Non-Invasive Glucose Estimation Device: A Hardware-Based Research

Diabetes management depends heavily on regular blood glucose monitoring, but conventional measurement methods generally require invasive or minimally invasive procedures that may cause discomfort, expense, and reduced user compliance. This paper presents the design of a hardware-based non-invasive glucose measurement research prototype that combines multi-spectral optical sensing, photoplethysmography (PPG), bioimpedance, temperature measurement, pressure sensing, inertial motion detection, and ambient-light compensation. The proposed device is based on an ESP32-S3 microcontroller and integrates an AS7265x 18-channel multi-spectral sensor, MAX86176 PPG sensor, AD5940 bioimpedance analog front end, MAX30205 temperature sensor, FSR402 pressure sensor, ICM-42688 inertial measurement unit, and TSL2591 ambient-light sensor. The collected multi-modal signals are processed using preprocessing, feature extraction, feature fusion, and personalized machine learning techniques. Random Forest and XGBoost are considered as baseline predictive models, with per-user calibration intended to address individual variation in skin properties, tissue characteristics, and physiological conditions. A reference glucometer is used during calibration and validation sessions to obtain paired reference glucose measurements. The proposed research methodology emphasizes measurement quality, motion and contact-artifact rejection, personalized calibration, and clinically informed evaluation using Mean Absolute Relative Deviation (MARD) and Clarke or Consensus Error Grid analysis. The proposed device is positioned strictly as a research prototype and not as a certified diagnostic or medical-reference instrument. The main contribution of the project is the integration of complementary sensing modalities into an accessible hardware platform for investigating non-invasive glucose estimation.