toolkit to analyze high-quality behavioral data and evaluate welfare in the Lesser Egyptian Jerboa ( Jaculus jaculus ). Movements of these small, nocturnal rodents are rapid and difficult to observe, potentially obscuring behavioral assessment. However, assessment became warranted when alopecia and jumping were noted. We compared trained human observers to machine learning trained computer vision algorithms, evaluating accuracy and precision in behavioral classification. Human observers categorized behaviors with an overall accuracy of 0.71 ± 0.11 and an intraclass correlation coefficient (ICC) of 0.92 ± 0.07, with greater odds of misidentifying behaviors lasting less than one second. Computer vision classifiers successfully met human-grade accuracy and ICC, with significantly less sensitivity to behavioral duration. As 34% of manually classified behaviors lasted less than 0.5 seconds, we used computer vision to annotate activity budgets of captive jerboas before and after adding novel enrichment. Alopecia was significantly associated with grooming, and while grooming was negatively associated with terrarium height and with opaque dividers between terraria, conventional rodent enrichment had no significant effect on behavior. Inflammatory causes of alopecia were not found with cytologic, molecular, or histopathologic analysis. These results suggest captive jerboa may demonstrate psychogenic alopecia. Furthermore, computer vision automation allows for fast, accurate analysis of large volumes of behavioral data that can be used to tailor species-specific husbandry practices and improve animal welfare.