Deep Learning Tool May Help Veterinarians Diagnose Equine Ophthalmic Disease
by Jenny Alonge
A recent study found that a deep learning, artificial intelligence (AI) software tool may aid veterinarians in diagnosing certain equine ophthalmic diseases, such as uveitis. The equine practitioner could determine if the horse needs emergency or specialized care to help save the affected eye. The study was published in the Equine Veterinary Journal.
The deep learning tool
Deep learning AI has been used extensively over the past few years as human and veterinary medicine assistant tools, revolutionizing how medical professionals approach data analysis and decision-making. By leveraging neural networks that mimic the human brain’s structure and function, deep learning enables machines to process vast amounts of data and make intelligent decisions. Deep learning involves training artificial neural networks to recognize patterns, understand complex inputs, and make predictions. Unlike traditional machine learning models that require manual feature extraction, deep learning models automatically learn features from the data. Therefore, deep learning is particularly powerful for tasks involving large and complex datasets, such as image recognition.
Deep learning algorithms are used diagnostically in human ophthalmology for conditions such as retinal pathologies, macular degeneration, and glaucomatous optic neuropathy. Systemic reviews demonstrated that these tools have equivalent sensitivity and specificity to health care professionals.
This study's deep learning AI tool used Convolutional Neural Networks (CNN) and was trained to recognize patterns and diagnose equine ophthalmic conditions using photographs of healthy equine eyes, uveitis, and other ophthalmic diseases. In total, the study used 2,346 training images, which was expanded to 9,384 images using augmentation. The AI tool demonstrated an accuracy of 99.82% in the training data.
Equine ophthalmic photographs
Horses included in the study were examined by a board-certified medicine specialist and a veterinarian with extensive equine ophthalmology experience. The horses’ pupils were dilated with tropicamide, and their eyes were examined by direct and indirect ophthalmoscopy, slit lamp biomicroscopy, and tonometry. Photographs were taken from various angles, and only images demonstrating significant ophthalmic findings were included in the study. Images used included 10 photographs of healthy eyes, 12 of uveitic eyes, and 18 of other ophthalmic diseases, such as various keratitis types, corneal ulcers, and glaucoma.
To meet the inclusion criteria, photographs of the uveitic eyes had to exhibit typical findings of inner eye involvement, such as anterior chamber fibrin or flare, miosis, inflammatory deposits, pupil irregularities, turbid greenish to yellow fundic reflex, and synechia.
Veterinary participants
The survey was sent to private practices and universities in Germany and other European countries. In total, 237 veterinarians returned the survey, but only 148 respondents completed the questionnaire, which was required for accurate statistical analysis.
The first five questions asked about the veterinarian’s field of practice, professional experience, and professional titles, such as veterinary specialist or diplomate. The next 40 questions asked the veterinarian to evaluate a photo of a horse’s eye and choose from one of three possible diagnoses for each image. The three choices were “healthy eye,” “uveitis,” or “other eye disease.”
The participants included 59% equine veterinarians, 20% mixed practice veterinarians, 18% small animal practitioners, and 3% poultry or ruminant veterinarians.
Results
The deep learning AI tool demonstrated a 93% probability for the correct answer. The misdiagnosed photographs included a keratitis eye that the tool classified as healthy, a healthy eye that the tool diagnosed as “other,” and two uveitic eyes also falsely categorized as “other.”
Equine veterinarians correctly diagnosed 76% of the photographs, while the non-equine veterinarians correctly diagnosed 67% of the images.
While these differences aren’t statistically significant, they demonstrate that the deep learning AI tool is at least equivalent to veterinarians in assessing ophthalmic diseases in photographs and may help veterinarians determine if a patient needs specialized care.
The deep learning tool is not meant to replace veterinary expertise. However, the tool can serve in addition to a full ophthalmic examination and can be especially useful in areas with little equine veterinary coverage for a second-opinion diagnostic tool.
Comments
Related Articles
More news
Copyright © 2025 - All Rights Reserved
ISSN 2768-198X
List
Add
Please enter a comment