Advancements in Non-Invasive Blood Glucose Monitoring Using Near-Infrared Spectroscopy: A Systematic Review.

Abstract

Non-invasive blood glucose monitoring offers pain-free and patient-friendly alternatives to invasive finger-prick testing and minimally invasive Continuous Glucose Monitoring (CGM). Among optical modalities, Near-InfraRed (NIR) Spectroscopy (NIRS) has attracted particular interest because it can probe glucose-related absorbance and scattering features in tissue while remaining compatible with compact, low-cost hardware. This systematic review synthesises developments in NIRS-based non-invasive blood glucose monitoring between 2019 and 2025. Major scientific databases (Scopus, PubMed, Web of Science, IEEE Xplore and ScienceDirect) were searched according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines and 16 studies met the inclusion criteria. The included works span laboratory-grade spectrometers, fiber- and free-space optical systems and light-Emitting Diode (LED)-photodiode sensor modules in transmittance, reflectance and interactance configurations, together with chemometric and machine-learning models for calibration. Across these studies, correlation coefficients between NIRS-based estimates and reference glucometers generally ranged from about 0.85 to 0.95, with Root Mean Square Error (RMSE) and Root Mean Square Error of Prediction (RMSEP) values typically between 12 and 25 mg/dL. Clarke Error Grid (CEG) analyses most often placed 85-95% of predictions in Zone A and the remainder mainly in Zone B, indicating clinically acceptable performance under controlled conditions. Recent trends include multi-wavelength NIR sources, LED-based hardware, multimodal sensing and advanced modelling approaches. However, accuracy frequently degrades across subjects, anatomical sites and free-living conditions and no NIRS-based system has yet obtained regulatory approval for insulin-dosing decisions. Overall, NIRS appears most suitable at present as an adjunct for trend tracking and alerts rather than a full replacement for invasive methods and future progress will require larger, more diverse clinical studies and the translation of spectroscopic and machine-learning advances into practical, low-cost reflective NIR sensor platforms.