AI tool could help veterinarians track heart disease more precisely

For dogs with heart disease, a fraction of a centimeter can make a life-changing difference.

Veterinarians often rely on chest radiographs to monitor subtle changes in heart size that help determine when treatment should begin. But those measurements can vary from one doctor to the next, making it difficult to know whether a dog’s condition is truly changing or whether the difference is simply the result of human interpretation.

Now, researchers at the Texas A&M College of Veterinary Medicine and Biomedical Sciences (VMBS) have shown that an artificial intelligence-powered tool can measure canine heart size on radiographs with accuracy comparable to trained clinicians, potentially helping veterinarians identify disease progression with greater consistency and confidence.

The study, recently published in PLOS ONE, evaluated RadAnalyzer, a web- and smartphone-based application that automatically measures vertebral heart size (VHS) and vertebral left atrial size (VLAS) — two metrics veterinarians use to assess heart enlargement and monitor disease progression.

The validation study represents the latest milestone for a technology that originated through Texas A&M’s Aggies Invent program, where students from a variety of disciplines collaborate to solve real-world problems. During one challenge, Texas A&M graduate student Thomas Reyes and veterinary student Tabitha Baibos-Reyes developed a prototype designed to automate heart-size measurements on canine radiographs. Since then, RadAnalyzer has evolved into a tool that could help veterinarians monitor heart disease more consistently.

Using more than 1,000 canine radiographs, researchers found that RadAnalyzer’s measurements closely matched those made by a highly trained observer, supporting its potential as a reliable clinical tool. The strong agreement between the two suggests the tool can perform measurements at a level comparable to experienced specialists.

“Most of the AI we’re talking about isn’t a black box,” said Dr. Sonya Gordon, a professor in VMBS’ Department of Small Animal Clinical Sciences and the principal investigator of the study. “It’s really a tool that helps people do their jobs better. It doesn’t replace us — it helps us.”

Measuring what matters

Myxomatous mitral valve disease, a condition characterized by a leaky valve between the left chambers of the heart, is especially common in older, small-breed dogs, many of whom develop heart murmurs long before showing signs of illness.

As the disease progresses, the heart chambers enlarge. Monitoring those changes in asymptomatic dogs helps veterinarians determine when treatment should begin and whether the disease is showing signs of progression.

Those decisions often fall to primary care veterinarians. According to Gordon, there are only about 400 veterinary cardiologists in the United States, making it impossible for specialists alone to meet the needs of every patient.

“Anything that makes the process easier for primary care veterinarians is valuable,” Gordon said. “The radiographs are done, the measurements are already there, and all you have to do is review them and make sure they look correct.”

While echocardiography remains the gold standard for evaluating the heart, it requires specialized equipment and expertise that may not be available in every veterinary clinic. Radiographs, by contrast, are more affordable and widely available, making tools that improve their accuracy and consistency especially valuable for veterinarians monitoring heart disease.

Because heart size can be difficult to assess by sight alone, veterinarians developed VHS and VLAS as standardized measurements that provide a more objective way to evaluate enlargement and track changes over time.

Those measurements help veterinarians determine when an older, small-breed dog with a heart murmur has a heart that is enlarged enough to warrant treatment with pimobendane, a medication known to delay the onset of clinical signs.

“It can extend symptom-free survival by an average of 60%, which is more than a year for most dogs,” Gordon said. “It’s all about extending health span by starting medication at the right time.”

Turning a tedious process into a simple one

Although VHS and VLAS are valuable clinical tools, obtaining them manually can be cumbersome.

Veterinarians must identify specific anatomical landmarks on a radiograph, place measurement markers, redraw those measurements over the vertebrae and calculate the final values. The process can be time-consuming and introduces opportunities for variation.

“A human would have to place multiple points on the picture and then estimate measurements to the nearest tenth of a vertebra — there are lots of places for error and variability,” Gordon said.

Gordon envisioned a simpler solution: a tool that could automatically perform the measurements while remaining accessible to veterinarians regardless of what imaging software they used.

“I wanted it to be a standalone tool, and I wanted it not to just be available on a computer — I wanted it to work with a phone,” Gordon said.

RadAnalyzer uses machine-learning algorithms trained on annotated radiographs to identify the same landmarks veterinarians use when measuring heart size. Users can simply upload or photograph a radiograph, review the automated markings and measurements and adjust if necessary.

The result is a faster process that produces the same measurements every time it evaluates the same image.

Consistency could be the real advantage

Gordon believes that one of the tool’s greatest strengths may be its consistency.

Previous research from Gordon’s group found that measurement variability decreases with experience but never disappears. AI-assisted measurements, however, remain consistent from one evaluation to the next.

“The AI tools have essentially zero measurement variability,” Gordon said. “Because they’re consistent from one measurement to the next, they may be able to identify disease progression more precisely than humans can.”

That precision becomes especially important when veterinarians are tracking small changes in heart size over months or years.

In some cases, a seemingly minor increase in heart size can indicate that a dog’s disease has progressed enough to warrant treatment. Consistent measurements may help clinicians identify those changes sooner and with greater confidence.

Building trust in veterinary AI

As artificial intelligence becomes increasingly common in veterinary medicine, Gordon believes rigorous scientific validation will be essential.

“We have to make sure we hold it to the same scientific scrutiny that we hold anything to, which means validation to document that these tools that are becoming available to us actually work,” Gordon said.

Gordon believes that the greatest value of AI lies in helping veterinarians make better-informed decisions rather than replacing human expertise.

“If people think of AI as something that’s making us obsolete, they’re looking at it the wrong way,” Gordon said. “It’s making us better.”