AI-Assisted Equine Research
How open-sourced artificial intelligence (AI) programs are helping to analyze high-definition video of animals.
In her work as a geneticist and associate professor of equine physiology at the University of Florida, Dr. Samantha Brooks uses the genome sequence and genetic technologies to better understand traits that are important to the health and well-being of horses and other large animals. One of the most important traits across a wide array of large animals is soundness, she said.
“When we talk about horses, we want them to be “sound and sane,” she said. “We want them to be safe to work around and mentally able to do their jobs. That’s true for most livestock too. We want them to be safe to work around and have good mental well-being as well as physical well-being.”
The challenge for researchers and veterinarians is that those types of traits are hard to measure. Eight years ago, UF scientists began efforts to better assess equine and livestock mobility. Graduate students interested in locomotor traits started by looking for anything unusual in an animal’s movement; polymorphisms and differences in the way of moving for the animals.
They discovered almost immediately that there were no tools available to do that from a scientific perspective – nothing was objective or repeatable. While there were a few systems on the market that might use things like instrument packages, motion sensors or GPS units, none of them met the scale that the UF researchers needed.
“So we started investigating new methods for doing digital video analysis, because it’s inexpensive to capture, it’s technology that most people are familiar with, and it is easily scalable, so we could do it in a large numbers of animals,” said Dr. Brooks. “As we worked in that area, though, it became apparent that there were applications for this type of tool beyond just the scientific experiments that we wanted to do, chief among those the detection of changes within an animal, rather than across animals, and that often includes body postures that indicate pain. There’s a real need for improved technologies in this area, because the gold standard scientific and clinical tool right now is the human eye, and the human eye has many limitations that make it less than ideal for this process.”
Enter artificial intelligence.
Through a university partnership with NVIDIA and research tools freely available through open-source software, UF researchers were able to analyze high-definition video of the animals much quicker and more in-depth. For example, an undergraduate researcher used AI-enabled tools to measure gate parameters in eventing horses; horses who were competing in three-day eventing. Prior to the competition, they examined measures like stride length, stance time and limb extension. Then, right before they went into the final phase of competition – show jumping – she observed them again.
What was discovered through video analysis was that in the morning before the horses competed in the show jumping phase, movements could be discerned of those horses most likely to be successful in show jumping and leave the ring with no penalties, versus those who were more likely to incur penalties in that phase. “We saw things like a slightly slower forward motion, which also correlates with a slightly shorter stride length and the length of time that that foot stayed on the ground was slightly longer in those who were not going to be successful or perfectly successful in show jumping,” Dr. Brooks said. “That’s exciting to see differences in their way of going hours before their later performance that that were statistically associated with how they performed.”
Possibilities and parameters
Dr. Brooks is hopeful the research project could lead to a low cost, open approach to better evaluation of animal mobility. For example, if an owner feels like they’re observing changes in their horse’s way of moving, they can take a video and get an analysis that might confirm the horse is walking a little different in the left hind, then forward that video and analysis to a veterinarian. The clinician can then observe the video remotely and decide whether to get involved in the care for the horse much earlier on. “Maybe we can begin to intervene and prevent injuries or apply an appropriate treatment before things have a chance to get irrevocably damaged,” she said.
In this project, AI is being used to recognize important anatomical landmarks on the picture of the horse. Each video is just a collection of individual frames, and the computer must figure out foundational principles such as what is a leg so that it can help track that leg throughout the video, making our job a little easier.
“My master student did a project labeling some over 70,000 or so frames of data, which would have taken her eight to nine months to do by hand, but we could do it within a few hours with the well-trained model,” Dr. Brooks said. “It gave us the ability to do things much, much faster. I think we’ll see this benefit from AI technology impacting many different aspects of research in the coming years.”
But one note of caution that Dr. Brooks would give to veterinarians is that applying an AI label to something does not magically make it “techier, more resilient or more effective,” she said.
“We focus a lot here at UF in making sure that these new AI methods are applied appropriately and applied ethically, because they’re going to be trained on data sets, and we need to know that the data set truly reflects the population, and that we are basing our interpretations of those data in a context that is appropriate for the utilization that we’re trying to create. We don’t want to assume, because it came out of fancy AI software that it is going to be exactly right and exactly right for this situation.”
Photo caption: Dr. Samantha Brooks, University of Florida
Photo credits:
istockphoto.com/Roman Stasiuk
Courtesy of Samantha Brooks, UF/Institute of Food and Agricultural Sciences
