USask researcher probing AI’s potential for health care

“I think there’s a disconnect from the health care world about what’s really happening, and what is possible,” said Lee, an associate professor in the Department of Medicine within the University of Saskatchewan’s (USask) College of Medicine, based in Regina. As an infectious disease specialist with an interest in machine learning, Lee is anticipating major, rapid changes to his field.

“I used to be of the opinion that health care would be one of the last fields touched by machine learning, but in fact I think it’s going to be one of the first.”

Lee recently published an article in JAMIA Open based on his work developing a chest X-ray machine learning model – using consumer-grade equipment he purchased at a local computer store. He received a College of Medicine Research Award (CoMRAD) to support similar work in the past.

His intent is to show how accessible it’s becoming to create this type of AI-guided diagnostic tool.

“This was basically showing the medical community how accessible this is,” he said.

Lee is perfectly poised to embrace the coming changes. After finishing his infectious diseases fellowship, he completed a Master of Science degree in health and clinical informatics, followed by specialization certificates in machine learning. During his undergraduate studies Lee focused on neuroscience, which is the basis for how the original “neural networks” were designed.

“Neural networks are really fascinating because they’re really modelled after the human brain,” he explained. In Lee’s study, instead of coding a machine to understand the X-rays, he created a convolutional neural network (CNN) to review the images. Loosely speaking, this is an artificial construct of the human brain.

Lee then fed a set of data into the neural network, and the machine learned to understand it, identifying chest X-rays as normal or abnormal.

While his creation of this type of model was not unique, what made the study interesting was the use of consumer grade hardware alongside public datasets and using methods to try and determine how the AI was thinking. This is important in health care as intergovernmental bodies such as the World Health Organization demand AI be explainable within a health care context.

There are significant barriers in the development of machine learning in health care, explains Lee. Unless universities and public institutions improve access to information and the large datasets needed to train models, those with the most resources stand to control the space.

“I think it’s very important that we have a voice in this, because as health-care providers, we have a responsibility to make sure AI gets deployed in a safe, fair, meaningful, and ultimately beneficial way for humankind,” he said.

“We need to be funding more research, and creating more educational opportunities to get people interested in this space. If you don’t understand it, then you know you’re not going to be able to cope with it and integrate it. There is a possibility of untold benefits for patients and health care but also a possibility of harm. How this plays out is still to be determined.”

Despite the limited tools and data available to him, Lee is determined to contribute to the advancement of AI in health care. He is optimistic about the potential of AI to improve his field, such as by helping to alleviate burnout, one of the biggest problems he sees working as a specialist.

“You’re so busy that you don’t have time to dream, you don’t have time to heal. You’re just trying to get the day done because there’s so many things waiting for you,” he said.

“Maybe machines can replace us in doing that, and you can become the dreamer again. You can think about how to make care better, you can become the healer again.”