A Portable AI-Based Non-Invasive Blood Glucose Monitoring System Using Multi-Wavelength NIR Spectroscopy

Abstract

Diabetes mellitus (DM) is a metabolic disorder characterized by increased blood glucose levels, which can cause significant health problems and premature death. Patients with diabetes must regularly monitor their blood glucose (BG) levels in order to estimate the insulin intake. Invasive methods are commonly used for blood glucose measurement because they are highly accurate but require skin puncturing, which is uncomfortable and increases the risk of infectious disease transmission. Alternatively, non-invasive techniques need no skin damage, making them simple, painless, and practical for routine monitoring. We present a non-invasive glucose monitoring system that analyses reflectance and transmittance of near-infrared (NIR) signals at three wavelengths (940 nm, 1050 nm, and 1300 nm). LEDs and photo-detector forms an optoelectronics circuit for NIR signals, while an ATmega32 microcontroller processes digital signals and displays the results. A cluster-based piecewise regression model is proposed to develop an optimized model from NIR discrete voltage values and invasive glucometer readings. A five-cluster piecewise linear multivariate regression (PLMR) model is employed on the training data set, which shows an excellent correlation of 92.24% and a low RMSE value of 17.28 mg/dl, indicating that it can be used in practice. The developed device is low-cost, portable and offers a painless alternative for regular monitoring.