Automated recognition of pain in cats

Abstract

Facial expressions in non-human animals are closely linked to their internal affective states, with the majority of empirical work focusing on facial shape changes associated with pain. However, existing tools for facial expression analysis are prone to human subjectivity and bias, and in many cases also require special expertise and training. This paper presents the first comparative study of two different paths towards automatizing pain recognition in facial images of domestic short haired cats (n = 29), captured during ovariohysterectomy at different time points corresponding to varying intensities of pain. One approach is based on convolutional neural networks (ResNet50), while the other—on machine learning models based on geometric landmarks analysis inspired by species specific Facial Action Coding Systems (i.e. catFACS). Both types of approaches reach comparable accuracy of above 72%, indicating their potential usefulness as a basis for automating cat pain detection from images.

Investigating subtle changes in facial expression to assess acute pain in Japanese macaques

“…Recently, automated recognition of pain has been shown to be accurate in horses, both discriminating the presence (88.3%) and level of pain (75.8%) 67 . Automated pain recognition was also successful in cats 68 . Computer vision and machine learning techniques hold promise for unveiling animal emotions with considerably more detail hidden from the human eye, less risk of anthropocentric biases, and the potential to outperform human evaluation 44 , 67 , 69 .…”

Explainable automated recognition of emotional states from canine facial expressions: the case of positive anticipation and frustration

“…The present study is the first to explore automated recognition of canine emotional states focusing on diverse facial expressions, whilst using a carefully designed controlled experimental setup for dataset creation and annotation. We present classifiers of two different types: deep learning based and DogFACS-based, both having a performance that is comparable to and in some cases outperforms those presented in previous studies addressing recognition of pain or emotional state from facial expressions, including mice 38 , 39 (> 89% and 93% respectively), cats 43 (> 72%), horses 42 , 46 (> 75% and 65% respectively) and sheep 55 (> 64%).…”

“…It should further be noted that as our dataset is limited to one breed, an immediate future research need is an assessment of the generalizability of the models to other breeds. If performance drops significantly when transferring the results to other breeds, alternative approaches to the deep approach used here are indicated, e.g., in Feighelstein et al 43 .…”

“…The majority of these works focus on the occurrence of pain. Species that have been addressed in this context include rodents 37 – 39 , sheep 40 , horses 33 , 41 , 42 and cats 43 . All of these works provide a binary classifier for pain/no pain, using machine learning techniques.…”

The history of pain measurement in humans and animals

“…Finally, a recent development in preclinical algesiometry, especially at later stages of preclinical research, is to consider assessing new treatments via their effect on clinical pain states (e.g., arthritis, cancer pain) in companion animals (see 94 , 95 ). Combined with more valid endpoints, such as automated measurement of grimacing in cats ( 96 ), this might represent a powerful way to predict clinically efficacy in human trials.…”