Dr. Robot: Like It or Not, AI Is Making Inroads into Veterinary Medicine

While concerns have been expressed about how artificial intelligence will affect the future of mankind, this technology is increasingly finding its way into our daily lives—and veterinary medicine is no exception. This article looks at several technologies that are harnessing the power of math and machines to change how work is performed.

by Maureen Blaney Flietner

Mention artificial intelligence (AI) and some might think of the film Terminator or TV’s Star Trek. However, many people today no longer see AI as futuristic sci-fi but rather part of their daily lives.

From online shopping recommendations to mobile banking apps to voice-enabled personal assistants like Alexa and Siri, AI is here.

Now AI is increasingly found in veterinary medicine.

What Is AI?

The term artificial intelligence was first coined in the 1950s. Today, it means any computer or machine that seems to display intelligence, according to John Craig, PhD, vice president of research and development for EponaMind in Creston, California. But you can’t read articles today about AI without coming across such terms as machine learning, neural networks, and deep learning.

AI is the broadest category, explained Craig, while machine learning is a subset that includes those techniques in which the computer “learns” by examining data. Deep learning, he said, is a subset of machine learning and refers to techniques that use neural networks arranged in layers. A neural network is a software program that uses “artificial neurons”—modeled after actual brain neurons—developed to “learn.” When neural networks are arranged in many layers to make a “deep” network, there is “deep learning.”

EponaMind offers Metron-DVM, a veterinary imaging product that uses the power of neural networks trained with deep learning techniques to process veterinary radiographs. Other companies, Craig said, might use deep learning for voice recognition, speech processing, text reading, and more.

Since “data is king,” said Craig, much of the work in making deep learning practical is in (a) knowing what problems to solve, (b) knowing what data will be required for those problems, and (c) cleaning the data and ensuring it is of high quality to avoid “garbage in.” Data cleaning, he explained, means that an expert human must curate all the data that will be used for the neural network to learn from so that labeling mistakes won’t confuse the neural network and reduce the quality of its output.

Overwhelmed? Think of AI as “augmented” rather than artificial intelligence, offered Mark Stephenson, DVM, chief veterinary officer of LifeLearn Animal Health in Guelph, Ontario, Canada. AI allows access to the most relevant information more quickly, thus creating an augmentation of the skills and services provided to clients, he explained.

AI-Powered Offerings

How is AI enhancing veterinary medicine? Consider these examples.

Urine sediment analysis is a powerful diagnostic tool, but it can be time consuming and requires experience and well-developed interpretation skills.

The SediVue Dx analyzer by IDEXX uses proprietary veterinary-specific algorithms similar to what is used in facial recognition technology, according to Tina Hunt, PhD, IDEXX corporate vice president and general manager for VetLab, Diagnostic Imaging, and Telemedicine.

“Machine learning is an essential component. The image library holds a broad spectrum of sediment examples from users worldwide. When veterinarians perform an analysis, those images are aggregated. Each analysis essentially helps the analyzer ‘get smarter’ over time,” she explained.

Natalie L. Marks, DVM, CVJ, medical director of AAHA-accredited Blum Animal Hospital in Chicago, Illinois, has used SediVue Dx for almost two years.

Doing the same thing over and over in the analysis process can be tedious and does not involve skills as much as going through the motions, she said. Technicians now get more information more efficiently, and new technicians benefit from viewing the images.

Results within five minutes allow for better diagnostics that are in the moment, said Marks. The technicians still double check, but with the patient still there, they feel more actively involved than before. Clients also appreciate the efficiency and convenience.

IDEXX’s latest Neural Network 4.0 software expanded the analyzer’s diagnostic capabilities to provide the analysis experience equivalent to 1,000 veterinary lifetime careers, said Hunt. The update included two new parameters—ammonium biurate and bilirubin—that aid in assessing liver function and blood disorders in critical care situations.

Helpful for Detecting Kidney Disease

A diagnostic tool that uses AI, machine learning, and data from 150,000 cats, RenalTech by Antech Diagnostics now provides the opportunity to obtain predictions with greater than 95% accuracy of whether or not a cat will develop chronic kidney disease (CKD) within two years.

As a practice making early use of RenalTech, AAHA-accredited Summit Boulevard Animal Hospital in West Palm Beach, Fla., has embraced the tool as another way to be proactive in caring for its patients, according to owner Iyampillai Arun, DVM. While clients are happy when they receive negative test results, he said, those with positive results can take steps to delay the onset of disease with more frequent visits and emphasis on oral health, renal-friendly diets, and caring for any underlying conditions. “AI is going to be a mainstay of medicine in the future. When clients hear that we can do this, it gives them another incentive to check their cats. Since we had a patient that was in renal failure at only two years of age, we recommend testing as early as one year. After all,” said Arun, “we don’t believe that only old people get sick.”

Getting quickly to the bottom of challenging cases is important for veterinarians.

