Artificial Intelligence and Diabetes Care
Three papers recently accepted for publication all stem from research led by two faculty members of the biomedical engineering program, Cindy Marling and Frank Schwartz.
Marling, associate professor of electrical engineering and computer science, and Schwartz, professor of endocrinology in the College of Osteopathic Medicine (COM), are working to design a system that automates the analysis of Type 1 diabetes patients' blood glucose levels and lifestyle data and, using case-based reasoning, makes treatment recommendations.
In collaboration with Jay Shubrook, assistant professor of family medicine (COM) and a trained diabetologist, this team conducted a six-week study of 20 patients on insulin pump therapy to develop this software. This study resulted in papers that will appear in the Journal of Diabetes Science and Technology1, Computational Intelligence2, and the proceedings of the European Conference on Case-Based Reasoning3. Their work, blending Marling's expertise in artificial intelligence with Schwartz's and Shubrook's clinical expertise in endocrinology, is emblematic of the interdisciplinary nature of the program and that of biomedical engineering in general.
The writing process included valuable input from all three investigators. Schwartz and Shubrook know what physicians will be looking for; Marling knows what artificial intelligence scientists need to learn.
"That's what's fun about this," Schwartz said. "We're totally dependent on each other."
"AI is always that way," Marling said. "Because what you're doing in AI is you're trying to understand how a human being thinks and make a computer think more like that human being….I'm always getting to interact with a human expert."
The goal of the research is to create software that will yield better information for patients and their doctors. Currently, blood glucose monitoring systems are available that supply patients' all-important glucose levels, but doctors are left to interpret data graphs that often look like Jackson Pollack paintings.
"What Frank calls it is a Rorschach (inkblot test)," Marling said. "Once in a while you'll get someone who says, ‘Oh, this makes sense to me.' But most people look at it and throw their hands up….Every single data point in there is a blood glucose data point, but none of the data points tell you what the patient's doing."
Events in patients' lives can have a great impact on their blood glucose levels. "Is your mother-in-law visiting?" Marling questioned. "Are you taking a midterm exam?"
In the study, their software produced graphs that included such life events, along with the blood glucose levels.
"We saw that what the program would pick up in a person over a six-week period was just fantastic," Schwartz said, "because what it does is it essentially gives you a frequency distribution of the major problems the patient had, so you can automatically sort out frequency versus priority in terms of severity risk for the patient….So for me as a practicing physician, it was just such a neat tool."
The information is helpful, too, for the patients.
One patient in the study was surprised to learn of the connection between the timing of his eating and his blood glucose levels. "He really didn't have any idea, because if he's only seeing the data from sticking his finger four times a day, he's missing it," Schwartz said. "So, when we could show him this, then he believed."
Marling and Schwartz envision their software growing to cover ever more problems, but the solutions it offers will be physician specific and patient specific. "We're hoping the software we are developing will be creative and flexible enough to actually learn the particular patient's problems," Schwartz said.
"That's why we're using case-based reasoning," Marling said, "because it's experiential rather than algorithmic programming. The patient-specific experiences are in the system, and the physician-specific experiences are in the system. Diabetes is a very individualized disease. Different patients react in different ways to the same environmental stimulus. And that is why this approach makes sense."
Marling refers to herself as a knowledge engineer, and she's training her graduate students to become knowledge engineers. "You're engineering knowledge in the sense that (Schwartz) has the knowledge, but it's not in a system," Marling said. "It's not captured. It's only in his brain. So, if he's asleep or if somebody needs his knowledge who's in Saudi Arabia or even in Tulsa, Oklahoma, that's a limitation. When you engineer knowledge, you get it from a person and you put it in a form that could be spread all around the world."
"In the near term," Schwartz said, "the software we are developing will require downloading to a central processor at Ohio University. But our work also provides insights into how to design the next generation of software, which could reside in a glucose monitor or insulin pump."
"We're trying to make the glucose monitors and pumps smarter," Marling said.
For more information on this group's work, see:
- "Use of Case-Based Reasoning to Enhance Intensive Management of Patients on Insulin Pump Therapy."
- "Towards Case-Based Reasoning for Diabetes Management: A Preliminary Clinical Study and Decision Support System Prototype."
- "Case-Based Decision Support for Patients with Type 1 Diabetes on Insulin Pump Therapy."