Automating Information Collection for Diabetes Management

Asim Smailagic, Human Computer Interaction
Dan Siewiorek, Human Computer interaction

Our hypothesis is that near automatic data acquisition to support management of diabetes is possible. The proposed system will gather information on exercise (calories burned), food intake (calories consumed) and readings of sugar concentration. We are developing an image-recognition system to determine a diabetes patient’s calories consumption. Our monitoring platform will be a camera-equipped cell phone augmented with accelerometers. We will use a picture of a meal to determine the food and serving size, derive the calories consumed from the image, and create a diabetic management system that records almost all data automatically. Unsupervised machine learning algorithms will monitor accelerometer data to determine type and intensity of user activity. For food caloric intake determination, we will detect food items in an image and classify food items by comparison with training samples.