A Minimal Model for Type-1 DM Patients: Meal and Exercise Adaptation

Abstract

Mathematical Modeling of glucose-insulin dynamics of Type-1 Diabetic Mellitus (T1DM) patients is an essential component of designing and developing Artificial Pancreas Systems. These model parameters, which exhibit significant inter-patient variability, are identified for each individual T1DM patient through standard tolerance tests. However, in addition to the inter-patient variability, each patient's model parameters vary according to the circadian rhythm. Therefore, the blood glucose response to a meal is different at breakfast when compared to lunch and dinner. In addition to this, the glucose-insulin dynamics vary when the T1DM patient exercises or during any physical activity. To account for these intra-patient variabilities and the variability due to exercise, a neuro-adaptive learning scheme is proposed in this work. The uncertainties are approximated as a product of a weight and a meaningful basis function. The model uncertainties are learned during meals and idle activity, whereas exercise learning requires an announcement from the patient and is only learned when the patient is exercising. This neuroadaptive learning scheme can prove to be of vital importance in designing model-based control laws for blood glucose regulation in Type-1 Diabetic patients.