Authentic AI Conversations

Stakeholders in veterinary medicine gathered to share their knowledge and identify research focus areas and priorities with artificial intelligence. What they talked about could have far-reaching implications.

Last spring, Cornell University hosted a Symposium on AI and Veterinary Medicine (SAVY). Various stakeholders across the industry – both academic and commercial – gathered to discuss a wide variety of topics related to AI. Parminder S. Basran, PhD, FCCPM, associate research professor at the University of Cornell College of Veterinary Medicine, spoke to Vet-Advantage about SAVY, what led to the symposium, what was discussed, and the potential for AI in research and medicine.

1. What was the motivation behind the symposium?

One of the key drivers was that there needed to be a space where researchers, students and general practitioners can come together and talk about artificial intelligence. At the time, there were no dedicated opportunities for people who had an interest in artificial intelligence and veterinary medicine to come together. Many people at the symposium have probably attended their specialty meetings like radiology or pathology, but there was nothing that really unified it from a big-picture perspective of the whole space of veterinary medicine and artificial intelligence.

So, it was really important to create a welcoming, safe environment where people of any background can come and chat and discuss with other folks about their experiences with artificial intelligence.

Another critical factor was to provide a vehicle for the next generation of scientists and leaders that will bridge this field of artificial intelligence in veterinary medicine. We wanted to give them an opportunity where they can meet, make connections with other researchers, showcase their results, and get feedback on the kinds of work they’re doing in a welcoming environment.

Many researchers in AI tend to come from computing science, and so making an impact in veterinary medicine with AI, you need multi-disciplinary teams to work on problems together. If you have computing science people embedded in a radiology department, for example, you’re probably going to be in a better position to see the fruits of AI in radiology. We wanted to try to create this environment in SAVY by bringing in cutting-edge AI researchers.

Researchers are interested in creating new things like AI, software and technologies. The translational part of that – trying to take those ideas and make them practical – requires creating an opportunity for researchers and clinicians to connect.

Many involved in this movement are undergraduate, graduate and PhD students. These are students doing population medicine, livestock health, and companion animal health in veterinary medicine. They are the younger generation of people that are really interested in tackling these problems and using these sophisticated tools. We wanted to find a way to provide them with networking opportunities and showcase their results, so a big part of our symposium involved poster presentations and oral presentations of preferred talks.

We also had an opportunity to hear from industry stakeholders so that they could share some of their AI-related research, work and products. We were not in the business of showcasing all the commercial products. If a company has an AI solution, and they’ve undergone rigorous testing, this was an opportunity for them to present their results.

So you can see three major audiences here; the academics or researchers in university environments who are in the computing science and the veterinary medicine side; the trainees, students, and PhD students trying to find opportunities or showcase service of their findings; and then industry professionals who wanted to listen and see what we see and also participate in the discussions.

That was the motivation behind it all. We were really pleased with how receptive everyone was to this symposium. By all indications, the message we got was this is an excellent idea.

2. The symposium’s events were organized around four pillars of veterinary medicine where AI is already having a large impact – patient-facing medicine for companion animals and livestock, population medicine and One Health. Why those four?

We wanted to have a big umbrella for this conversation. We’re acutely aware that each of our specialties might be getting introduced to AI in some way or form, but piecing it together into a big picture, a global perspective, is important. We hope AI can solve some of the difficult problems we face in society and veterinary medicine.

We also know that veterinarians are involved in many things besides companion animal care. For instance, veterinarians are regularly employed in the FDA to manage pandemics and epidemics of diseases. It was important to try to capture all the different ways that veterinarians can interact with society. We wanted to make sure we added a space for them to discuss and break down some silos to try to look at many of our problems through a different lens. Often, we find one particular silo may have a solution to a problem using some artificial intelligence methods, but another silo may have never thought of using those tools.

In one instance, a talk was given on using machine learning techniques for blood serum analysis. Well, there’s no reason you can’t use those same machine learning techniques for handling other problems, like population health challenges using demographic information. We wanted to try to mix it up and allow that cross-fertilization, so ideas can get bubbling, and we can create new collaborations.

3. What were your takeaways from the industry roundtable that covered AI challenges and solutions?

One of the key points that we learned from that round table was there is a need for the kinds of skill sets that bring AI to the table in veterinary medicine. What does that look like? Does that mean we change the curriculum of veterinary medicine for students? How do we integrate the concepts of AI in veterinary medicine education to produce the skill sets needed to operationalize AI in the next coming years?

Another issue was improving the collaboration among data scientists, information technology and veterinary medicine in general. We’re still in this infancy where we don’t have an institute of artificial intelligence and veterinary medicine. We have lots of researchers in this space. But we should also create environments where computing scientists and data scientists and veterinary medicine folks are actually in the same room, working out problems and developing solutions.

4. What are you hearing as far as what we need to do as an industry to create better collaboration and guardrails for AI adoption?

There are a lot of things that we ought to think about. One of the critical things that has really jumped out is access to data. Data can be used to benchmark algorithms and establish performance standards for AI algorithms.

For example, one of the things that doesn’t get the credit it deserves is the creation of ImageNet by a Stanford scholar who essentially said, “We have all these fancy algorithms that are trying to determine if this thing in this image is a car, a dog or a horse. Why don’t we create a data set that actually labels all this information?” So, she and her team of researchers created the ImageNet database and published it online.

Once you have that reference data set, anybody can develop their own algorithm and test it.

Someone can test their algorithm and see that it performs 50% or 80% better than the original. Another person can make a small improvement to that algorithm and increase the performance even more. You can see how the availability of this well-curated and labeled data really helps push the technology forward.

That is really hard to do in veterinary for several reasons. There are all kinds of issues related to the data. Who owns it? Can you make it public? If you’re an academic entity, you can jump through a few hoops and publish your data set online so anybody can access it. But if you’re a private company with a vast data set, can you – or should you – make that data set public? There are some questions there. Maybe they don’t, and you can’t blame them, but the quantity and quality of good label data sets for veterinary medicine is sorely lacking, and we need to do something about that.

As a community, we need to come together and find ways to create good quality data sets that people can test on their own. Say someone markets a radiology AI tool they say can save a veterinarian hundreds and hundreds of hours. Wouldn’t it be nice to be able to test it with an independent data set to see how well it performs, and then I can make a good decision about how comfortable the veterinarian is in purchasing product A or product B? These things exist in human medicine but don’t exist yet in veterinary medicine. So that’s a big challenge for us.

Parminder S. Basran, PhD, FCCPM

Dr. Basran is an associate research professor and medical physicist at Cornell College of Veterinary Medicine, providing research, education, and clinical support. He is passionate about sharing knowledge of what medical physicists do with the public and improving global access to cancer care.

Photo credits:

istockphoto.com/Thanadon Naksanee

istockphoto.com/ruizluquepaz