AI in veterinary radiology: why the future depends on radiologists.

Technology can support expertise; it does not replace it.

Veterinary radiologists bring clinical judgment, contextual understanding, and years of specialized training to every interpretation. While machine learning can identify patterns across large datasets, it cannot replace the nuanced decision-making that experienced specialists apply every day.

That's why our approach to AI development starts with radiologists and stays grounded in their expertise. We build AI and machine learning tools alongside practicing specialists, with the goal of reducing repetitive tasks and helping radiologists spend more of their time where it matters most: delivering high-quality patient care.

We also see technology as an important way to expand access to specialist expertise. Every day, veterinarians make decisions about whether advanced diagnostic interpretation is feasible for their practice and their clients. Over time, responsible applications of AI have the potential to improve efficiency and help make specialty radiology services more accessible. 

But efficiency alone is not the goal. Increased productivity has value only if diagnostic quality is maintained or improved. That’s why radiologists remain deeply involved throughout the development, evaluation, and review process. For us, success isn’t measured by how much work technology can automate. It’s measured by how effectively technology helps radiologists extend more expert care to patients without compromising quality. 
 

Why veterinary imaging deserves veterinary-specific solutions.

Another principle shaping our approach is that veterinary medicine is not simply a smaller version of human medicine. Research evaluating state-of-the-art human radiology foundation models on veterinary imaging data found that models trained for human radiology did not reliably transfer to canine thoracic radiographs, highlighting a significant gap between human and veterinary imaging domains.¹ The findings reinforce the importance of veterinary-specific datasets, veterinary-specific expertise, and veterinary-focused model development. 

For us, this underscores an important point: Meaningful innovation in veterinary radiology must be built for veterinary medicine. AI is most effective when it is built with deep knowledge of the profession it is meant to serve. 

Our approach draws on large collections of veterinary data alongside the expertise of board-certified radiologists who contribute to everything from establishing reference standards and reviewing cases to evaluating model performance and guiding future development. The result is a development process grounded in veterinary medicine itself, informed by IDEXX’s 40+ years of proprietary research and board-certified specialist expertise. 

Veterinary specialists understand the unique clinical presentations, imaging characteristics, and diagnostic challenges of animal patients. Their expertise is not an optional input into development. It is foundational to the process.
 

Expert review matters.

Published research continues to demonstrate that even among board-certified veterinary radiologists, interpretation can involve meaningful interobserver variability.² This reality does not diminish the value of specialist expertise. Rather, it highlights why expert review, collaboration, and consensus remain so important in advancing both diagnostic quality and responsible AI development.

The strongest AI systems are not those that attempt to eliminate expert judgment. They are those that learn from it, incorporate it, and continually improve through partnership with the specialists who use them.
 

Responsible AI requires scientific rigor.

Many imaging findings exist along a spectrum of certainty. Experienced radiologists may disagree, not because someone is right or wrong, but because medicine often involves ambiguous signals and complex clinical contexts.
This reality has important implications for how AI systems are developed and evaluated. Before an AI model can be assessed, researchers must establish a “ground truth” reference, typically by having one or more radiologists independently interpret and label imaging studies. The quality of those expert labels directly influences how accurately an AI system’s performance can be measured. 

Recent research conducted by IDEXX demonstrated that the number of radiologists involved in creating that reference standard matters. AI validation methods relying on a single reader or even dual-reader approaches can significantly underestimate true model performance compared with triple-read consensus methods.³ In some cases, key performance metrics differed by 10 to 30 percent depending on the reference standard used. These findings highlight how critical rigorous labeling and validation practices are when assessing imaging AI. 

These findings reinforce a principle that guides our broader approach to AI development: The quality of an AI system depends on the quality of the clinical expertise behind it. Creating reliable veterinary AI requires more than large amounts of data. It requires carefully curated datasets, rigorous clinical oversight, robust reference standards, and ongoing partnership with board-certified specialists.

The broader lesson extends beyond any individual model. As one IDEXX research team concluded, "The current state of radiological diagnoses and AI calls for honesty and transparency when assessing prediction quality in the face of ambiguous signal strength." 

We believe that honesty and transparency are foundational principles for veterinary AI. Responsible development demands more than impressive numbers. It requires careful evaluation, strong governance, clear communication about limitations, and ongoing involvement from the experts who understand the clinical reality behind the data.
 

Building the future of the specialty.

AI is a part of veterinary radiology's future, but it is only one part of the story. At IDEXX, our commitment to the profession rests on three pillars.
 

People + technology.

We are building technologies designed to support radiologists, not replace them. Our AI efforts are guided by responsible development practices, strong governance, and direct collaboration with the specialists whose expertise remains central to quality patient care.
 

Education and specialty growth.

We believe advancing the profession means investing in the next generation of radiologists. Through initiatives such as CaseConnexx, ongoing rounds and educational programming, support of IDEXX’s Diagnostic Imaging CE Event (DICE), and sponsorship of Rounds in Veterinary Diagnostic Imaging (RIVDI), we are committed to helping strengthen the veterinary radiology community and expand access to learning opportunities.
 

Giving back.

We also believe that showing up for our communities matters. Every IDEXX team member receives paid volunteer time off, reflecting a broader commitment to service that extends beyond the workplace and into the communities where we live and work.
 

A future worth building together.

Veterinary radiology is entering an important new chapter. The opportunities created by AI are significant, but realizing their full potential will require scientific rigor, transparency, and continued leadership from veterinary radiologists themselves.

At IDEXX, we are committed to building that future responsibly. One where technology supports and amplifies expertise, where specialists remain at the center of patient care, and where innovation is guided by the people who understand the profession best.

For radiologists who share that vision, we believe there's never been a more exciting time to help shape what's next.