These days, providing our pets with the most state-of-the-art veterinary care is more attainable than ever. Between identifying a swallowed stray children’s toy or assessing the root cause of unusual clinical signs, veterinary radiology is an absolutely crucial aspect of care and truly saves pets lives each and every day. Working in such an essential line of pet care, veterinary radiologists are in high demand, often leaving them overworked and prone to subjectivity, bias and fatigue. So, it goes without saying we need tools to help our veterinary teams get the most accurate, fastest, and most cost-effective results. Our pets’ lives could depend on it!
At SignalPET and PetScan, we see a clear path to enhancing veterinary care. AI has the potential to help radiologists in report writing and make pet radiology more accessible. Rather than solely trying to boost the number of radiologists to ease the demand, why not offer tools that can assist those we have, and maybe even, one day, improve their work?
For example, AI does not suffer from excess variation due to human judgment. Of course, when a tired, overworked, or even a freshly minted radiologist is reading reports, we can run into this kind of subjectivity or error. It happens to anyone in nearly any field. We are all human! But when it comes to saving the lives of our pets, a second opinion from AI could prove very effective in assisting veterinarians in their difficult and important jobs.
To see just how effective this technology is and assess its potential for improving pet health care, we supported a study to establish the new gold standard. The study conducted a comparison of radiological interpretation made by veterinary radiologists and AI software designed to analyze radiographic studies of cats and dogs. So, which interpretations were more effective? Let’s analyze the details of the study to learn more!
The study was established with the purpose of uncovering how commercially available veterinary radiology AI software compares to veterinary radiologists in reporting radiographic findings. The researchers predicted that the AI will be, on average, more accurate than the radiologists, and more accurate than even the highest performing radiologists in the study. They used SignalPET’s SignalRAY® technology to create the comparison and test their hypotheses.
So, how did they do it? The study matched AI tests with the radiologists’ interpretations. Each of the findings was then classified as normal or abnormal. They tested validity based on the sensitivity, specificity, and accuracy of the readings. The study established the ground truth, or the correct reading, as the consensus among the majority of the radiologists and used statistical testing to determine the significance of the results.
As the researchers expected, the AI technology outperformed a significant number of radiologists in the study. In addition, there was no significant difference in accuracy between AI and the best-performing radiologist. Believe it or not, the AI performed almost as well as the best radiologist in all settings for radiographic findings! The study also found that the AI software was biased towards claiming abnormal results, especially when it came to the unclear findings.
Essentially, the AI takes a conservative and ultra-safe approach. This affirms that early veterinary radiology AI’s strength lies more in detecting abnormalities than confirming normalcy. This bias can save time for radiologists who may not need to overly-scrutinize normal results but will want to take a closer look at abnormal results discovered by the AI.
Ultimately, the study revealed that this early-generation veterinary radiology AI can serve as an aid, rather than a replacement for a radiologist in vet practices. While it doesn’t save a radiologist from analyzing the case, it can improve the quality of care by offering a second opinion with a more standardized approach. When the AI flags something as abnormal, it encourages a second look and a more detailed analysis — a great safety feature!
This can be an excellent tool to save time for radiologists and assist them in their reporting. The research found that the AI technology is especially useful for non-controversial cases of abnormality or when assessing borderline findings. As time goes on and the technology continues to develop and learn through its experiences, we expect to see major improvements in its ability to assist and enhance veterinary care.
This preliminary study was a great way to start assessing the benefits that AI technology can have for vets. But there is still more research to be done to understand the full extent of what AI can really do now, and in the future, to advance veterinary care and save pets’ lives.
As it often goes with research, the study design was not a fully accurate representation of a real-world environment. Synthesis of radiographic findings into diagnosis is the most important part of the diagnostic process. Since AI does not provide a diagnosis or an assessment of the medical history of the animals being analyzed, this study could only compare radiographic findings. In addition, the AI technology provides findings per single radiograph while radiologists can look at multiple images for a more holistic view of the pet’s condition.
This early-generation technology shows great promise, even as veterinary radiology AI is still in its beginning stages. The results are truly inspiring when it comes to the usefulness of AI for current applications in veterinary care, and even more so for its potential to redefine what the most state-of-the-art veterinary care and technology looks like.
In the future, the research suggests that it will be worth exploring how AI could be provided to veterinary radiologists to actually enhance their performance beyond what is possible for a human alone. SignalPET is already working closely with radiologists to provide them with this value. As the world catches on to how all kinds of AI can make our lives easier, more efficient, and more accurate – early adoption of AI technology has the potential to truly revolutionize the way we live, work, and even how we care for our furry friends.
Schwarz T, Ndiaye YS, Chernev C, Ockenfels AS, Crampton P (2023)
Comparison of radiological interpretation made by veterinary radiologists and AI software for canine and feline radiographic studies.
ACVR Annual Scientific Meeting, New Orleans, LA, USA, October 25 – 28, 2023.
(oral presentation) abstract in: Proceedings; Veterinary Radiology & Ultrasound in press.