A Dual-Channel Wearable Ring With MODWT-Based Signal Processing for Non-Invasive Blood Glucose Estimation: A Pilot Study

Abstract

Diabetes management necessitates regular blood glucose level (BGL) monitoring, but invasive finger-prick glucose monitoring suffer from poor patient compliance due to discomfort, infection risks, and high costs. This study presents a non-invasive BGL monitoring system that utilizes a ring device with maximal overlap discrete wavelet transform (MODWT)-based multiscale signal analysis. The ring device incorporates an anatomically-optimized photoplethysmography (PPG) sensor on finger digit, capturing signals at 850 nm and 523 nm wavelengths. The MODWT framework simultaneously denoises and decomposes PPG signals and their derivatives, including velocity plethysmogram (VPG) and acceleration plethysmography (APG), into multi-frequency scales while maintaining full temporal resolution. High-quality segments are identified through an advanced signal quality index. After data processing, statistical features, including nonlinear measures targeting to glucose-relevant vascular, autonomic, and respiratory mechanisms and complementary linear features, were extracted and optimized through filter-based feature selection. Four machine learning regression models were evaluated using data from 47 participants across three clinical groups (healthy, hypertension, and hypertension with diabetes). Linear support vector regression achieved the superior performance, particularly when using PPG and APG signals with MODWT decomposition, yielding RMSE = 16.023 mg/dL, and 100% accuracy in Clarke plot analysis within Zone A+B under five-fold stratified cross-validation. Fractal dimension features from MODWT-decomposed green channel signals, which characterize the structure of the vasculature, emerges as the most discriminative features for glucose estimation. This integrated approach combining anatomy-based sensor design and physiological-related features from multiscale signal decomposition demonstrates feasibility of a practical approach for wearable and non-invasive glucose monitoring.