LifeLearn’s Sofie, a veterinary medical search tool, helps them do just that. Launched formally in February 2018, Sofie harnesses the power of IBM’s Watson—a question-answering computing system that made headlines in 2011 by defeating two human competitors on the quiz show Jeopardy! It works through natural language processing, which is much different from keyword search, explained Stephenson.

Mark Stephenson, DVM, chief veterinary officer of LifeLearn Animal Health in Guelph, Ontario, Canada, uses Sofie on a tablet.

“Sofie is able to learn through ongoing use by veterinarians, by upgrades to software functionality, and by the addition or updating of new content a few times annually,” he said. This fall, for example, exotics (birds, reptiles, and small animals) will be added, involving extensive new training in areas of existing materials that were not included previously and three new textbooks dedicated to those animals.

Robin Downing, DVM, MS, DAAPM, DACVSMR, CVPP, CCRP, director of AAHA-accredited Windsor Veterinary Clinic in Windsor, Colorado, started using Sofie a few years ago during beta testing.

She found it attractive for several reasons, including:

  • Quick, easy access to multiple current veterinary references
  • Use of everyday language
  • Easy-to-find information after initial query
  • Speedy look-up process
  • Access to journals for which the hospital does not have a subscription
  • Access to meeting proceedings to dive deeper into a topic

“Recently, I saw a dog who probably has diabetes insipidus. I needed to remind myself about differentiating this illness from other potential explanations for increased drinking and urinating, and I needed to fine-tune my list of differential diagnoses quickly. I appreciated the fact that, while my client was still in the exam room, I could use Sofie to skim through multiple sources and better formulate both my diagnostic plan and my presumptive treatment plan,” explained Downing.

Downing said she uses Sofie to help fill gaps in her memory about diseases and treatments and to flesh out thoughts about specific patients—all within minutes without navigating multiple websites.

At AAHA-accredited Headon Forest Animal Hospital in Burlington, Ontario, Canada, co-owner Patricia Murphy, DVM, also uses Sofie. An example? A canine patient had an aural hematoma and she and her colleagues had differing opinions on best options.

“Sofie’s language interface did take a little getting used to after using VIN for so many years. It is very easy to use, clean, and simple. Sofie is all published information versus opinions from other veterinarians. With VIN, you need to make sure the person answering the question has recognizable credentials,” she noted.
Diagnostic images can be difficult and time consuming to evaluate.

Vetel Diagnostics, the US distributor for Metron-DVM, uses its deep learning to offer an automated measurement of the vertebral heart score that can assess a routine radiograph and bring additional diagnostic content in seconds, according to James Waldsmith, DVM, president of Vetel Diagnostics in San Luis Obispo, California.

By assessing the normality of the patient’s heart size and comparing it against a database of breed-specific measurements, it can alert the veterinarian to the possibility of subclinical heart disease.

In its newest product, Vetel Diagnostics offers assessment measurements of the canine pelvis. With all of this done quickly, the veterinarian experiences greater efficiency in the diagnostic process, more informed patient assessment, and better clinical outcomes, Waldsmith noted.

Veterinary hospitals without after-hours care leave clients to choose between a possibly unnecessary or expensive visit to a 24/7 emergency hospital and seeking information on the internet.

In May in the United Kingdom, Vet-AI launched an AI app called Joii. Free to download, Joii offers a free symptom checker and the option of a video consultation with a veterinarian. The company plans to expand Joii’s availability in a year or so with North America a key target because there are no language barriers and American pet parents view their pets similarly to those in the United Kingdom, explained cofounder Robert Dawson, BVMS, MRCVS.

Dawson said that algorithms beneath the symptom checker allow pet owners to determine their best course of action—often directing them into a practice if the pet is likely to need a physical examination.

He said the AI is being trained to recognize images, videos, and other input that, in time, will allow some scoring to be automated. Until the AI has been trained with large amounts of data and it can demonstrate clear statistical accuracy, it won’t play any part in helping with diagnosis.

These photos show screenshot examples of Vetel Diagnostics’ automated measurements of radiographs.

How Software “Learns”

James Waldsmith, DVM, president of Vetel Diagnostics in San Luis Obispo, California, explained the basics of how intelligent medical imaging software “learns.”

The neural network is taught to recognize objects—images, in the case of Vetel Diagnostics—and then define each image so it can be assigned to a group. A simple example of a category with six levels of depth would be as follows: Radiograph>Canine>Thorax>Lateral>Right side>Normal.

  • The software recognizes the image.
  • It learns to separate “normal” from “not normal.”
  • As the database of “not normal” grows, each “not normal” gets a name.
  • The neural network is taught to separate “not normals” into their own categories of pathological condition/diagnosis.
  • After the diagnosis is applied, a degree of severity is assigned.
  • The software then searches the literature for treatment alternatives, efficacy rates, and so on.

