Objective The objective of this study was to investigate the existence of an equine pain face and to describe this in detail. Study design Semi-randomized, controlled, crossover trial. Animals Six adult horses. Methods Pain was induced with two noxious stimuli, a tourniquet on the antebrachium and topical application of capsaicin. All horses participated in two control trials and received both noxious stimuli twice, once with and once without an observer present. During all sessions their pain state was scored. The horses were filmed and the close-up video recordings of the faces were analysed for alterations in behaviour and facial expressions. Still images from the trials were evaluated for the presence of each of the specific pain face features identified from the video analysis. Results Both noxious challenges were effective in producing a pain response resulting in significantly increased pain scores. Alterations in facial expressions were observed in all horses during all noxious stimulations. The number of pain face features present on the still images from the noxious challenges were significantly higher than for the control trial (p = 0.0001). Facial expressions representative for control and pain trials were condensed into explanatory illustrations. During pain sessions with an observer present, the horses increased their contact-seeking behavior. Conclusions and clinical relevance An equine pain face comprising ‘low’ and/or ‘asymmetrical’ ears, an angled appearance of the eyes, a withdrawn and/or tense stare, mediolaterally dilated nostrils and tension of the lips, chin and certain facial muscles can be recognized in horses during induced acute pain. This description of an equine pain face may be useful for improving tools for pain recognition in horses with mild to moderate pain.

Pain compromises the welfare of animals. A prerequisite for being able to alleviate pain is that we are able to recognize it. Potential behavioural signs of pain were investigated for dairy cattle with the aim of constructing a pain scale for use under production conditions. Forty-three cows were selected and fifteen different behaviours were scored, subsequently a clinical examination was performed to allocate the cows to a pain and non-pain group. The animals were then treated with an analgesic or a placebo and after a resting period the cows were re-scored by two observers blinded to the treatment. Six behaviours were found to be significantly different between the pain and non-pain group and robust enough to be included in the pain scale: ‘attention towards the surroundings’ ‘head position’, ‘ears position’, ‘facial expressions’, ‘response to approach’ and ‘back position’ (a seventh, piloerection, was also significant but seemed difficult to use as it changed rapidly; p<0.05 for all measures). The Cow Pain Scale is the sum of the score for the aforementioned behaviours. For each individual animal before and after treatment, it was significantly lower after analgesic treatment (p=0.003) in the ClinPain group but not after placebo treatment (p=0.06); the pain score did not differ significantly before compared to after treatment with analgesic or placebo for the non-pain group (p=0.2; p=0.1). A second study was conducted to further validate the Cow Pain Scale. Cows from two herds were randomly selected (n=119) and their behaviour scored by two observers. Subsequently the cows were clinically examined and allocated to a pain and non-pain group (n=96, 23 cows were excluded because of incomplete examination). The cows from the pain group scored higher on The Cow Pain Scale compared to the non-pain group for both observer I (p<0.0001) and observer II (p=0.0001). For the two observers the sensitivity of the Cow Pain Scale was calculated to 0.61/0.75 and the specificity to 0.75/0.75 with a weighted Kappa of 0.62. In conclusion the Cow Pain Scale has the potential to be applied for the assessment of pain in dairy cattle under production conditions

Objective—To compare the effects of intra-articular (IA) versus IV administration of morphine on local and systemic inflammatory responses in horses with experimentally induced acute synovitis. Animals—8 horses. Procedures—Each horse received the following 2 treatments 4 hours after synovitis was induced: IA administration of morphine (0.05 mg/kg) with IV administration of 1 mL of saline (0.9% NaCl) solution/100 kg, and IA administration of 1 mL of saline solution/100 kg with IV administration of morphine (0.05 mg/kg). Treatments were administered in randomized order with a washout period of 3 weeks between treatments. Before each treatment, aseptic synovitis was induced by injection of lipopolysaccharide into a radiocarpal joint. For the second treatment, the contralateral radiocarpal joint was selected. Joint swelling and skin temperature over the treated joints were recorded. Clinical examinations were performed, and blood WBC count, serum amyloid A (SAA) concentration, serum cortisol concentration, synovial fluid WBC count, synovial fluid total protein (TP) concentration, and synovial fluid SAA concentration were measured before and repeatedly during each of the two 168-hour study periods. Data were analyzed by use of ANOVA with repeated measures. Results—IA administration of morphine resulted in significantly less joint swelling and lower synovial fluid TP and serum and synovial fluid SAA concentrations, and blood WBC count than did IV administration of morphine. Conclusions and Clinical Relevance—IA administration of morphine exerted anti-inflammatory properties in horses with experimentally induced acute synovitis, supporting its use as a part of a balanced analgesic protocol.

A prerequisite to successfully alleviate pain in animals is to recognize it, which is a great challenge in non-verbal species. Furthermore, prey animals such as horses tend to hide their pain. In this study, we propose a deep recurrent two-stream architecture for the task of distinguishing pain from non-pain in videos of horses. Different models are evaluated on a unique dataset showing horses under controlled trials with moderate pain induction, which has been presented in earlier work. Sequential models are experimentally compared to single-frame models, showing the importance of the temporal dimension of the data, and are benchmarked against a veterinary expert classification of the data. We additionally perform baseline comparisons with generalized versions of state-of-the-art human pain recognition methods. While equine pain detection in machine learning is a novel field, our results surpass veterinary expert performance and outperform pain detection results reported for other larger non-human species.