The study was led by Ruby Chang, Associate Professor of Statistics at the RVC, and carried out by Dr Hanya Ahmed.
The team compiled a databank of images, including 100 equine fracture cases from two UK equine hospitals and published literature, alongside 70 feline cases from hospital databases and around 4,000 human fracture images from a public database.
Using these images, the researchers built a three-stage AI system that first identifies the scan type, then recognises the image angle, before detecting and precisely localising fractures.
The system used transfer learning, enabling it to be trained on the large human dataset before being adapted for equine cases.
Using this method, the system achieved a reported fracture localisation accuracy of between 71 and 84% without requiring an unrealistically large number of equine images.
The RVC said the findings demonstrate the potential for AI-assisted tools to strengthen fracture diagnosis across veterinary practice.
It said faster and more reliable detection could reduce uncertainty in clinical decision-making and enable earlier treatment for racehorses and companion animals.
Building on the work, the team has expanded its collaboration with the Hong Kong Jockey Club to investigate whether AI can identify early bone changes before fracture occurs.
The study has been shortlisted for the STEM for Britain 2026 award and was funded by the Horserace Betting Levy Board.
Reference
PS: Whilst you're here, take a moment to see our latest job opportunities for vets.