AI tools show promise for veterinary diagnosis

AI tools can support vets in diagnosing disease, research has shown.

A computer model has been trained to recognise a common disease in dogs with an encouraging success rate, underscoring its potential as a veterinary tool.

The approach supports the use of AI for relatively straightforward tasks, to allow vets to concentrate on more complex tasks, according to the team behind the study. 

Scan images

Researchers from the Royal (Dick) School of Veterinary Studies, the Roslin Institute and the University of Edinburgh’s School of Informatics developed their model using CT scan images of dogs’ middle ears – a part of the ear that is frequently affected by disease.

The model was trained to recognise disease using about 500 images, some that showed signs of disease and others that did not, which had been manually interpreted by veterinary experts. This number is fewer than the several thousands that would typically be needed for many AI studies, which would be resource-intensive to generate.

The AI tool was able to make correct diagnoses in 85 per cent of cases. Experts say the result is robust for a relatively small sample size, demonstrating the validity of the approach, and could be improved if additional images were annotated.

Diagnosis potential

The application of AI in this way, which is established in human medicine, could be made to various veterinary medicine challenges, potentially including those that are traditionally difficult to diagnose by eye. This would save time and money in making diagnoses, the team says.

A bank of veterinary medical images, annotated by vets, would be required to enable adoption of AI in veterinary diagnoses, and could be applied across a range of conditions, the team suggests. Vet schools and hospitals such as the R(D)SVS have banks of images, but a sufficient number have to be annotated to train algorithms.

The study was conducted by a masters degree student in the School of Informatics.