UC Davis Researchers Exploring Data and AI Tools for Animal Health Diagnosis and Treatment

By Neelanjana Gautam

The development and use of artificial intelligence (AI) is rapidly growing and offering new possibilities, as well as concerns, in how it may shape the future of industries and interactions. AI is already used in many everyday digital applications like voice assistants, travel apps and ecommerce, but applications in other areas like healthcare are also emerging quickly.

The rise of AI based technology may play an important role in human healthcare from diagnostics to treatment. Using a data-driven approach, AI may be able to help doctors analyze and assess diseases more efficiently.

Researchers at the University of California, Davis, are now exploring ways to use AI for the benefit of animal health.

Stefan Keller, an assistant professor and pathologist in the School of Veterinary Medicine, specializes in diagnosing disease in various animal species. A large part of his research is directed toward how to improve outcomes by using data to make more informed diagnostic decisions in healthcare. Toward that end, Keller, with colleagues from the Artificial Intelligence in Veterinary Medicine Interest Group at UC Davis, is exploring ways to use AI in three different projects while providing insight to address some of the apprehensions associated with it.

Identifying Patterns with AI Tools

Keller and team are developing a machine learning algorithm (called a “classifier”) that uses historical patient data to reduce errors in the interpretation of blood tests and avoid inaccurate diagnosis.

The project –– funded by the UC Davis Venture Catalyst Science Translation and Innovative Research (STAIR^TM) grant program in 2022 –– is focused on identifying disease patterns in animals using laboratory data and AI tools, and helping clinicians consider different diagnoses to act upon. Typically, when pets come to a veterinarian for a physical exam, standard tests are conducted including blood work. “We have thousands of these blood test data from over the past decades; if we take all that and run it through our algorithm, we can predict what disease the pets might have and what the prognosis might be,” Keller said. While the initial application is for dogs, the team sees an opportunity to adapt the tool for other species as well, including cats and horses.

Standardizing the Assessment of Inflammation

Being a pathologist, Keller’s diagnostic work involves, for the most part, looking at tissues through the microscope. In his second project, the team is investigating inflammatory bowel disease in aged cats to assess the level of inflammation. “In the current system using microscopes, there’s a discrepancy in how pathologists assess or grade inflammation,” said Keller.

Keller sees an opportunity to standardize the assessment of inflammation and share it with clinicians by digitizing the data and running AI algorithms on it. “What we’re hoping to do with this project is to train the classifier to recognize different inflammatory cells, which will allow us to standardize the assessment of inflammation,” said Keller. “It is important because it affects how clinicians treat patients.” This research is facilitated by a graduate student fellowship for ‘Digital pathology and AI’, jointly funded by Charles River Laboratories and the UC Davis Office of Research. This fellowship provides a unique opportunity to strengthen the ties between industry and academia as well as training the next generation of scientists. And this leads to the third project of hosting the classifiers on a platform to help with clinical decision making.

Automating Patient Diagnosis in Real-Time

Currently, scientists have to manually pull the data from a patient database onto a computer, run the classifier on that data locally, and then provide the information back to the clinician. What Keller’s team is working on is to cut down that time-consuming, manual process by hosting the classifiers on a shared platform called Animal Health Analytics (ANNA) and feeding the patient data directly into it so the patterns can be detected in real-time. “What we’re trying to do is automate the process where the clinician, by just a push of a button, can run a classifier on the patient data and get the results back immediately,” Keller said.

The project is conducted in collaboration with the IT service in the School of Veterinary Medicine that involves clinicians and colleagues including Professor Krystle Reagan who has developed the classifiers. “We are at the very final stages now of deploying our platform. The platform is accessible through our patient database,” said Keller. “It’s essentially just compute infrastructure to route data from a database and apply machine learning classifiers to it and then display the data. That’s the basic idea behind it.”

Addressing Challenges in Adopting of AI in Veterinary Medicine

In terms of adoption, Keller thinks veterinary medicine presents an interesting landscape, adding that training data is more easily obtained in veterinary medicine than in human medicine and there are so far no rules around the use of AI algorithms for diagnostic use. However, the pitfalls of the technology and the attitude of users are very similar in veterinary and human medicine. “If you’ve done things the traditional way for decades, you might be reluctant to adopt an AI algorithm, and then believe in the results,” said Keller. “Transparency of the method is important but with many AI algorithms there is some black box analysis that comes into play, and that is potentially concerning for users. We have to see how we can address that.”

Keller considers thorough testing, feedback and objective analysis as crucial to make sure we are critical users of technology.