Development of a Noninvasive Blood Glucose Monitoring System Prototype: Pilot Study

Maria Valero; Priyanka Pola; Oluwaseyi Falaiye; Katherine H Ingram; Liang Zhao; Hossain Shahriar; Sheikh Iqbal Ahamed

doi:10.2196/38664

Development of a Noninvasive Blood Glucose Monitoring System Prototype: Pilot Study

JMIR Form Res. 2022 Aug 26;6(8):e38664. doi: 10.2196/38664.

Authors

Maria Valero¹, Priyanka Pola¹, Oluwaseyi Falaiye², Katherine H Ingram³, Liang Zhao¹, Hossain Shahriar¹, Sheikh Iqbal Ahamed⁴

Affiliations

¹ Department of Information Technology, Kennesaw State University, Marietta, GA, United States.
² Department of Software Engineering and Game Development, Kennesaw State University, Marietta, GA, United States.
³ Department of Exercise Science and Sport Management, Kennesaw State University, Kennesaw, GA, United States.
⁴ Department of Computer Science, Marquette University, Milwaukee, WI, United States.

PMID: 36018623
PMCID: PMC9463623
DOI: 10.2196/38664

Abstract

Background: Diabetes mellitus is a severe disease characterized by high blood glucose levels resulting from dysregulation of the hormone insulin. Diabetes is managed through physical activity and dietary modification and requires careful monitoring of blood glucose concentration. Blood glucose concentration is typically monitored throughout the day by analyzing a sample of blood drawn from a finger prick using a commercially available glucometer. However, this process is invasive and painful, and leads to a risk of infection. Therefore, there is an urgent need for noninvasive, inexpensive, novel platforms for continuous blood sugar monitoring.

Objective: Our study aimed to describe a pilot test to test the accuracy of a noninvasive glucose monitoring prototype that uses laser technology based on near-infrared spectroscopy.

Methods: Our system is based on Raspberry Pi, a portable camera (Raspberry Pi camera), and a visible light laser. The Raspberry Pi camera captures a set of images when a visible light laser passes through skin tissue. The glucose concentration is estimated by an artificial neural network model using the absorption and scattering of light in the skin tissue. This prototype was developed using TensorFlow, Keras, and Python code. A pilot study was run with 8 volunteers that used the prototype on their fingers and ears. Blood glucose values obtained by the prototype were compared with commercially available glucometers to estimate accuracy.

Results: When using images from the finger, the accuracy of the prototype is 79%. Taken from the ear, the accuracy is attenuated to 62%. Though the current data set is limited, these results are encouraging. However, three main limitations need to be addressed in future studies of the prototype: (1) increase the size of the database to improve the robustness of the artificial neural network model; (2) analyze the impact of external factors such as skin color, skin thickness, and ambient temperature in the current prototype; and (3) improve the prototype enclosure to make it suitable for easy finger and ear placement.

Conclusions: Our pilot study demonstrates that blood glucose concentration can be estimated using a small hardware prototype that uses infrared images of human tissue. Although more studies need to be conducted to overcome limitations, this pilot study shows that an affordable device can be used to avoid the use of blood and multiple finger pricks for blood glucose monitoring in the diabetic population.

Keywords: deep learning; diabetes; glucose concentration; glucose monitoring; machine learning; noninvasive monitoring; optical sensors.

©Maria Valero, Priyanka Pola, Oluwaseyi Falaiye, Katherine H Ingram, Liang Zhao, Hossain Shahriar, Sheikh Iqbal Ahamed. Originally published in JMIR Formative Research (https://formative.jmir.org), 26.08.2022.