The presentation will focus on algorithmic aspects of a continuous glucose monitoring (CGM) system used for glucose estimation in real-time. The CGM system is composed of a glucose sensor and an estimation algorithm to measure blood glucose levels. The glucose sensor is composed of an electrode that is implanted in the interstitial fluid and generates a noisy electrical current in response to interstitial glucose concentrations, with a relationship which varies with sensor wear time. An estimation algorithm then infers blood glucose concentrations from the noisy electrical signal, in addition to intermittent capillary glucose measurements to account for time-variability. In this paper, we propose a novel Kalman filter based estimation algorithm through (i) noise filtering step, (ii) compartment matching step and (iii) parameter estimator step. The initial estimate and covariance of the parameter estimation step are extracted from offline dataset using a nonlinear cubature Kalman filter. The new algorithm has been compared to a standard algorithm based on minimum error variance estimator using a dataset obtained from 04 patients who were tested on 3 in-clinic days over a 7-day sensor wear period (days 1, 4, and 7). The comparison is based on mean absolute relative difference (MARD) and percentage of estimates with deviation less than 20% in reference to gold standard glucose analyzer measured glucose references on clinical days. The MARD is 8.02, 9.39 and 9.11 for the proposed algorithm and 9.50, 14.95 and 14.35 for the standard algorithm while the percentage of estimates under 20% deviation is 94.16, 91.74 and 90.53 for the proposed algorithm compared to 90.77, 73.16 and 80.08 for the standard algorithm.
Bienvenue à tous!