Advances in Non-Invasive Glucose Sensing: A Comprehensive In Vitro Analysis
Disciplines
Biomedical Engineering and Bioengineering | Biomedical Informatics | Biotechnology | Computer Engineering
Abstract (300 words maximum)
With over 800 million adults living with diabetes worldwide and nearly one in three adults in the United States diagnosed with prediabetes, there is an urgent need for accessible blood glucose monitoring solutions. Traditional glucose measurement methods are often invasive, painful, and costly, posing significant barriers to routine monitoring and early intervention. Non-invasive glucose sensing offers a promising alternative. This study presents a comprehensive in vitro investigation combining voltage-based sensing and infrared (IR) imaging for glucose concentration estimation using synthetic blood samples beneath a skin-mimicking layer. Our system employs near-infrared (NIR) light emission at 1550 nm, with voltage signals captured via an InGaAs photodiode and digitized using an ADS1115 ADC. In parallel, an IR camera captures images of the light transmission through the medium. These images are then analyzed using deep learning models to predict glucose concentration. Glucose levels range from 70 to 300 mg/dL in 10 mg/dL increments. The dataset is systematically processed and modeled using both statistical regression and convolutional neural networks to compare performance. Clinical relevance is assessed using Zone A of the Clarke Error Grid Analysis. By evaluating and comparing both voltage-based and image-based predictions, this study contributes to the development of multi-modal, non-invasive glucose monitoring technologies that leverage optical sensing and AI-driven analysis.
Academic department under which the project should be listed
CCSE - Computer Science
Primary Investigator (PI) Name
Dr. Maria Valero de Clemente
Advances in Non-Invasive Glucose Sensing: A Comprehensive In Vitro Analysis
With over 800 million adults living with diabetes worldwide and nearly one in three adults in the United States diagnosed with prediabetes, there is an urgent need for accessible blood glucose monitoring solutions. Traditional glucose measurement methods are often invasive, painful, and costly, posing significant barriers to routine monitoring and early intervention. Non-invasive glucose sensing offers a promising alternative. This study presents a comprehensive in vitro investigation combining voltage-based sensing and infrared (IR) imaging for glucose concentration estimation using synthetic blood samples beneath a skin-mimicking layer. Our system employs near-infrared (NIR) light emission at 1550 nm, with voltage signals captured via an InGaAs photodiode and digitized using an ADS1115 ADC. In parallel, an IR camera captures images of the light transmission through the medium. These images are then analyzed using deep learning models to predict glucose concentration. Glucose levels range from 70 to 300 mg/dL in 10 mg/dL increments. The dataset is systematically processed and modeled using both statistical regression and convolutional neural networks to compare performance. Clinical relevance is assessed using Zone A of the Clarke Error Grid Analysis. By evaluating and comparing both voltage-based and image-based predictions, this study contributes to the development of multi-modal, non-invasive glucose monitoring technologies that leverage optical sensing and AI-driven analysis.