Reliable Noninvasive Glucose Sensing via CNN-Based Spectroscopy

El Arbi Belfarsi, Henry Flores, Maria Valero

Confirmation: I have read and agree with the IEEE BHI 2025 conference submission's policy on behalf of myself and my co-authors.

Keywords: Noninvasive glucose monitoring, spectroscopy, convolutional neural networks, random forest regression, photodiodes, feature extraction, wearable sensors

Abstract: In this study, we investigate two distinct methods based on short wave infrared (SWIR) spectroscopy for non-invasive glucose monitoring. The first method employs a multi- wavelength SWIR imaging system combined with convolutional neural networks (CNNs) to extract spatial features associated with glucose absorption. The second method utilizes a compact photodiode voltage sensor with machine learning regressors (e.g., random forest) applied to normalized optical signals. Both methods were evaluated on synthetic blood phantoms and skin-mimicking materials across physiological glucose concentrations (70–200 mg/dL). The CNN-based system achieved a mean absolute percentage error (MAPE) of 4.82% at 650 nm, with 100% Zone A coverage in the Clarke Error Grid. In contrast, the photodiode system achieved 86.4% Zone A accuracy. Together, these results demonstrate distinct sensing strategies that balance clinical accuracy, cost considerations, and potential for wearable integration, contributing to the development of reliable continuous non-invasive glucose monitoring.

Track: 1. Biomedical Sensor Informatics

Registration Id: Z5N7HZ6BD4J

Submission Number: 217