A cloudy eye can be an early sign of infectious bovine keratoconjunctivitis (IBK), commonly known as pinkeye. Spotting affected cattle early can help producers and veterinarians respond, but keeping an eye on every animal can be difficult when cattle are spread across large pastures.
Could artificial intelligence (AI) help? Researchers are exploring two approaches: analyzing cattle muzzle images to identify animals that may be developing the disease and examining eye images for visible signs of infection. Both could eventually help identify cattle that need a closer look.
Using Muzzle Images to Identify Emerging Cases
A 2023 study investigated whether AI could flag cattle developing IBK before veterinarians noticed clinical signs.
Researchers developed a system that analyzes cattle muzzle images for patterns associated with emerging disease. Interestingly, they could not identify a specific muzzle change that explained the system’s predictions, even though the AI successfully flagged animals before veterinarians identified the disease.
The team tested the system on 870 cattle across four locations in Kansas and Missouri. Veterinarians subsequently diagnosed 170 animals with IBK. Images were collected using different camera setups. In 2021, they photographed cattle held in a squeeze chute using cell phones or digital cameras. The following year, they used a GoPro mounted on a feeder truck.
The AI had flagged 169 of those animals before veterinarians identified the disease. It also flagged 17 animals that did not develop IBK during the study. This method had 99.4% sensitivity and 97.6% specificity suggesting that muzzle-image analysis could offer an early warning of disease, potentially giving producers more time to respond.
The researchers emphasize the potential value of acting on early warnings. Producers could respond by strengthening fly control, providing more shade or separating suspect animals when practical. However, because the team could not identify a specific muzzle change associated with emerging IBK, further research is needed to understand what the AI is detecting and how consistently it can identify animals at risk.
Recognizing Pinkeye From Eye Images
A 2026 study takes a more direct approach: using AI to examine cattle eye images and classify them as healthy or infected.
Researchers compared two AI models, YOLOv11 and YOLOv26, using 408 eye images collected from yearling Angus beef steers at South Dakota State University’s Cottonwood Field Station during the summers of 2023 and 2024.
Of those images, 323 were classified as healthy and 85 as infected. Researchers based their classifications on visible signs such as inflammation, discharge and corneal cloudiness.
The researchers faced one challenge: they had far more images of healthy eyes than infected ones. To help the AI recognize pinkeye, they tested two ways to give infected images more weight during training. One involved using the same infected images more than once, while the other created new versions by adjusting things like image angle, brightness and contrast.
Several versions of the AI models achieved 99% accuracy, and the best-performing versions correctly identified all 17 infected images in their test sets. The researchers also found that how the models were trained influenced their ability to detect pinkeye. Reusing existing infected images generally improved detection more than creating modified versions.
The results suggest that AI can recognize visible signs of pinkeye in photographs. However, the study did not establish whether the models could detect disease before clinical signs appeared.
What Comes Next for AI Pinkeye Detection?
Both studies show potential, but more work is needed before these tools can be used reliably on farms. The muzzle-based system needs further evaluation across cattle populations, including dairy cattle. The eye-image models need testing with larger datasets to determine whether their results hold up beyond the original study.
Researchers also need to establish how well these systems perform under everyday conditions, where lighting, camera angles and animal movement can affect image quality. For the eye-image approach, further work is needed to determine whether the models can distinguish IBK from other eye conditions and recognize cases at different stages of disease.
If these approaches prove reliable in the field, they could give producers and veterinarians another way to identify cattle that need closer examination.
