On Animal Health AI and Innovation
The Veterinary Innovation Council’s Dr. Aaron Massecar says the animal health industry needs to be asking the right questions when it comes to AI’s potential uses.
As executive director of the Veterinary Innovation Council, Aaron Massecar, Ph.D., is well aware that people in the industry are buzzing about artificial intelligence.
“That is the hot topic,” he said. “Everyone wants to talk about AI. There are so many different ways in which it’s hitting the space. It’s fascinating to think about.”
How do practitioners use AI in their clinical work? What about their business workflows? And how will companies leverage the technology for things like diagnostics? The list is endless.
In an interview with Veterinary Advantage, Dr. Massecar discussed AI’s potential, as well as what other trends to keep an eye on as the industry implements new technology.
Vet-Advantage: Can you tell us a little bit about the Veterinary Innovation Council and your work there?
Dr. Massecar: We have a sponsor-driven board with companies such as Mars, IDEXX, PetSmart Charities, PetCo Love, and others. They sponsor the work that the Veterinary Innovation Council does, including the Veterinary Innovation Summit that happens every year and that’s the crowning achievement of our work. The goal of the Summit is to get veterinarians, veterinary professionals and the industry more broadly to understand what the future looks like and bring that a little bit closer.
We look at topics like AI and big data, personalized medicine, genomics, etc., topics that are poised to impact the profession 5 to 10 years from now. How can we, as a profession, make sense of them? How do we incorporate them into our practice? How do we make it a collaborative effort rather than something that happens to the profession?
We also work with startups, helping them through basic education and pitch competitions, such as the one at the Veterinary Innovation Summit and another at NAVC’s VMX.
I’m also working to better understand where all the private equity and venture capital is coming from. Can we better understand that so that we understand who the companies are, and then do some matchmaking to drive innovation?
We want to drive innovation forward. We don’t want to be reactive; we want to be proactive. So, we’re looking at developing something like an XPrize where we create a particular set of criteria and then reward the company who fulfills those criteria first. For example, can we cure veterinary burnout? There is, of course, a lot of work that would be required to make that happen, such as getting a clear definition of what burnout is and how it can be cured, but that’s the kind of moonshot that we want to start advocating for.
Vet-Advantage: Have you noticed an uptick in conversations about artificial intelligence?
Dr. Massecar: That is the hot topic. Everyone wants to talk about AI. There are so many different ways in which it’s hitting the space. It’s fascinating to think about. If I’m a practitioner, how do I use AI? Then there are the big companies who are looking at leveraging next generation AI for things like diagnostics, for example.
There’s a lot of conversation around the use of AI in radiology as a starting point. Everything that needs to be done in radiology, everything from creating guidelines and regulations around development and use, if it’s done right, will set up all the other uses of AI that are going to hit the market soon. For example, there are a lot of questions about how the algorithms are being trained. If they’re being trained appropriately and with transparency, then we can be sure that the outcomes from those algorithms are going to be accurate. Without knowing how an algorithm was developed, what data they used to train the algorithm, then we don’t have the level of trust that we need. Creating guidelines and regulations will only increase trust in the use of AI.
The learning curve for understanding what’s happening with AI and radiology is so steep that I think it scares a lot of people away. There are some really basic questions that people need to be asking. If we can educate people on what those basic questions are, then when AI is introduced into ultrasound, cytology, endoscopy – any area where it uses visual imagery or text – we’re going to have a better understanding of what’s happening. We’re going to understand whether or not the tool was built appropriately.
Vet-Advantage: What are the questions that need to be asked?
Dr. Massecar: One of the basic ones is, how is the model being developed? Is the model being developed with a particular data set that forms the ground truth, and is that data set publicly available? Or is the model being developed in an iterative fashion, where every single time somebody uses it, it’s getting better and more refined? If that’s the case, then you have an issue where you’re pulling data from the individual practice. How is that data anonymized and protected? Who owns the data? What is the role of the veterinarian in the production of that data? There are data privacy issues that exist around that.
And there are questions around, how exactly was it trained to begin with? What is the ground truth? If you take, for example, a data set of 50 radiographs, and you put those 50 radiographs in front of a team of board-certified radiologists, you’re not going to get perfectly consistent responses. There are some things that are going to be pretty obvious, and other things that are going to be less obvious. How are you making decisions around the ground truth of the data set you’re using? You’re moving from subjectivity to objectivity that will then inform all future diagnoses, so how do you make that choice as a company?
