By Diane Wilson, DVM, DACVR • Mars Science & Diagnostics - Antech It’s no doubt that AI can sometimes feel like a black box. But despite that, it continues to transform traditional practices, and veterinary medicine is no exception. AI is now regularly used in various workflows, including in assisting at the physician level in triage, and in imaging and pathology such as counting mitotic figures. When it comes to diagnostic imaging, AI can help speed up radiologist throughput by properly orienting images on the screen; it can help with a tool like assisted reading where the radiologist receives information about what the machine has found on the images, then can edit those findings, if needed. The machine can also assist in writing reports for the radiologist that the radiologist is then free to edit as necessary. Driving Success in AI-assisted Medical Imaging Successfully implementing AI in diagnostic imaging requires four key elements: Domain Experts: These are the board-certified radiologists. The image interpretation by a board-certified radiologist is the current gold standard in diagnostic imaging. Because machine learning is brand new technology, we weigh it against that gold standard, and against its usefulness to the practitioner. Data Scientists with Expertise in AI: These are not computer scientists; they have expertise in artificial intelligence and machine learning and employ many tools such as natural language processing. Vast amounts of data: Data for training, separate data for testing, and still different data for validation. At Antech Imaging Services (AIS™), we found that in very general terms you need 4,000 to 5,000 examples for each finding to have reliable accuracy data. End User Education: Most essential to successful use of AI in veterinary medical diagnostic imaging is what we call ‘end user education’. AI is brand new technology for our profession, and it's extremely important for anyone using the tool to educate themselves on what it can—and perhaps more importantly what it cannot—do. AIS RapidRead™ is not a mechanical radiologist; there is a finite list of abnormalities that it can detect and understanding what it can do and coupling that with what type of information you're looking to add to the case is crucial. AI at Antech Imaging Services (AIS) AIS has incorporated the AI technology of machine learning into more than just a screening tool by blending the speed and efficiency of the AI with the expertise of board-certified veterinary radiologists and our team of data scientists to create AIS RapidRead™, an AI-powered interpretation tool that provides accurate evaluation for a number of radiographic findings. Whether it’s a pre-anesthetic screen, radiographs for a patient admitted through the emergency department and requiring a quick confirmation of a diagnosis, or immediate answers for a patient in surgery. How AIS RapidReadTM works Ensuring that technological advancements like machine learning are developed with patient care at the forefront is essential. AIS RapidRead was trained on a database of over 16 million radiographs, incorporating dog and cat breeds of all shapes and sizes, including a wide range of films, to gain the highest accuracy possible. This extensive training enables the AI to achieve results that are at least 95% accurate when compared to findings by board-certified veterinary radiologists. AIS RapidRead offers screening for the most common radiographic findings, with results in less than 10 minutes. This is beneficial as a speedy pre-anesthetic screen as well as where timely decisions are critical, such as in clinics offering emergency or urgent care services. Clinics with young veterinarians have found that the extra support and reassurance builds confidence. Patient care is paramount, so ongoing robust quality control is essential. Board certified radiologist evaluation is the current gold standard for image interpretation. Following this gold standard, AIS RapidRead is backed by AIS’ broader team of 140+ radiologists providing quality control reviews and overreads. We want to ensure that no patient goes to surgery or receives any sort of life altering treatment without confirmation by a board-certified radiologist. As a result, potentially life altering findings such as gastric dilatation volvulus (GDV) and GI obstruction are automatically routed to a board-certified radiologist for STAT confirmation at no additional cost to the clinic. Why education must be an essential part of AI adoption It's important that all of us understand that we are responsible to educate ourselves on the proper use of AI in veterinary radiology. A tool like AIS RapidRead is not a mechanical radiologist, it’s an interpretation tool that only learns what our team of board‑certified radiologists and data scientists teach it. It’s not like other forms of artificial intelligence such as generative AI that learn on their own. To promote education around AI interpretation tools, we require hospitals go through specific training to use AIS RapidRead. The training is designed to help the clinician understand when and how to use it, as well as when not to use it. AIS RapidRead looks at radiographic images of canine and feline thorax, abdomen, and limbs for 50+ findings; it cannot (yet) provide comparison studies, evaluate species other than canine and feline, and it doesn't evaluate the pelvis, spine or skull, so knowing when to use the tool is essential to its success. AIS Rapid Read represents a significant leap forward in veterinary radiology, combining the speed and efficiency of AI with the expertise of board-certified radiologists. But we must remember that this technology is in its infancy and best practices are still being explored. Proper use of tools like AIS RapidRead can both improve patient care and play a crucial role in shaping the future of veterinary medicine. Dr. Wilson was awarded her DVM degree from Tuskegee University and her DACVR status after her residency at Auburn University. As Senior Director of Special Services, she focuses on machine learning (ML). Dr. Wilson places the utmost importance on ensuring that technological advancements, like machine learning are developed with patient care as the pinnacle. She is excited to take part in advancing the radiology profession into the next phase.
