Explainable automated pain recognition in cats

Abstract

Manual tools for pain assessment from facial expressions have been suggested and validated for several animal species. However, facial expression analysis performed by humans is prone to subjectivity and bias, and in many cases also requires special expertise and training. This has led to an increasing body of work on automated pain recognition, which has been addressed for several species, including cats. Even for experts, cats are a notoriously challenging species for pain assessment. A previous study compared two approaches to automated ‘pain’/‘no pain’ classification from cat facial images: a deep learning approach, and an approach based on manually annotated geometric landmarks, reaching comparable accuracy results. However, the study included a very homogeneous dataset of cats and thus further research to study generalizability of pain recognition to more realistic settings is required. This study addresses the question of whether AI models can classify ‘pain’/‘no pain’ in cats in a more realistic (multi-breed, multi-sex) setting using a more heterogeneous and thus potentially ‘noisy’ dataset of 84 client-owned cats. Cats were a convenience sample presented to the Department of Small Animal Medicine and Surgery of the University of Veterinary Medicine Hannover and included individuals of different breeds, ages, sex, and with varying medical conditions/medical histories. Cats were scored by veterinary experts using the Glasgow composite measure pain scale in combination with the well-documented and comprehensive clinical history of those patients; the scoring was then used for training AI models using two different approaches. We show that in this context the landmark-based approach performs better, reaching accuracy above 77% in pain detection as opposed to only above 65% reached by the deep learning approach. Furthermore, we investigated the explainability of such machine recognition in terms of identifying facial features that are important for the machine, revealing that the region of nose and mouth seems more important for machine pain classification, while the region of ears is less important, with these findings being consistent across the models and techniques studied here.

Automated Pain Recognition in Cats using Facial Landmarks: Dynamics Matter

“…For this study, we used two datasets: (i) Finka et al 25 and (ii) the TiHo Cat Pain dataset 35 . The first dataset was collected under the ethical approvals of the Institutional Animal Research Ethical Committee of the FMVZ-UNESP-Botucatu (protocol number of 20/2008) and the University of Lincoln, (UID: CoSREC252) as per Finka et al 25 .…”

“…The TiHo dataset was collected at the Department of Small Animal Medicine and Surgery of the University of Veterinary Medicine Hannover (TiHo) 35 . Owners provided written informed consent to share data that can be used for research, regulated by the law and regulations for research in Lower Saxony (Germany).…”

“…While both approaches reached a comparable accuracy of approximately 72% in pain recognition, a significant limitation was that the study population was highly homogenous, limited to young female cats of a single breed, and evaluated using only one type of postoperative pain condition. In a follow-up study, Feighelstein et al have addressed this gap using another dataset with a more naturalistic or 'noisy' population (with variations in breed, sex, and painful conditions) 35 . The dataset was collected at the Department of Small Animal Medicine and Surgery of the University of Veterinary Medicine Hannover and included individuals of different breeds, ages, sexes, and with varying medical conditions/medical histories.…”

“…The above achievements create a yet unprecedented opportunity to explore the temporal dimensions of landmark-based approaches, moving from single frames to videos, where each individual frame can be automatically annotated. Such an approach does not require any manual labor-intensive efforts: neither in landmark annotation nor in the selection of "good" quality images, which were necessary for the AI models developed in 34,35 . The latter issue is particularly impractical for clinical applications of pain recognition models, analogous to apps like PainChek for humans ?…”

“…The current study uses the cat pain datasets previously explored in 34,35 to investigate the problem of automated landmarkbased cat pain recognition from video. The presented video-based AI pipeline reaches over 70% and 66% accuracy respectively in the two investigated datasets, outperforming previous landmark-based approaches using single frames under similar conditions, leading to the conclusion that dynamics matter in cat pain recognition.…”