Real data-based personalization of an automatic glucose control system

This paper overviews the application of machine learning and data analysis methods in medicine. The problem of constructing a closed personalized automatic control system for blood glucose level is considered. Such a system focuses on a particular patient and involves glucose level measurements in the interstitial space by a sensor. We describe a modification of the glucose level regulation model for the blood of a patient during the intake of glucose with meals and the supply of exogenous insulin into the bloodstream. Also, we propose an isolating search method for a group of personalized model parameters to be identified individually. As an example, model parameters are identified for a patient with type 1 diabetes based on real data, and the optimal PD control law of exogenous insulin supply is applied in the identified model. The result is compared with the actual glycemic curve after a single administration of insulin to the patient as recommended by a physician. As shown, the optimal PD control law effectively stabilizes blood glucose level to avoid the development of hypoglycemia. The results of this paper can be used to design automatic glucose control systems for humans (insulin pumps).