Only 424 board-certified veterinary radiologists belong to the European College of Veterinary Diagnostic Imaging (ECVDI), but more than 139 million households across Europe own pets. Many of those pets could benefit from a radiologist’s expertise, but the demand for radiology services far outpaces our ability to supply trained specialists.
Advanced imaging is increasingly common in first opinion practices but interpreting those images quickly and accurately can be difficult without specialist input. Artificial intelligence (AI) tools may help to bridge this gap. With rapid image analysis, well-trained AI programs can support clinical decision-making and help veterinary teams prioritise cases.
As AI radiology tools rapidly evolve, many veterinary professionals have questions about safety, accuracy, and how to blend AI into daily workflows to affect positive change and improve patient access to radiologist support. Learning how veterinary radiology AI is developed, the level of accuracy it can achieve, and the benefits it can bring to practices across the UK and Europe can help to answer those questions.
Veterinary radiology AI development
Accuracy in radiology is critical for diagnosis, treatment planning, and patient outcomes. At Vetology, our veterinary radiology AI is developed in close collaboration with European and American board-certified radiologists, as well as imaging-trained veterinarians. These experts, both boarded and non-boarded, dedicate thousands of hours to labelling and verifying images, establishing the “ground truth” data that trains our AI. While Vetology’s approach reflects this rigorous process, the internal methods of other companies may differ.
AI development in veterinary imaging relies on several deep learning techniques, such as:
– Convolutional Neural Networks (CNNs) – These deep learning models are designed to recognise images and detect patterns. In radiology, CNNs can quickly and accurately analyse radiographs, identify anatomical structures, and flag abnormalities. Multi-output CNNs can also determine species (e.g., dog or cat) and focus on the relevant region of interest.
– Confusion matrices – These performance evaluation tools compare AI predictions against human-verified results. They measure accuracy, sensitivity, specificity, and precision by breaking results into true positives, false positives, true negatives, and false negatives.
– Quality assurance (QA) regression testing – Developers use this process to compare AI outputs against known labelled images, identifying errors and refining classifiers.
– Large language models (LLMs) – Once the AI has analysed the image, LLMs can generate professional reports by describing conditions and summarising findings.
While a new AI classifier is in development, every Vetology report generated by AI is carefully verified against the expert opinion of a board-certified veterinary radiologist. This safeguard ensures that findings and conclusions are consistent with the standards of human interpretation. Because interpreting flat images of three-dimensional structures requires complex reasoning, radiologist oversight remains essential for recognizing subtle variations and clinical nuance.
This is the rigorous process we follow at Vetology; however, we acknowledge that other organisations may employ different methodologies.
Variability in veterinary radiology
Unlike laboratory diagnostics, which often yield objective numerical results, radiology interpretation is subjective. Experienced veterinary radiologists may disagree on the meaning of a particular imaging finding, especially when changes appear subtle or multiple issues are present on the film.
This cross-reviewer variability matters for developers of AI imaging tools. The models learn from human-labelled data, so when specialists disagree, the “ground truth” the AI relies on contains some degree of subjectivity.
Generally speaking, advanced or severe processes (e.g., fractures, heart enlargement) generate the most agreement, while subtle or early-stage changes (e.g., mild degenerative joint disease, diffuse lung patterns) generate disagreement. Recognising differences in interpretation allows our AI system to learn in a realistic setting.
Building something new: How AI compares to humans
Vetology’s goal for veterinary radiology AI is not to replace human expertise, but to reach a level of performance that supports real-world veterinary teams.
Developers compare AI results to agreement rates between board-certified veterinary radiologists. If the AI approaches the agreement rates amongst the veterinarians, then that model can be considered clinically useful.
Independent evaluations of AI tools can help determine reliability and identify where these systems are most useful in practice. When available, published sensitivity and specificity data from confusion matrices could help veterinary teams understand when AI is most dependable and when a case would benefit from human expertise.
In one study at Tufts University, the AI imaging system tested detected canine pleural effusion with 88.7% accuracy, 90.2% sensitivity, and 81.8% specificity.1 In research conducted at the Animal Medical Center in New York, the tested AI imaging tool achieved 92.3% accuracy, 91.3% sensitivity, and 92.4% specificity in identifying canine cardiogenic pulmonary edema.2
Choosing tools with proven accuracy ensures AI can act as a rapid and consistent screening tool in everyday practice. While not a replacement for the nuance and clinical context a board-certified veterinary radiologist brings to the table, it can help identify anomalies quickly, support rapid triage, and inform decisions about when to seek a human specialist review.
Combining AI screening reports with teleradiology
Teleradiology has transformed how veterinary teams access specialist input and second opinions on complex cases. However, sending every image for review isn’t practical or affordable for many clients.
AI screening tools can provide a middle ground with similar accuracy to a human radiologist for the right cases. With a subscription-based digital service, veterinary hospitals can upload images and receive an AI-generated report in a fraction of the time it takes for human review. Teams can use that rapid feedback to:
– Confirm a suspected finding before moving forward with treatment
– Help prioritise urgent cases for faster intervention
– Reduce delays in decision-making during busy periods
If clinicians are still uncertain, they can submit the same images for a full review by a board-certified veterinary radiologist, using the same digital platform. A layered, step-wise approach that combines clinician judgement, AI analysis, and specialist review can increase diagnostic confidence while keeping costs and turnaround times manageable.
For UK and European practices, where specialist access may be limited by geography or general scarcity, combining AI screening with teleradiology offers a practical way to improve case management, workflow efficiency, patient outcomes, and veterinary team member confidence.
No AI system can replace the knowledge, experience, and clinical judgment of a board-certified veterinary radiologist. However, AI-assisted image interpretation and teleradiology can play an important role in supporting veterinarians and radiologists facing cognitive overload and high case/patient volumes. These tools are designed to augment and support human expertise, not replace it, helping to bridge the gap between the growing demand for imaging services and the relatively limited availability of board-certified radiologists.
Getting Started with Vetology AI
Vetology’s AI-powered radiology technology was introduced to UK veterinary practices through a partnership with VetIT. This collaboration has made Vetology’s diagnostic tools available to practices across the UK, offering AI-supported radiographic assessments that assist clinical decision-making. Practices interested in using Vetology AI can contact the VetIT team to learn more about how the service works, request a demo, and explore how it can support their diagnostic workflows https://www.vetit.co.uk/our-veterinary-software/vetology-ai/
Cory Clemmons is the Chief Technology and Strategy Officer at Vetology, bringing over 25 years of experience in technology leadership and business strategy. Prior to joining Vetology, Cory served as Chief Technology Officer at Stretto, where he was responsible for establishing the company’s technological vision and directing its development. His career spans senior roles in technology and financial services, with a focus on innovation, strategic growth, and operational efficiency.
References
- Müller TR, Solano M, Tsunemi MH. Accuracy of an artificial intelligence software for detection of confirmed pleural effusion in thoracic radiographs in dogs. Vet Radiol Ultrasound. 2022; 63: 573–579. https://doi.org/10.1111/vru.13089
- Kim E, Fischetti AJ, Sreetharan P, Weltman JG. Fox PR Comparison of artificial intelligence to the veterinary radiologist’s diagnosis of canine cardiogenic pulmonary edema. Vet Radiol Ultrasound. 2022; 63: 292–297. https://doi.org/10.1111/vru.13062
