Ohio University computer science and medical faculty pioneer new blood glucose prediction processes
By Joe Barbaree and Colleen Carow
ATHENS, Ohio (Jan. 30, 2012) – An interdisciplinary team of biomedical engineering researchers from Ohio University’s Fritz J. and Dolores H. Russ College of Engineering and Technology and Heritage College of Osteopathic Medicine received a $350,000 grant from the National Science Foundation (NSF) for their work on predicting blood glucose levels for Type 1 diabetes patients.
Led by Associate Professor of Computer Science Cindy Marling with colleagues Assistant Professor of Computer Science Razvan Bunescu, Professor of Endocrinology Frank Schwartz and Associate Professor of Family Medicine Jay Shubrook, the project “Machine Learning Models for Blood Glucose Prediction in Diabetes Management” aims to use artificial intelligence to predict blood glucose levels.
A person's blood glucose levels naturally fluctuate throughout the day, whether or not they have diabetes. However, the fluctuation is greater in people with diabetes, who must be careful not to let their blood glucose levels go too high or too low.
Sustained high blood glucose levels can lead to blindness, heart attacks and stroke, while low ones cause weakness, confusion, dizziness, sweating, shaking and, if not treated in time, seizures, coma – or death.
Marling explained that being able to predict blood glucose levels means that the patient can act before problems occur. “Our team’s research goal is to predict blood glucose levels 30 to 60 minutes in advance -- plenty of time for patients to take preventative action.”
In order to predict blood glucose levels, the team is building regression models specific to each patient using support vector regression (SVR) – more commonly used to predict stock market prices and utility loads. The team is the first to use the overall prediction process for the medical profession with human subjects.
Regression models will be “trained” for each patient using current and past data about their blood glucose levels, insulin dosages, food intake and exercise. Such training requires continuously examining past data, making preliminary predictions and then using errors in the predictions to refine the model to make it more accurate.
Funded through the National Science Foundation’s Smart Health and Wellbeing program, the project forms the cornerstone of a new SmartHealth Lab in the Russ College, which will facilitate additional Smart Health and Wellbeing research.
Marling noted that collaboration among disciplines is the key for this type of research.
“No single discipline has all of the answers,” she said.
Schwartz explained this is why he approached Marling years ago. “I had seen software that could merge music tracks and thought, ‘We could use software to analyze blood glucose patterns for diabetes management,’” explained Schwartz.
Marling recalled that Schwartz needed someone who could make sense of a vast amount of data.
Schwartz agreed. “The cool thing is that you need each other to ask the right questions before you can even begin to find answers,” added Schwartz.
The researchers have been performing extensive research with patients since 2004, thanks to three internally funded studies, gathering data and testing artificial intelligence software for diabetes management.
The new grant will provide funding for the next three years, but the team’s plans run for a longer term.
“We’ll work on it until we retire, and we’re not retiring any time soon,” Marling said.