Development of an Optical-Based, Non-Invasive Blood Glucose Level Assessment System

Glucose monitoring is crucial for effective management of diabetes, enabling timely intervention and personalized treatment to prevent complications and improve patient outcomes. However, traditional self-monitoring blood glucose (SMBG) methods are invasive, causing discomfort and increasing the risk of infection. This project proposes an IoT-based non-invasive glucose monitoring system using Near-Infrared (NIR) spectroscopy for convenient diabetes management. The device includes a 940-nm TSAL4400 NIR emitter, TEFT4300 phototransistor, ESP32 microcontroller, and an isolated printed circuit board housed in a 3D-printed casing. Features such as voltage drop normalized by finger thickness, Body Mass Index (BMI), and fasting status were used to train a classification model to predict whether a measurement falls into the normal or prediabetes class based on finger measurements. ESP32's integrated Wi-Fi module enables internet connectivity and facilitates data transmission between Arduino Cloud, allowing data to be stored both locally and in the cloud. The proposed device was developed and validated through in vitro testing and with 25 subjects. The correlation between NIR absorbance by glucose molecules and glucose concentration was established. The Random Forest (RF) classifier outperformed the Support Vector Machine (SVM) model with an accuracy of 87% compared to 55% for SVM. This project successfully demonstrated the ability of the non-invasive glucose monitoring device to effectively monitor glucose status for public users at any time.