Accuracy continuous glucose monitoring method, system, and device

1 . A method for improving the accuracy of a continuous glucose monitoring system (CGM) comprising:

obtaining an output CGM signal at a time t from a CGM device;

removing, substantially in real time, random noise from said CGM signal by estimating a value of the CGM signal at a time t+PH using a noise state estimation, wherein:

PH is a real-time short-time prediction horizon selected based, at least in part, on blood glucose-to-interstitial glucose (BG-to-IG) kinetics; and

the noise state estimation is based on a noise-estimation algorithm developed in a stochastic context and implemented using a Kalman filter, wherein one or more unknown parameters within the Kalman filter are estimated by using a Bayesian smoothing criterion;

substituting in real time the estimated value of the CGM signal at time t+PH for the output CGM signal at time t;

using the estimated value of the CGM signal at time t+PH as the true value of the CGM signal at time t.

2 . The method according to claim 1 , comprising:

using a prediction horizon PH of less than 20 minutes.

3 . The method according to claim 2 , comprising:

compensating part of a delay introduced by low-pass characteristic of a blood glucose-to-interstitial (BG-to-IG) kinetic system.

4 . The method according to claim 1 , comprising:

compensating part of a delay introduced by a low-pass characteristic of a blood glucose-to-interstitial (BG-to-IG) kinetic system.

5 . The method according to claim 1 , comprising:

substituting a current CGM value given in output by a CGM sensor at time t, named CGM (t), with the glucose concentration predicted by the noise-estimation algorithm PH minutes ahead in time.

6 . The method according to claim 1 , wherein the PH is a value that matches that of a diffusion constant of blood glucose-to-interstitial glucose (BG-to-IG) kinetics for the CGM device, wherein the estimated value of the CGM signal at time t+PH is a closer approximation of blood glucose than the output CGM signal at time t.

7 . The method according to claim 1 , wherein the one or more unknown parameters are variance of the process and/or measurement noise.

8 . A method for improving the accuracy of a continuous glucose monitoring (CGM) sensor comprising:

improving accuracy of CGM readings by reducing random noise and calibration errors in said readings using a real-time short-time prediction horizon (PH); and

denoising said CGM readings by using a Kalman filter (KF) coupled with a Bayesian smoothing criterion for the estimation of its unknown parameters, wherein the estimation of unknown parameters via the KF provides the short-time prediction;

wherein the PH is a value that matches that of a diffusion constant of blood glucose-to-interstitial glucose (BG-to-IG) kinetics for the CGM sensor, wherein an estimated value of a CGM signal at time t+PH is a closer approximation of blood glucose than an output CGM signal at time t.

9 . A system for improving the accuracy of a continuous glucose monitoring sensor comprising:

a digital processor;

a continuous glucose monitoring (CGM) sensor in communication with the digital processor, the continuous glucose monitoring (CGM) sensor configured to generate a glucose signal; and

a denoising module, configured to receive the glucose signal from the continuous glucose monitoring (CGM) sensor, and generate an improved accuracy CGM signal by reducing random noise and calibration errors using a real-time short-time glucose prediction horizon (PH) to estimate the real time denoised value of the CGM signal based on a noise-estimation algorithm developed in a stochastic context and implemented using a Kalman filter;

wherein the PH is a value that matches that of a diffusion constant of blood glucose-to-interstitial glucose (BG-to-IG) kinetics for the CGM sensor, wherein an estimated value of a CGM signal at time t+PH is a closer approximation of blood glucose than an output CGM signal at time t.

10 . The system according to claim 9 , wherein the denoising module is configured to use a prediction horizon PH of less than 20 minutes.

11 . The system according to claim 9 , wherein the denoising module is configured to compensate part of a delay introduced by a low-pass characteristic of a blood glucose-to-interstitial (BG-to-IG) kinetic system.

12 . The system according to claim 9 , wherein the denoising module is configured to substitute a current CGM value given in output by a CGM sensor at time t, named CGM (t), with the glucose concentration predicted by the noise-estimation algorithm PH minutes ahead in time.

13 . The system according to claim 12 , wherein a Bayesian smoothing criterion is coupled with the Kalman Filter for estimation of unknown parameters.

14 . The system according to claim 12 , wherein the noise-estimation algorithm uses CGM data only for real-time application.

15 . The system according to claim 9 , wherein a Bayesian smoothing criterion is coupled with the Kalman Filter for estimation of unknown parameters.

16 . The system according to claim 9 , wherein the noise-estimation algorithm uses CGM data only for real-time application.

17 . A method of increasing accuracy of a continuous glucose monitoring (CGM) sensor signal measured at a time t, by using a short-time prediction horizon (PH) to estimate a value of a CGM signal at time t+PH by applying CGM data obtained at time t to a Kalman filter for time t+PH, and substituting the estimated value of CGM signal at time t+PH for the real time CGM signal value obtained at time t, wherein the estimated value of CGM signal is determined using a noise state estimation based on a noise-estimation algorithm developed in a stochastic context, wherein the PH is a value that matches that of a diffusion constant of blood glucose-to-interstitial glucose (BG-to-IG) kinetics for the CGM sensor, wherein an estimated value of a CGM signal at time t+PH is a closer approximation of blood glucose than an output CGM signal at time t.