Texas A&M AI tool may sharpen canine heart disease tracking
Bottom line
Texas A&M researchers say their AI-powered RadAnalyzer app can automatically measure vertebral heart size and vertebral left atrial size on canine thoracic radiographs with accuracy comparable to trained clinicians, potentially giving veterinarians a more consistent way to monitor cardiac enlargement over time. In the validation study, published in PLOS ONE on May 13, 2026, the team evaluated high-quality radiographs from 1,058 client-owned dogs across 80 breeds and found strong correlation between the software and a trained observer, especially for vertebral heart size. Texas A&M has framed the tool as a way to support decision-making in common conditions such as myxomatous mitral valve disease, particularly in settings without easy access to a cardiologist. (pmc.ncbi.nlm.nih.gov)
Why it matters: For general practitioners, the main value isn’t that AI replaces clinical judgment, but that it may reduce the interobserver variability that can complicate serial radiographic follow-up. The study authors note the app is designed so clinicians can review the landmarks the model selected and adjust them if needed, which could make it more practical as a workflow aid than a black-box result. That matters in primary care, where thoracic radiographs are widely available, echocardiography is not, and small shifts in measured heart size can influence whether a dog is monitored, staged differently, or referred. The authors also note the validation was done on high-quality radiographs without pulmonary infiltrates, so performance in messier real-world cases still needs careful attention. (pmc.ncbi.nlm.nih.gov)
What to watch: Watch for broader real-world validation, comparisons with echocardiography and lower-quality films, and whether adoption grows now that RadAnalyzer is available as a commercial web and smartphone tool. (pmc.ncbi.nlm.nih.gov)
Key facts
- Tool
- RadAnalyzer
- Developer
- Texas A&M researchers
- Use
- Automatically measures vertebral heart size and vertebral left atrial size on canine thoracic radiographs
- Study type
- Validation study
- Publication
- PLOS ONE
- Publication date
- May 13, 2026
- Sample size
- 1,058 client-owned dogs
- Breed count
- 80 breeds
- Key finding
- Software showed strong correlation with a trained observer, especially for vertebral heart size
- Limitation
- Validation used high-quality radiographs without pulmonary infiltrates
Texas A&M’s RadAnalyzer is moving from student-built concept to peer-reviewed clinical tool, with new PLOS ONE data suggesting the AI application can measure canine heart size on chest radiographs about as well as trained clinicians. The study, published May 13, 2026, tested the software on more than 1,000 canine cases and positions the platform as a practical aid for tracking heart disease progression in everyday practice. (researchgate.net)
The background here matters. RadAnalyzer began as a Texas A&M student project aimed at automating vertebral heart score calculations, a task that is routine, useful, and often variable from one reader to another. Texas A&M’s earlier reporting described the tool as a response to the time and uncertainty involved in manual measurements, with Sonya Gordon and colleagues helping refine the algorithm using cases from the university’s Small Animal Teaching Hospital. That origin story helps explain the product’s current pitch: not AI as a replacement for cardiology, but AI as a way to standardize a measurement many primary care teams already use. (vetmed.tamu.edu)
In the new validation study, the investigators retrospectively analyzed high-quality right lateral thoracic radiographs from 1,058 client-owned dogs representing 80 breeds. According to the paper, RadAnalyzer showed stronger agreement for vertebral heart size than for vertebral left atrial size, with correlation coefficients of 0.917 and 0.873, respectively. The authors concluded that the web-based, smartphone-optimized application generated measurements comparable to a trained veterinarian and could reduce the effect of interobserver variability. They also emphasized an important guardrail: the study population excluded radiographs with pulmonary infiltrates, and clinicians can review or adjust the AI-selected landmarks after measurement. (pmc.ncbi.nlm.nih.gov)
That fits with the broader state of veterinary imaging AI. A 2022 review in Veterinary Radiology & Ultrasound argued that AI tools in radiology should be evaluated with the same attention to validation, bias, and intended use as other diagnostic aids, especially because image quality, case mix, and workflow integration can shape real-world performance. More recent veterinary literature has also underscored that vertebral heart size and vertebral left atrial size are useful, widely used radiographic metrics, but they are not perfect substitutes for echocardiography. In other words, RadAnalyzer appears most useful as a consistency tool inside a broader cardiac workup, not as a stand-alone staging test. (onlinelibrary.wiley.com)
Direct outside commentary on this specific paper was limited in accessible sources, but the industry reaction around the launch has been fairly consistent: the strongest use case is in practices that already take thoracic radiographs but do not have immediate specialist access. Texas A&M’s materials and RadAnalyzer’s own product information both stress accessibility through web and smartphone workflows, while the paper itself highlights the ability to inspect landmarks before acting on a result. Taken together, that suggests the commercial strategy is aimed squarely at first-opinion practice, urgent care, and settings where serial monitoring needs to be fast and repeatable. That’s an inference based on the study design and product positioning, rather than a direct claim from an independent analyst. (vetmed.tamu.edu)
Why it matters: For veterinary professionals, the real significance is operational as much as clinical. Heart disease monitoring often depends on comparing today’s radiograph with one taken months earlier, and small differences in how a measurement is placed can muddy that comparison. If an AI tool can narrow that variability, it may improve confidence in trendlines, support more consistent recommendations to pet parents, and help teams decide when a case should move from monitoring to referral or treatment adjustment. At the same time, the paper’s limits are important: this was a single-center, retrospective validation on selected high-quality films, so clinics should be cautious about overgeneralizing performance to noisy, rotated, poorly exposed, or pulmonary edema cases. (pmc.ncbi.nlm.nih.gov)
There’s also a business and regulatory angle worth noting. RadAnalyzer is already being presented as a commercially available tool, with developer-facing documentation and clinic access pages describing automated VHS and VLAS measurement through an app and API. That means this is not just an academic proof-of-concept anymore; it is entering the market at a moment when veterinary practices are increasingly evaluating AI tools for workflow support, second reads, and client communication. (radanalyzer.com)
What to watch: The next meaningful milestones will be external validation in more diverse practice settings, head-to-head comparisons with echocardiographic findings and other AI platforms, and evidence showing whether use of the tool actually changes referral timing, treatment decisions, or outcomes in dogs with suspected or confirmed cardiac disease. (pmc.ncbi.nlm.nih.gov)