CONICET | Buscador de Institutos y Recursos Humanos

Recent technology breakthroughs towards a fully automated artificial pancreas give rise to the need of new monitoring tools aiming at increasing both reliability and performance of a closed-loop glycemic regulator. Based on error grid analysis (EGA), an insightful monitoring tool is proposed to assess if a given closed-loop implementation respects its specification of an optimally performing glycemic regulator under uncertainty. The optimal behavior specification is obtained using Linearly Solvable Markov Decision Processes (LSMDP) whereby the Bellman optimality equation is made linear through an exponential transformation which allows obtaining the optimal control policy in an explicit form. The specification for the desired glucose dynamics is learned using Gaussian processes for state transitions in an optimally performing artificial pancreas. By means of the proposed grid, the specification is vis-à-vis compared to glucose sensor readings so that any significant deviation from the expected closed-loop performance under abnormal or faulty scenarios can be detected.