Epidemiological Dynamics of a Visually Apparent Disease: Camera Trapping and Machine-Learning Applied to Rumpwear in the Common Brushtail Possum.

Abstract

Visually apparent diseases are valuable for investigating and monitoring the occurrence and prevalence of pathogens in wildlife populations through passive monitoring methods like camera trapping. Rumpwear, characterized by visible clinical signs of hair breakage and damage on the lumbosacral region, affects common brushtail possums (Trichosurus vulpecula) across Australia. However, the etiology of rumpwear remains unclear, and the spatiotemporal factors are understudied. This study investigated the epidemiology of rumpwear in common brushtail possums at Adamsfield, Tasmania (Australia), and predicted rumpwear distribution across the Tasmanian landscape. We visually classified images of rumpwear clinical signs in 6908 individual possums collected from a 3-year camera trapping network. Our results revealed that: (1) adults were twice as likely to show signs of rumpwear compared to young possums; (2) rumpwear occurrence increased with the relative activity of possums at a site; and (3) prevalence of rumpwear was seasonal, being lowest in May (3.2%-late autumn) and highest in December (27.1%-early summer). Collectively, these findings suggest that the occurrence of rumpwear may be density dependent, the putative etiological agent seems to be influenced by seasonal factors or site use. Additionally, a convolution neural network (CNN) was trained to identify rumpwear automatically based on the manually (human-expert) classified camera trap images. Applying the trained classifier to 38,589 brushtail possum images from across Tasmania, the CNN predicted that rumpwear is widespread, with an overall prevalence of 18.6%. This study provides new insights into rumpwear epidemiology and identified factors for further investigating within this host-pathogen system.