Ophthalomolgy

AI Comes to the Stables

By Adam Marcus

Artificial intelligence has come to the stables. Researchers in Germany have created an AI model for detecting eye disease in horses that appears to significantly outperform veterinarians.

In a recent study, published in Equine Veterinary Journal, the algorithm correctly identified eyes with equine uveitis and other ocular ailments in 93% of cases. Equine veterinarians answered correctly about three-fourths of the time (76%), whereas other veterinarians reached the right diagnosis for roughly two-thirds of the images (67%).

For the study, Anna May, DVM, of the Equine Clinic at Ludwig Maximilians University, in Oberschleissheim, and her colleagues wrote the algorithm and trained it on 2,300 photographs of horse eyes, which they modified slightly into nearly 9,400 pictures to enlarge the data set for the model.

They then tested the algorithm on 40 images of horse eyes with several ophthalmic disorders, including 12 with uveitis, 18 with other conditions, and 10 healthy eyes.

They also provided the same pictures to a group of 148 vets, including equine and small animal specialists, as well as generalists, and asked for their diagnoses for each.

“We showed 40 pictures which were not known to the AI algorithm to the software tool and 148 vets,” Dr. May said.

“The model was especially good in detecting uveitis because it had been trained on multiple forms of that disease,” she said. “Other conditions that were represented less frequently in the data set or where clinical findings were not that clearly visible were harder to detect for the AI.”

Dr. May said the success of the AI tool wasn’t surprising, since a previous study showed it to be about 96% accurate at identifying eye diseases. “Nevertheless, the AI cannot replace the clinician as other things—including medical history and any other medical conditions—have to be taken into account and the AI is not able to examine all parts of the eye. It can therefore only serve as an additional tool helping unexperienced veterinarians or as a guidance on whether a condition should be treated as an emergency when there is no vet available.”

Since this study, the researchers have trained their model on many more ophthalmic diseases, “and we will definitely test its performance again,” Dr. May said. MeV

For more information:
Scharre S, et al. Comparison of veterinarians and a deep learning tool in the diagnosis of equine ophthalmic diseases. Equine Vet J. 2025;57:47–53. https://beva.onlinelibrary.wiley.com/doi/10.1111/evj.14087