Giving a Voice to Those Without Words: The AI That Reads Horses' Pain

Researchers at Tel-Hai College have developed an AI system that detects pain in horses and explains where it is located. The method, published in the International Journal of Computer Vision, uses a fixed 3D model of the horse's face to stabilize explanation maps. The model achieved high accuracy on three datasets, and its alignment with veterinarians' assessments was statistically significant. The development may also be used in human medicine, for example to assess pain in infants or dementia patients.

Researchers at Tel-Hai College in Kiryat Shmona have developed a groundbreaking AI system that detects suffering in horses and explains exactly where they hurt. The study, led by Dr. Marcelo Feigelstein, was published in the scientific journal International Journal of Computer Vision. The system is based on previous tools developed by the team for "reading" facial expressions and body language of animals, and now horses have been added to the list. The main innovation is a method called SHIC-XE, which projects the model's attention onto a fixed 3D model of the horse's face. This provides a consistent and anatomical explanation even when the camera angle or head position changes. The model was tested on three datasets: post-surgical pain, inflammatory-orthopedic pain, and acute mechanical pain, and achieved high accuracy metrics. A statistically significant correlation was found between the model's focus areas and expert veterinarians' assessments, especially in the ears and cheek muscles. The implications extend beyond the animal world: the need to understand what an AI system "saw" in video also exists in human medicine, for example in assessing pain in infants, dementia patients, or anesthetized patients. The method may be used in the future for clinical video analysis. The study also involved Prof. Anna Zamansky, Prof. Ilan Shimshoni, students from the Technion, and partners from universities in Bern, Milan, São Paulo, and Newcastle.