Most practices that are evaluating the use of the technology are asking financial questions: What’s the cost per use? Most practices don’t ask about the error rate of the technology. For example, how often is an algorithm returning a false result? Does the algorithm return a confidence score? If the AI is doing basic triage work, and there is a need for a stat read, then where does the doctor go for more information? If a doctor wants to know how the algorithm determined that a fracture was present or why there is a particular area of interest, where can the doctor go for more information? Being able to ask those questions upfront and get answers to those questions is only going to help increase confidence in the use of the AI.
If you don’t know what the ground truth is, if you don’t know how the algorithm is being trained, if you don’t know how it was originally defined and whether it’s iterative, if you don’t know what the data and privacy issues are around the technology – if you don’t know any of those things – you’re opening yourself up to potential challenges. For example, we’re just starting to become aware that, ultimately, it’s the individual doctor using the technology that is responsible for any diagnoses that come from that technology. If a doctor doesn’t understand how a diagnosis is given, then we have a case of “black box” medicine that is, at best, irresponsible.
But if we start from the beginning and say, these are the things we need from these companies, and we set those up in advance and ask who’s meeting these criteria, then we’re going to be in a much better position to utilize this long term.
Vet-Advantage: Why is radiology a good place to start? Why is that important as far as being on the ground floor of AI implementation in veterinary medicine?
Dr. Massecar: There are a few reasons. One, it has been used in human medicine for a while now, and so there are many lessons that we can quickly learn from them. Second, it’s a flat image, so it’s easier to develop and use the algorithms on that instead of, for example, 3D images. Third, because there are not enough radiologists to meet the current market demand, AI is filling that market gap. In addition, the lack of radiologists means that there aren’t enough radiologists to meet the market demand and the educational demand. The lack of radiologists then becomes a compounding growth problem where the market demand is going to increase exponentially because there aren’t enough academic radiologists training young vets, young veterinarians are not going to feel like they’re adequately trained and so they’re going to require more stat reads, which will put pressure on existing radiologists and drive up pay for those specialists, which will then pull them out of academia, which will then make fewer teachers and more young vets lacking confidence.
Because of the human capital shortfall, AI software technology is filling the gap. It’s happening to radiology now, but there are many other areas that will see the exact same cycle in the near future, everything from ultrasound and clinical pathology to diagnostics and decision support. AI is going to continue to fill the widening gaps that are only beginning to emerge, which is why it’s so important that we get things right with radiology first and set up frameworks for proper development and implementation now.
Vet-Advantage: How would you conceptualize the potential AI has for the veterinary industry?
Dr. Massecar: AI is a broad category that contains within it all sorts of buzzwords: Large Language Models, Generative, Deep Learning, Machine Learning, etc. In the near future, we’re going to be more careful with the language that we use when we’re talking about AI, but for right now, it’s the buzzword that not only gets practitioners interested, but also gets venture capital very interested.
If we look at the history of AI, we can see that there was something called the “AI Effect,” which essentially means that things that were thought of as AI in the past have actually just become common use. For example, when optical character recognition first hit the market, it was called AI. Now it’s just called optical character recognition. Over time, things that we hastily grouped under one category, will slowly get parsed out into more specific domains. It reminds me of a quote from Arthur C. Clarke in 1962: “Any sufficiently advanced technology is indistinguishable from magic.” That’s where AI is living right now.
In terms of the effect on the industry, we are just at the very beginning of change. Flat image recognition is here, dynamic image recognition is on the second horizon (e.g. endoscopy), computer vision is also here, which will help with diagnosing orthopedic issues. Decision support is here through AI PIMS like clini.ai or Veterinary Electronic Assistant – they are both provided automated SOAP notes and are starting to dig into the literature to be able to come up with treatment plans that rely on the most recent research. Decision support tools are being built so that vets have the entire body of knowledge in a more accessible way, right on their phones. We are truly just at the very beginning of the innovation trigger phase of the hype cycle when it comes to the use of AI and nobody, myself especially, has a full grasp of the scope and nature of the changes that are coming to vet med in the next 2-4 years.
Photo credit: istockphoto.com/FG Trade
