Fractures are a major concern in Thoroughbred racehorses. These injuries can affect both a horse’s welfare and its racing career, and in severe cases they can be life-threatening. Studies estimate that around 10% of racehorses experience a fracture during training, while bone injuries occur in roughly 1.3 per 1,000 starts in flat racing. Because of this risk, early and accurate detection of bone damage is extremely important.
However, diagnosing fractures is not always straightforward. Veterinarians usually rely on radiographs (X-rays) to assess suspected bone injuries. While X-rays are very useful, identifying fractures on these images can be challenging. Small cracks or subtle changes in bone structure may be difficult to see, and image quality or the angle at which the X-ray is taken can also affect interpretation. As a result, there is growing interest in using technology such as AI to support clinical decision-making.
The RVC research team developed an AI system designed to analyse medical images and identify fractures. The study was led by Associate Professor of Statistics Dr Ruby Chang, with the research carried out by Dr Hanya Ahmed. To train the system, the researchers created a large database of images that included 100 equine fracture cases collected from two UK equine hospitals and from published studies. They also included 70 feline fracture cases and around 4,000 human fracture images from a public database.
The AI system works in three stages. First, it identifies the type of medical image being analysed, such as an X-ray, CT scan, or MRI scan. Next, it determines the angle or projection of the image. Finally, it analyses the image to detect whether a fracture is present and to pinpoint its exact location.
One particularly interesting feature of the study is the use of a technique called transfer learning. In transfer learning, an AI model is first trained on a large dataset (in this case, thousands of human fracture images). The knowledge it gains is then adapted to a smaller dataset from another field; in this case, veterinary medicine. Because there are far fewer veterinary medical images available for training, this approach helps overcome one of the main challenges of developing AI systems for animal healthcare.
Using this method, the AI system was able to detect and locate fractures in horses with accuracy levels between 71% and 84%, despite having a relatively small number of equine images to learn from. The system also achieved very high accuracy when identifying image types and projections, reaching more than 96% accuracy in some stages of the analysis.
The results suggest that AI could become a valuable support tool for veterinarians. By helping identify fractures more quickly and reliably, AI-assisted systems may reduce uncertainty in diagnosis and allow treatment to begin earlier. This could improve recovery outcomes for horses and other animals.
The research team is now expanding the project through collaboration with the Hong Kong Jockey Club. The next goal is to investigate whether AI can detect early bone changes before a fracture occurs. If successful, this could help prevent serious injuries in racehorses and improve welfare within the sport.
Although the current study focuses on horses, the approach could also be adapted for other species such as cats, dogs, and potentially humans. Overall, the research demonstrates how advances in AI and medical imaging could play an increasingly important role in the future of veterinary diagnostics.
For more details, see:
Ahmed, Hanya T., Dagmar Berner, Qianni Zhang, Kristien Verheyen, Francisco Llabres-Diaz, Vanessa G. Peter, and Yu-Mei Chang. 2026.
Bridging Species with AI: A Cross-Species Deep Learning Model for Fracture Detection and Beyond
Bioengineering 13, no. 2: 213.
https://doi.org/10.3390/bioengineering13020213
No comments:
Post a Comment