Unable to speak about their pain, sheep can suffer in silence for long periods of time before their pain is detected by humans. Researchers at the University of Cambridge have recently demonstrated that an artificial intelligence (AI) system can determine whether sheep are in pain.
Essential to maintaining the welfare of animals is assessing their pain levels, a task that can be a very difficult and time-consuming process. By detecting pain in animals, diagnosis of diseases and subsequent treatment to alleviate pain can occur. In sheep, pain typically indicates they are suffering from diseases such as foot rot (an infectious disease which as the name suggests, rots away the foot) and mastitis (inflammation of the breast tissue caused by an infection). Recent research suggests that early detection of these diseases can be achieved through efficient AI assessment tools.
The Sheep Pain Facial Expression Scale (SPFES), a standardised measure to assess the pain level in sheep using their facial expressions, was recently introduced. It has been shown to assess face pain in sheep with high accuracy. However, training people to become SPFES scorers is time consuming and a limitation of this tool is the possibility of individual bias that results in inconsistent scoring.
Robinson and colleagues at the University of Cambridge devised a new technique with an AI system to analyse sheep facial expressions in the hope of improving efficiency and consistency in pain estimations. They introduced a preliminary classification system based upon the SPFES and adapted computer vision techniques commonly used in human emotion recognition to estimate pain levels.
The study considered frontal face measurement using a multi-level approach to automatically assess the pain level in sheep. Using the classification system which included the ears, nose and eyes as key features to evaluate the pain level, a total of 480 images of sheep were analysed. Of the 480 images, the data sets were divided into two subsets, with 380 sheep from farm and 100 images of sheep from the internet.
Overall the study showed 67% accuracy, which is comparable to the accuracy of human assessments. The experimental evaluation confirmed that the presented taxonomy was reasonable and able to successfully assess the pain levels of sheep, with ears giving the most noticeable cues. By refining the procedure, Robinson and colleagues were able to increase accuracy.
Introducing this AI system could mean earlier diagnosis of diseases, leading to quicker treatment and less suffering for sheep. It is expected that future investigations will focus on the generalizability of this AI system to assess pain in other animals.
Written By: Lacey Hizartzidis, PhD