Committed moving horizon estimation for meal detection and estimation in type 1 diabetes

We introduce a model-based meal detection and estimation method for the treatment of type 1 diabetes that automatically detects the occurrence and estimates the amount of carbohydrate (CHO) intake from continuous glucose monitor (CGM) data. Meal detection and estimation play a critical role in closed-loop insulin control by enabling automatic regulation of post-meal insulin dosing in artificial pancreas systems without manual meal announcements by the patient. Our approach to meal detection is based on a novel technique we call Committed Moving Horizon Estimation (CMHE), an extension of Moving Horizon Estimation (MHE). While MHE alone is not well-suited for disturbance estimation and meal detection, CMHE aggregates the meal disturbances estimated by multiple MHE instances to balance future and past information at decision time, thus providing timely detection and accurate estimation. We evaluated our CMHE-based meal detection and estimation method in-silico, using a nonlinear ODE gluco-regulatory model and random meal profiles to generate blood glucose and CGM signals. CGM data is used to detect meal occurrences and to estimate their onset, duration, and CHO amount. At the optimal operating point of the detector, we achieve an 88.5% daily detection rate and, more importantly, a 100% detection rate, with an average of 18.86 minutes onset deviation, and 70.50% CHO amount estimation accuracy for the main meals (i.e., excluding snacks).