by Maureen Blaney Flietner

IF YOU’RE WONDERING WHAT YOUR VETERINARY FUTURE HOLDS, you can bet that artificial intelligence (AI) will be involved in it.

But just what form AI will take may be difficult to determine, partly because ideas about what “intelligence” is keep changing. Or, as computer scientist John McCarthy, one of the founding fathers of AI, reportedly said, “As soon as it works, no one calls it AI anymore.”

“Artificial general intelligence” refers to that bit of fantasy reserved for machines that might be able to exhibit human-level intelligence—learning, understanding, and forming connections and generalizations across different areas.

Instead it is “artificial narrow intelligence”—task-focused AI that is far removed from artificial general intelligence—that powers the many tools now ubiquitous in our lives. And experts suggest that those working in veterinary medicine have roles to play to increase the benefits available from it.

Understanding the Process

Nathan Bollig, DVM

“Every veterinarian, practice owner, administrator, project information management system developer, and pet owner should advocate for high-quality animal health data,” said Nathan Bollig, DVM, postdoctoral fellow in computation and informatics in biology and medicine and PhD student in computer sciences at the University of Wisconsin–Madison. “Because when data is well structured, complete, and easily accessible, it becomes a rich resource that has the potential to improve the quality of life for people and animals.”

How? One of the more dominant subsets of AI that is driving many breakthroughs is machine learning (ML), and it has a massive hunger for data. ML uses programmed algorithms—sets of rules or steps—to analyze inputs and learn patterns to predict outputs. The machine learns and improves its operations to advance performance, developing “intelligence” over time.

“That data also needs to be mostly accurate and similar to the type of data that the model will encounter in the future. This means that institutions or vendors that want to incorporate more ML-based analytics or clinical-decision support into their systems need to make sure that they are collecting and storing data in an ML-amenable form. Storing patient data in a direct, structured format can make all the difference,” explained Bollig.

“If veterinary medicine is going to see a digital revolution as robust as we are seeing in human clinical informatics, we need to talk about how to manage and curate data to support this revolution. The issues at the forefront should be data sharing, interoperability, standardized terminologies, and structured coding of diagnoses and other clinical information. If diagnoses are consistently and correctly coded with the use of SNOMED CT (VetSCT), it would open more possibilities for extracting value and insight.”

Bollig offered one project he led as an example. He and his team investigated how ML models—computer programs that have been trained to recognize certain types of patterns—could be used for syndromic surveillance. Using 33,567 necropsy reports in free-text format with no structured diagnoses from the Wisconsin Veterinary Diagnostic Laboratory, the team applied different ML models to check for broad categories of disease rather than specific diseases. Their models identified a spike in gastrointestinal illnesses from one source. The study’s top-performing model relied most heavily on words that were medically appropriate. The key, he said, is that they did not have to specify the terms in advance. The model learned—refining and updating its knowledge automatically.

“Syndromic surveillance could become relevant to many veterinary practitioners. If the profession continues to make steps toward increased data sharing, we will find ourselves with increasingly large quantities of data that could present more opportunities for disease monitoring. Additionally, if veterinary software vendors incorporate more ML-based analytics into their systems, they might be used by an individual hospital or clinic for surveillance or other clinical-decision support applications.”

General practice veterinarians interested in the clinical applications of AI might want to check out the Association for Veterinary Informatics (avinformatics.org), suggested Bollig.

Recognizing Disease Patterns

Krystle Reagan, DVM, PhD, DACVIN (SAIM)

At the University of California, Davis, School of Veterinary Medicine, Krystle Reagan, DVM, PhD, DACVIN (SAIM), and Chen Gilor, DVM, PhD, DACVIN (SAIM), were bothered by the fact that Addison’s disease has predictable but subtle abnormalities in bloodwork that often make it difficult to recognize. They wondered whether AI could find the elusive patterns more effectively.

They gathered the results of routine screening tests from more than 1,000 dogs—with and without Addison’s—who had been evaluated at UC Davis and entered them into their specifically designed algorithm. The goal was to see whether the algorithm would create an alert system to inform veterinarians when the data showed the disease as likely and that more investigation was needed.

Their AI-powered algorithm did just that—with an accuracy rate greater than 99%.

The team filed a nonprovisional patent through the UC Davis Office of Research. Reagan said the technology is expected to be validated in large commercial laboratories in the coming year, with the ultimate goal of getting it into the hands of veterinarians.

“AI truly has the potential to revolutionize how we practice medicine, ranging from diagnostics, monitoring of patients from afar with wearable devices, improving our efficiency in the hospital with administrative tasks, and more,” she said. “We should understand that AI tools can be used to improve our patient care and connection with our clients and patients.”

“My hope is that, as clinicians, we can help identify areas that could benefit from the addition of AI and help guide the development of these tools so they can be used in the most efficient and safest way possible,” said Reagan.

“Ultimately, the incorporation of AI technologies into the veterinary field will change the industry and lead to cost savings for practices.”—Jamie Perkins, DVM

Helping Veterinary Education

Jamie Perkins, DVM

Jamie Perkins, DVM, assistant professor, College of Veterinary Medicine, University of Arizona, was a veterinary student when she came up with the concept of using AI to help improve veterinary education.

Cell phone coverage issues with a computer that transcribed Perkins’s brainstorming ideas as she talked with family and colleagues while driving sparked an idea: Create a hands-free training tool that would use the Amazon Alexa. Alexa uses the subsets of AI known as natural language generation and processing and ML.

