Automated Pain Recognition in Horses from Facial Images

The Horse Grimace Scale has been shown to be an effective and reliable method of assessing pain from facial images following routine castration in horses. Yet being dependent on a human observer, it is still subject to human bias and subjectivity. This leads to the interest in the development of automated approaches in this domain. In this study we compare two different deep learning approaches to automated pain recognition in horses from lateral facial images using a dataset of n=39 horses undergoing a routine castration. The first method directly classifies from image embeddings and achieves over 73% accuracy in pain recognition. The second method involves regression from embeddings to Facial Action Unit (FAU) scores, surpassing the first with an accuracy exceeding 79%. In terms of accuracy and precision, both methods are comparable and surpass human Horse Grimace Scale (HGS) scoring, with the latter method demonstrating higher recall.