A Grid-Based Tool for Optimal Performance Monitoring of an Artificial Pancreas

Abstract

Due to its safety-critical condition, continuous performance monitoring of an artificial pancreas (AP) is of paramount importance for both patients and health care personnel. Based on error grid analysis (EGA), a monitoring tool is proposed to assess if a given control policy implementation respects the specification of an optimally controlled glycemic system under uncertainty. The optimal behavior specification is obtained using Linearly Solvable Markov Decision Processes (LSMDP) whereby the Bellman equation is made linear through an exponential transformation such that the optimal control policy is obtained in an explicit form. The system specification is learned using Gaussian processes for state transitions in a well-performing glucose regulator.