An In-Ear PPG-Based Blood Glucose Monitor: A Proof-of-Concept Study

Abstract

Monitoring diabetes saves lives. To this end, we introduce a novel, unobtrusive, and readily deployable in-ear device for the continuous and non-invasive measurement of blood glucose levels (BGLs). The device is equipped with a low-cost commercially available pulse oximeter whose infrared wavelength (880 nm) is used for the acquisition of photoplethysmography (PPG). For rigor, we considered a full range of diabetic conditions (non-diabetic, pre-diabetic, type I diabetic, and type II diabetic). Recordings spanned nine different days, starting in the morning while fasting, up to a minimum of a two-hour period after eating a carbohydrate-rich breakfast. The BGLs from PPG were estimated using a suite of regression-based machine learning models, which were trained on characteristic features of PPG cycles pertaining to high and low BGLs. The analysis shows that, as desired, an average of 82% of the BGLs estimated from PPG lie in region A of the Clarke error grid (CEG) plot, with 100% of the estimated BGLs in the clinically acceptable CEG regions A and B. These results demonstrate the potential of the ear canal as a site for non-invasive blood glucose monitoring.

Keywords: NIR spectroscopy; blood glucose; continuous monitoring; diabetes; hearables; in-ear PPG; machine learning; non-invasive; photoplethysmography (PPG).

Figure 1

The principle of light reflectance and absorption through the different layers of the skin.

Figure 2

The in-ear PPG sensor used in our study. (a) The different components of the in-ear sensor. (b) The PPG sensor chip embedded on a viscoelastic earbud attached to an ear hook. (c) Sensor placement within the ear canal.

Figure 3

The recording protocol in this study.

Figure 4

Pre–processing steps for the raw PPG. Top to bottom: raw signal; band–pass signal with the lower envelope detected; upsampled signal with the lower envelope removed; pre–processed signal with the peaks and troughs identified.

Figure 5

Matched filtering of PPG cycles. (Top left) Exemplar template used for matched filtering of PPG cycles. (Top right) Overlay plot of all PPG cycles acquired during one recording. (Bottom left) Exemplar low–quality PPG cycles, with a correlation < 0.9 with the template. (Bottom right) Overlay plot of the remaining PPG cycles after matched filtering.

Figure 6

Averaged low–glucose (blue) and high–glucose (red) PPG cycles with the corresponding shaded areas indicating the confidence intervals. The green dots designate time instances where the null hypothesis of no statistical difference between the low–glucose and high–glucose PPG cycles is rejected.

Figure 7

Features extracted from every PPG cycle.

Figure 8

The distribution of the true glucose levels of the calibration and prediction data for each subject (S1–S4), recorded using the gold standard glucometer. Histograms (horizontal bars) show the relative number of measurements at different glucose levels (y-axis) and the box plots indicate the summary statistics of all recordings for each subject.

Figure 9

Percentage difference between the high and low glucose levels, based on the values of the ten most significant features across the four subjects. The % difference indicates the importance of a feature.

Figure 10

Clarke error grid analysis (EGA) of the glucose predictions for a non-diabetic subject S1, pre-diabetic subject S2, type I diabetic subject S3, and type II diabetic subject S4 (black dots: Day 8 samples, red dots: Day 9 samples, orange line: linear least-squares regression of the reference versus the predicted glucose concentrations).