Ethical issues are raised by the application of AI in veterinary medicine

Because there aren't enough practicing veterinary radiologists to meet demand, AI is presently promoted to veterinarians for radiography and imaging. Cohen argues that an experienced radiologist analyzing photographs in light of an animal's medical history and particular circumstances is not the same as AI image analysis. Users must be aware of potential restrictions even though some X-ray situations may be accurately identified by AI. The AI might not be able to reliably distinguish between disorders that appear identical on X-rays but require different treatments, for instance, or it might not be able to recognize every possible condition.

According to Eli Cohen, associate clinical professor of radiology at NC State's College of Veterinary Medicine, "one significant difference between veterinary and human medicine is that veterinarians can euthanize patients—which could be for a variety of medical and financial reasons. Therefore, the stakes of diagnoses provided by AI algorithms are very high." Cohen analyzes the moral and legal issues that are brought up by the veterinary AI products that are now in use in a review for Veterinary Radiology and Ultrasound. He also identifies important distinctions between veterinary AI and AI employed by human clinicians. "Veterinary AI products are currently not subject to any regulatory scrutiny, whereas human AI goods must be verified before entering the market."

Cohen claims that AI is frequently a "black box," which means that not even the developer is aware of how it diagnoses or makes decisions. "When you combine that with lack of transparency by companies in AI development, including how the AI was trained and validated, you're asking veterinarians to use a diagnostic tool with no way to assess whether or not it is accurate. "Since veterinarians frequently get a single visit to diagnose and treat a patient and don't always get follow-up, AI could be providing incorrect or incomplete diagnoses and a veterinarian would have limited ability to identify that unless the case is specifically addressed.

Before AI technologies are made available to the general public, Cohen advises that veterinary specialists collaborate with AI developers to guarantee the caliber of the data sets used to train the algorithm. He also suggests that third-party validation testing be carried out. "There is a market void, so AI is being positioned as a replacement or as having value comparable to a radiologist interpretation. The ideal use of AI moving forward, and undoubtedly during this early phase of deployment, is what Cohen refers to as "a radiologist in the loop," where AI is used in combination with a radiologist, not in place of one.

This is the most moral and reasonable way to use this emerging technology, according to the author: by using it to increase access for veterinarians and animals to radiologist consultations, but most importantly by having subject-matter experts troubleshoot the AI and stop bad outcomes and patient harm. According to Cohen, almost every diagnosis a veterinarian could make from radiographs has a medium- to high risk of resulting in modifications to medical care, surgery, or euthanasia, either because of the clinical diagnosis or client financial restrictions. The FDA uses that risk level as the cutoff in human medicine to decide whether a radiologist should be kept in the loop. We should follow a similar model as a profession.

Unlike in human medicine, the FDA currently does not regulate AI in veterinary products. Without any additional regulation save that given by the AI developer and/or corporation, veterinary products may be put on the market. AI is a potent instrument that will transform how medicine is carried out, but the ideal approach moving forward will be to work with radiologists to use it in conjunction with them to enhance patient access and treatment quality rather than to replace their consultations.