PATIENT-SPECIFIC CONTROLLER FOR AN IMPLANTABLE ARTIFICIAL PANCREAS

Abstract

The aim of the artificial pancreas is to improve the quality of diabetics' lives. An implantable artificial pancreas employing intravenous glucose monitoring and insulin delivery is proposed. Model Predictive Control can be extended for hypoglycaemia avoidance and take constraints into account. Patient specific parameters of the Minimal Model were estimated using CGMS glucose readings from 11 free-living patients. Scalable micropump units were designed to reduce flow rate to meet insulin delivery requirements. The insulin delivery subsystem was tested under various reservoir fill levels and pump settings. Optimization in MPC can be solved using gradient-based methods or search heuristics. A hybrid optimization scheme comprising of Primal-Dual-Interior-Point and a modified Ant System search heuristic is used for MPC. Simulation of the optimization scheme showed promising results. In vivo experiment on a porcine model was conducted with an implantable artificial pancreas prototype. Normoglycaemia was reached in 100 minutes of peak glucose levels.