Comparing the performance of deep learning video-based models and trained veterinarians in cattle pain assessment

Accurate pain assessment in animals is crucial for ensuring animal welfare and guiding veterinary interventions. Traditional pain evaluation relies on scoring of pain behaviours by veterinarians, which can be influenced by observational variability and individual expertise. There is a growing interest in using AI tools, and the question whether Artificial Intelligence (AI) can outperform humans in animal pain recognition is only beginning to be explored. This study is the first to address cattle pain recognition in this context. Namely, we compare the performance of trained veterinarians in the task of pain recognition in cattle using video-based analysis. Our results show that machine learning models achieve high accuracy in pain classification and demonstrate performance comparable to trained veterinarians, with some advantages in video-based assessments. These findings highlight the potential of machine learning to enhance pain assessment in veterinary medicine, offering a scalable and more objective tool for improving animal welfare.