Design and Implementation of a Flexible Simulation Framework for Glucose Metabolism

Abstract

The UVa/Padova research group has over the last decade developed a computer simulator (the UVa/Padova Type 1 Diabetes Metabolic Simulator (T1DMS)) that has been approved by the U.S. Food and Drug Administraion (FDA) as a substitute to animal trials for preclinical testing in Artificial Pancreas (AP) studies. As a consequence of being a commercial simulator, its implementation is unknown to the user. This limits its usability in the context of academic research. Motivated by this, the Artificial Pancreas Trondheim (APT) research group has implemented the APT-simulator. Its goal is to resemble the UVa/Padova T1DMS in means of equations, parameters and inputs. The main contribution of this thesis is an iteration in the software development life cycle of the APT-simulator. Based on an evaluation of the existing implementation, a new design that emphasizes maintainability and modularity has been made. The metabolic model has been updated according to the newest publication from the UVa/Padova research group (Man et al., 2014), and a harmonized interface that enables the APT-simulator to use input of the same format as the UVa/Padova T1DMS has been implemented. A framework that simplifies the process of running and comparing simulations between the APT-simulator and the UVa/Padova T1DMS was developed. The APT-simulator applies the numerical solver ode15s, available in MATLAB ® , in order to simulate the metabolic model. The second contribution of this thesis is a review of the UVa/Padova T1DMS. In the process of implementing the APT-simulator, several questions and potential flaws concerning the implementation and documentation of the UVa/Padova T1DMS arose. Simulations presented in this thesis indicate that the UVa/Padova T1DMS implements a metabolic model that is slightly different than what is published by the UVa/Padova research group. As a conclusion, simulations performed in open-loop demonstrate that the APT- simulator resembles the UVa/Padova T1DMS. Root mean square error (RMSE)- calculations prove that the simulators have a high correlation when only a basal insulin rate is injected. The RMSE is slightly higher in scenarios where an insulin bolus is injected in relation to a meal. This is assumed to be a consequence of the different equation used by the UVa/Padova T1DMS. Further testing and development of the APT-simulator is recommended in order to increase its usability.