Design of a Wearable Finger PPG-Based Blood Glucose Monitor.

Abstract

Blood glucose monitoring is crucial for controlling diabetes. However, traditional fingertip pricking methods usually cause discomfort to patients and cannot achieve continuous monitoring. To overcome these limitations, we developed a novel, non-invasive, and wearable device for continuous blood glucose levels (BGLs) monitoring. The device is equipped with a pulse oximeter, which has a visible wavelength (660 nm) and an infrared wavelength (880 nm) light-emitting diode (LED) to acquire finger photoplethysmography (PPG). The BGLs from PPG were estimated using a multi-layer perceptron (MLP) machine learning model, which was trained on dual-wavelength PPG intensity pertaining to various BGLs. We also analyzed the effect of MLP training parameters on the accuracy of blood glucose prediction. Experimental results indicate that 99.33% of the BGLs estimated from PPG lie in the clinically acceptable Clarke error grid (CEG) regions A and B, suggesting a high potential for accurate blood glucose monitoring with minimal clinical risk. Additionally, our 24-hour monitoring test further validates the device's capability to effectively track daily glucose fluctuations, which verifies its reliability in daily blood glucose monitoring. In conclusion, our novel wearable device for continuous blood glucose monitoring has shown feasibility and effectiveness. By leveraging PPG signals and a machine learning model, we have developed a promising alternative to traditional invasive blood glucose monitoring methods. This device has the potential to significantly improve the quality of life for diabetes patients by providing a more comfortable and continuous monitoring option.