gSENSE: A PPG-based Automated Framework for Continuous Blood Glucose Level Monitoring Using Deep Learning Approach

Traditional blood glucose testing method, like finger pricking, is painful, time-consuming, and expensive for people who need regular monitoring. In recent years, photoplethysmography (PPG) has become more popular as a non-invasive alternative. PPG measures changes in blood volume using light, offering valuable insights into cardiovascular health. This study proposes gSENSE, a deep learning-based approach to estimate blood glucose levels from PPG signals. A Continuous Wavelet Transform (CWT) technique extracts glucose-relevant features, converting time-series PPG data into 2D scalograms. These scalograms are then processed by a deep learning regression model to capture complex patterns for precise glucose estimation. Evaluated on a public PPG dataset, our approach outperformed existing methods, achieving an R² of 0.89 and an MSE of 0.002. To bridge research with real-world use, we deployed our model on an Android app, where a smartphone’s camera and flashlight capture PPG signals for real-time glucose estimation. This method offers a non-invasive, cost-effective, and accessible solution for continuous glucose monitoring, benefiting both diabetic patients and individuals at risk of diabetes.