Artificial intelligence is being applied to a stubborn and costly problem in beef production, through tools that will help detect disease and other health issues in cattle to reduce losses.
At the University of Wyoming, a doctoral student has created a computer vision model to analyze cattle hearts for signs linked to congestive heart failure. At the University of Arkansas, another team has built a system that uses thermal imaging and artificial intelligence to estimate a calf’s temperature and flag fever without the stress of manual checks.
The economic stakes are substantial. Cattle production is the largest sector of U.S. agriculture by cash receipts- in 2024 it accounted for about 22 percent of the $515 billion in total cash receipts for agricultural commodities. The U.S. also has the world’s largest fed-cattle industry and is the world’s largest total consumer of beef, particularly high-value, grain-fed beef.
Within that vast industry, even small improvements in detecting illness earlier can carry real financial benefits for producers. That is especially true for bovine congestive heart failure, or BCHF, a fatal condition that has drawn growing alarm in the cattle industry in recent years, particularly in the Western Great Plains.
Warnings about the problem were sounding well before the newest AI tools emerged. In 2018, researchers, veterinarians, cattle producers and other industry stakeholders gathered in Clay Center, Neb., for the Bovine Pulmonary Hypertension and Congestive Heart Failure Collaborators’ Workshop, a meeting that laid out the seriousness of the disease and the urgent need for better tools to identify it.
The workshop’s executive summary described BCHF as “an untreatable, fatal condition” and said that for some producers it had become “the single most costly health-related problem,” with losses “exceeding $250,000 annually in individual operations, even surpassing those from bovine respiratory disease.” Participants identified among the most pressing needs “diagnostic tests that allow rapid, early, and affordable identification of diseased individuals.”
Gary Darnall, proprietor of Darnall Ranch Inc. and Darnall Feedlot, wrote that over a five-year span his operation had seen 518 head of BCHF, including dead animals and realizers, amounting to “a financial loss of $943,553.00.”
Pete McClymont, executive vice president and treasurer of the Great Plains Veterinary Educational Center, wrote that BCHF was “increasingly recognized as an emerging condition of cattle in the Western Great Plains of the United States and Canada” and that it had “a significant impact on the economic viability of the operations of our members.”
Significant research was sparked by that workshop.
Chase Markel, a doctoral candidate in animal science at the University of Wyoming, has developed what the university describes as a first-of-its-kind AI model trained to predict congestive heart failure risk based on images of a cow’s heart. Markel, who grew up in Wheatland, Wyo., in an agricultural community, came to artificial intelligence not through computer science, but through livestock research.
“I’m not a computer scientist. I’m not an AI guy,” Markel said in materials published by the university. “I’m someone who is studying heart failure (in cattle) and just happened to have the right conversation and made the connection to build something that I think can be useful.”
His earlier graduate work focused on pulmonary hypertension, also known as brisket disease or high-altitude disease, which has been linked to congestive heart failure. As pressure builds in the right ventricle, the heart can become thickened and misshapen, creating visible indicators of risk. Markel’s idea was to see whether artificial intelligence could be trained to recognize those patterns in a systematic way.
To build the model, a team of researchers assembled nearly 7,000 heart images from commercial processing plants in Nebraska and Colorado. Each image was scored by hand on a one-to-five scale, and those rankings were used to train the system. According to the university, the model now assigns the correct score 92 percent of the time when shown an image it has not seen before.
“Anything we can do to improve traceability and individual animal identification back as far as we can go in the production cycle to try to prevent these things is a net benefit for the industry,” Markel said. He is also developing a similar model to examine liver images for abscesses, another common problem in feedlot cattle.
At the University of Arkansas, researchers are developing a tool to detect fever in cattle without physically handling each animal.
Their tool, called CattleFever, combines thermal imaging, RGB color cameras and AI to estimate body temperature. The project was led by Trong Thang Pham, a doctoral student in the university’s Artificial Intelligence and Computer Vision Lab, which is directed by Ngan Le, an associate professor of electrical engineering and computer science.
Today, cattle temperatures are typically taken rectally, a method that is labor intensive, time consuming and stressful for the animals. The Arkansas team’s goal is to create a system that can detect fever remotely, offering ranchers a less intrusive way to monitor herd health.
To do that, the researchers had to build their own dataset. They recorded short videos and thermal images of thousands of calves at the Arkansas Agricultural Experiment Station’s Savoy Research Complex, while also taking rectal temperatures to establish a benchmark. The team marked 13 facial landmarks on the images, including the eyes, ears, muzzle and mouth, then used those annotations to train a model that could automatically identify the same features across thousands more frames.
Their research found that temperatures near the eyes and nostrils most closely matched readings from a thermometer. The resulting system was able to estimate an animal’s temperature within 1 degree of the rectal reading, according to the university.
Fever is often one of the earliest signs that an animal is sick. In large herds, where individual illness can be difficult to detect quickly, a camera-based system could help ranchers spot problems sooner, respond earlier and reduce unnecessary handling.
Pham noted that the current system was developed using images of calves facing the camera and that the next challenge is to gather more data in real-world settings, with animals moving naturally and appearing from different angles.
“We probably need to take more photos of them in the real-world settings, such as running around, to capture their motion in the field,” Pham said.
