"Smart" continuous glucose monitoring sensors: on-line signal processing issues

Abstract

The availability of continuous glucose monitoring (CGM) sensors allows development of new strategies for the treatment of diabetes. In particular, from an on-line perspective, CGM sensors can become "smart" by providing them with algorithms able to generate alerts when glucose concentration is predicted to exceed the normal range thresholds. To do so, at least four important aspects have to be considered and dealt with on-line. First, the CGM data must be accurately calibrated. Then, CGM data need to be filtered in order to enhance their signal-to-noise ratio (SNR). Thirdly, predictions of future glucose concentration should be generated with suitable modeling methodologies. Finally, generation of alerts should be done by minimizing the risk of detecting false and missing true events. For these four challenges, several techniques, with various degrees of sophistication, have been proposed in the literature and are critically reviewed in this paper.

Keywords: calibration; diabetes; filtering; model; prediction; time-series.

Figure 1.

Representative type 1 diabetic subject. Top: BG references (stars) vs. original FreeStyle Navigator® CGM profile (continuous line). Bottom: BG references (stars) vs. CGM profile (blue line) profile recalibrated by the method of King et al. [32].

Figure 2.

Representative type 1 diabetic subject (data taken from Kovatchev et al. [30]). Top: measured CGM (blue line), SMBG samples used for recalibration (copper circles), recalibrated CGM (green line), and other available SMBG samples plotted as reference (asterisks). Bottom: estimated deviation of the sensor gain from the unit value.

Figure 3.

Two representative type 1 diabetic CGM time series. Top: FreeStyle Navigator® time series (1 min sampling), taken from Kovatchev et al. [30]. Bottom: GlucoDay® time series (3 min sampling) (taken from Maran et al. [44]).

Figure 4.

Same data as in Figure 3: original profile (blue) and outcome (green) of a MA filter with M = 5 and μ = 1.

Figure 4.

Same data as in Figure 3: original profile (blue) and outcome (green) of a MA filter with M = 5 and μ = 1.

Figure 5.

Application of the Kalman filtering method of Facchinetti et al. [43] to two FreeStyle Navigator® time-series exhibiting a different SNR (the green line is the filter output). Top panel: same time-series as in the top panel of Figures 3 and 4. Bottom panel: same time-series as in the top panel of Figure 1.

Figure 6.

FreeStyle Navigator® time-series in a type 1 diabetic subject taken from Kovatchev et al. [30]. Original (blue line) vs. predicted (green line) CGM time-series by using an AR model of order 1 with PH = 30 and μ = 0.85 (too small) and μ = 0.99 (too large).

Figure 7.

Same data of Figure 6: Original (blue) vs. predicted (green) profiles. Top panel: μ = 0.95 and PH = 30. Bottom panel: same μ as in top panel, but PH = 60.

Figure 8.

Risk of generating a false hypo-alert from the original CGM profile (blue line) at time 19.2 mitigated by employing an on-line filtered profile (green line) together with its confidence interval (shaded area).