Perkins’s first design was rough, but it could be demonstrated to others. She started networking, met the founder of Voiceflow, used its tools to create skills faster, and received funding for the project.

“Originally, the projects included apps and eBooks with instructional videos and whiteboard video explanations. Demoing those early projects helped me secure a faculty position shortly after graduation. Since then, I have been granted the flexibility to develop projects ‘in the sandbox,’ which eventually led to the Alexa projects.”

One goal was to ease the challenge faced by veterinary educators: teaching communication and clinical reasoning skills, which requires experienced personnel and sufficient time with those personnel. Many systems are screen based or require too much human capital to be sustainable for more than one or two sessions per semester per student, she explained.

Perkins noted that her Alexa skill, which allows a student to practice at any time in any place, is consistent, always available, and comparatively inexpensive. However, it does have limitations. Since it is a voice-only skill, it does not train the user to interpret body language or other nonverbal cues and does not understand or pronounce all medical terminology correctly.

Eventual robust and well-designed AI systems could provide many benefits to universities and their students, said Perkins, noting that she has received unofficial positive feedback from students who have tested it.

The ultimate goal is for students to pilot-test the Alexa skill, with the college providing 25 Alexa devices and making the skill available to any of its interested College of Veterinary Medicine students with an Alexa-enabled device. There are also plans to acquire headphones and video-enabled devices for other Alexa research projects.

Perkins said she sees the future of AI as limitless.

“We are working on some great projects for veterinary education, particularly systems that change student assessment. Education will eventually move to a system in which students are not assessed by evaluating a learning outcome that is temporarily mastered and then forgotten. Our AI system will allow students to engage with a computer-based system in an oral format that ‘learns’ and redirects a student back to areas the individual needs to have reinforced. Such a system would allow for detailed characterization of content for the students and the educators working with each student.”

Perkins believes that AI systems may soon manage some of the technical aspects in private veterinary practices.

“There are already systems that can interpret radiographs and some pathology findings. These systems will be significantly improved in the coming years, leading to increased productivity and consistent diagnosis of some conditions. AI systems will be improved to manage hands-free, voice-activated access to content for both students and practitioners. Some of this content will be improved drug formularies, reference materials, and client-education tools. While some AI technology may not be groundbreaking, it will increase productivity in a clinical setting.”

Perkins said there is a lot on the horizon with AI in the veterinary industry.

“Some developers like me are building AI specifically for the veterinary-education market, but others are developing technology that will benefit clinicians in their daily lives. There are also companies that have developed AI systems to interpret radiographs, provide medication dosages on the go, and troubleshoot issues with clinical instrumentation. Ultimately, the incorporation of AI technologies into the veterinary field will change the industry and lead to cost savings for practices.”

Freeing Up Time for Vital Tasks

Yao-Yi Chiang, PhD

AI technologies are helping to automate many routine tasks and reduce the required expert intervention in the process, explained Yao-Yi Chiang, PhD, associate professor (research) of spatial sciences, at the University of Southern California Viterbi School of Engineering. With research in artificial intelligence and data science, he develops computer algorithms and applications that discover, collect, fuse, and analyze data from heterogeneous sources to solve real-world problems.

“AI technologies can help veterinarians collect data from existing cases, generate potential treatment plans, and conduct causality analysis using large datasets,” Chiang said. “However, we should see these technologies as a set of very useful tools but not as a replacement of human experts.

“Using these technologies can help free precious expert time so that they can focus on important tasks that only humans can do well, such as maintaining good communications between doctors and [clients], designing treatment plans, and such.”

Chiang said he also does not want people to get unrealistic hopes about AI technologies.

“When the technologies do not live up to their expectations, they might think AI is overhyped and quickly dismiss it entirely. In analogy, sometimes patients think medicines are the whole cure for their conditions but, in reality, very often medicines only help relieve symptoms and prevent the existing condition from worsening. If the patients do not fully understand the benefit of their medicines, they might be disappointed and seek alternative treatments that are clinically less useful.”

It’s important, too, said Chiang, to remember that AI is not infallible. AI technologies typically require a huge amount of training data, especially for the deep-learning models that are currently everywhere, doing face recognition, text recognition, speech recognition, and autonomous driving and guiding recommendation systems, he explained.

“And yes,” he said, “there will always be bias in the results of any kind of AI technologies, whether intentionally or unintentionally generated. As users of AI technologies, we should at least understand the risk and never overestimate the capability of a certain technology.”

But with AI’s ability to help quickly organize huge amounts of data and prepare it for analysis, Chiang said he expects many benefits. He predicted that one direct result would be new types of drugs or new applications of existing drugs. There also would be the capability to efficiently detect relationships between certain conditions and diet or drugs, he noted, pointing to the Food and Drug Administration investigation into the potential link between certain diets and canine dilated cardiomyopathy.

“So now,” said Chiang, “the question to veterinary medicine practitioners is, ‘If technologies can free you from doing tedious, repetitive tasks, how would you use the saved time to advance the field?’”

Maureen Blaney Flietner is an award-winning freelance writer living in Wisconsin.

Photo credits: poba/E+ via Getty Images, KrulUA/iStock via Getty Images Plus; Nathan Bolig, courtesy of Nathan Bolig; Krystie Reagan, courtesy of Krystie Reagan; Morsa Images/E+ via Getty Images, KrulUA/iStock via Getty Images Plus; Jamie Perkins, courtesy of Jamie Perkins; LifestyleVisuals/E+ via Getty Images, KrulUA/iStock via Getty Images Plus; Yao-Yi Chiang (photo by Lois Park)