Artificial Pancreas under a Zone Model Predictive Control based on Gaussian Process models: toward the personalization of the closed loop

This work introduces a novel zone model predictive control (MPC) based on Gaussian Process models (GPs) for the artificial pancreas (AP). The main novelty of the proposal is to exploit a GP that is trained on previously collected metabolic data of type 1 diabetes mellitus (T1DM) patients, to regulate the blood glucose levels by means of a personalized MPC strategy that automatically adjusts the basal insulin and the insulin boluses to be injected to the patients. The average closed-loop performance is improved in terms of classical indexes such as time in range, avoidance of critic hypoglycaemia episodes and avoidance of long-term hyperglycaemia events. The controller was evaluated in-silico by means of the FDA-accepted UVA/Padova metabolic simulator on 11 adult T1DM patients, showing promising results.