Source: Super Genes: Unlock the Astonishing Power of Your DNA for Optimum Health and Well-Being, by Deepak Chopra, MD, and Rudolph E. Tanzi, PhD

“Our view is that we now live in an age where owners have access to online care in a way that simply wasn’t available at the time that regulations were drawn up, and the profession is struggling to keep up with the changes in tech,” said Dawson. “Clearly there are conditions that do require a physical examination to make a diagnosis, but there are many where this isn’t the case or where the long-term management of a condition can be performed as well or often better remotely, where we get to see the pets acting normally in their home environment.”

GuardianVets, based in Chicago, Illinois, provides a different after-hours option offering a triage and answering service operated by licensed veterinarians and technicians for client veterinary hospitals, according to John Dillon, founder and chief executive officer. The team schedules visits for cases that can wait, provides advice after hours, and directs pet owners to come in for emergency care if necessary. It does not establish veterinarian–client–patient relationships (VCPRs).

Triage members have access to protocol guides that incorporate machine learning to curate questions based on its database of consultations, explained Dillon.

“The goal of our software is to be comprehensive in triage. We believe technology can be a tool to supplement veterinary care but not replace the need for a physical examination in establishing a VCPR, nor can it be used standalone as a substitute for a licensed professional in providing triage.”

These photos show screenshots of the Joii app.

AI Challenges and Concerns

Dillon has reservations about AI, particularly about the technology behind chatbot veterinary consultations.

“We consider machine learning a ‘nice-to-have’ tool to supplement the quality of GuardianVets’ triage but not currently a ‘must have.’ Fortune 500 companies already try to implement intelligent interactive voice response systems and their website chatbots claim to use sophisticated AI, but the experience can be suboptimal. The algorithms are very linear, whereas human communication and many pet symptoms are not. Amplify that problem with a pet owner trying to interpret what is going on with a pet who doesn’t speak . . . it’s not a risk we are willing to take in an emergency triage setting.”

“We don’t consider our licensed [technicians] and DVMs an operational inefficiency to eventually be programmed away,” he said. “To the contrary, we double down and invest in their training, feedback, and professional development. Talented, empathetic licensed professionals are a critical part of building trust with clients.”

Dillon is not alone in his concerns. In the broader world context of AI, red flags have been raised about lapses and biases in algorithms and more.

Craig suggested that veterinarians keep an eye on AI and learn about it.

“What is the alternative? Lobby the government to outlaw AI? Then the rest of the world develops it, and we don’t,” Craig said. “There is no avoiding what comes in the future.”

A New Game in Town

“So many media articles jump right away to fears of ‘automatic diagnosis,’ but there are so many other things to do for the veterinarian—such as work flow and ease of use—that they won’t mind handing over to a machine,” said Craig.

Hunt compared concerns about AI-enhanced products to the time before the widespread in-house adoption of hematology analyzers in North America when blood was always analyzed under the microscope.

Introducing analyzers didn’t replace technicians, she said; it only allowed practices to get more efficient and higher-quality results.

“AI is a new game,” said Waldsmith, “and some of the rules of how it will be used have yet to be developed. An AI program is only as good as the person(s) [who] wrote the code of the deep learning engine, the quality of the databases that were used to train the neural network, and design of the software in how it is used by the practitioner. If you accept the premise that time is the only thing there is no more of, then one should embrace proper AI solutions as the vehicle to a life in which we can accomplish more.”

Robert Dawson, BVMS, MRCVS, cofounder of Vet-AI (photo courtesy of Vet-AI).

John Dillon, chief executive officer and founder of Guardian Vets (photo courtesy of Guardian Vets).

Robin Downing. DVM, MS, DAAPM, DACVSMR, CVPP, CCRP, director of the AAHA-accredited Windsor Veterinary Clinic, with Tommy (photo courtesy of Robin Downing).

Tina Hunt, PhD, IDEXX corporate vice president and general manager for VetLab, Diagnostic Imaging, and Telemedicine (photo courtesy of IDEXX).

Natalie L. Marks, DVM, CVJ, medical director of the AAHA-accredited Blum Animal Hospital in Chicago (photo courtesy of Natalie Marks).

Patricia Murphy, DVM, and Huey (photo courtesy of LifeLearn).

Mark Stephenson, DVM, chief veterinary officer of LifeLearn Animal Health in Guelph, Ontario, Canada (photo courtesy of LifeLearn).

James Waldsmith, DVM, president of Vetel Diagnostics in San Luis Obispo, California (photo courtesy of Vetel Diagnostics).These photos show screenshots of the Joii app.

Maureen Blaney Flietner is an award-winning writer and longtime Trends contributor living in Wisconsin.

 

Photo credits: ©iStock.com/Ekkasit919, photo courtesy of LifeLearn, photos courtesy of Vetel Diagnostics, photos courtesy of Joii

Advertisement

Close

Subscribe to NEWStat