Noninvasive Glucose Measurements in Tissue Simulating Phantoms Using a Solid-State Near-Infrared Sensor

TL;DR

A new solid-state laser sensor for noninvasive glucose measurement in skin is benchmarked against a traditional FT spectrometer, showing promising results.

Contribution

A solid-state laser-based near-infrared sensor is evaluated for noninvasive glucose monitoring using tissue simulating phantoms.

Findings

The solid-state laser platform had a median RMS noise level of 327.8 µAU, higher than the FT spectrometer's 667.2 µAU.

The laser prototype achieved an SECV of 7.82 mg/dL, comparable to the FT spectrometer's 6.62 mg/dL.

Higher spectral noise in the laser prototype required two additional PLS factors for accurate glucose predictions.

Abstract

Benchmark data are reported for a solid-state laser-based near-infrared spectrometer designed for noninvasive measurements in human skin. These data were obtained using a set of aqueous phantoms composed of polystyrene beads, triton X-100, saline, and glucose. The performance of this prototype solid-state laser platform was compared to parallel results obtained with a Fourier-transform (FT) spectrometer. The fundamental spectroscopic performances of the two spectrometer systems were quantified by an analysis of 100% lines determined by ratioing back-to-back spectra collected over time for each phantom. Root mean square (RMS) noise levels were computed for each dataset and the median RMS noise levels were 327.8 µAU and 667.2 µAU for the FT spectrometer and prototype laser platform, respectively. The analytical utility of the solid-state laser platform was assessed through a series of leave-one-phantom-out partial least squares analyses. Results for the laser prototype data included a standard error of cross validation (SECV) of 7.82 mg/dL for an optimized PLS model with 10 factors over a spectral range of 1401–2238 nm. This compares favorably with the results from the FT spectrometer of an SECV of 6.62 mg/dL with 8 factors and a spectral range of 1551–2378 nm. The additional two PLS factors for the laser prototype were shown to be a consequence of its higher spectral noise. Selectivity of these PLS models was demonstrated by comparing models associated with correct and random glucose assignments to each spectrum. Overall, these findings benchmark the analytical utility of this solid-state laser prototype.

Figures