Modeling of Exercise Induced Effects on Blood Glucose Dynamics in T1DM Patients

Abstract

Keeping Blood Glucose (BG) concentrations in a healthy range is despite long ongoing research a difficult task in individuals suffering from Type 1 Diabetes Mellitus (T1DM). Especially, the effects that physical activity imposes on the BG dynamics are still one of the major challenges. While first attempts have been made to model the effects of aerobic exercise on BG. No one has yet focused on the anaerobic exercise and its unique impact on glycemic control. Although anaerobic activity is common in the general population, mostly in form of resistance exercise, and is seen to improve glycemic control, no one yet has focused on modeling its unique impact on glycemic levels. Together with the recent development of wrist-worn fitness trackers, it has become easier than ever to classify exercise type, intensity, and duration non accurately and non-invasively. Thus, this project focuses on learning models from data obtained by fitness trackers to classify exercise accurately. This is done by fusing Heart Rate (HR) data with motion data. Additionally, a stand-alone exercise model is proposed capable of capturing aerobic and anaerobic activity. The performance of this model is validated with HR data during different exercise types. The exercise model is built in a LPV-like manner and can thus be used for direct controller design or to enhance Virtual Patient Population (VPP) simulators such as the famous FDA approved UVa-Padova simulator. Lastly, the proposed exercise model is used as an extension to the well known Bergman Minimal Model and evaluated on in-silico data. It is seen to capture the opposing effects of aerobic and anaerobic exercise as widely mentioned in the literature. Thus it lays pioneering work in the field of capturing anaerobic effects. Nevertheless, the group is further working on validating the model’s performance with in-vivo data obtained from a specifically designed clinical trial.