A Single Wavelength LED Photoplethysmography Used for Non-Invasive Glucose Monitoring

Diabetes is a chronic disease, occurs because of insufficient insulin production from pancreas. Continuous monitoring of glucose levels is critical so that it can prevent damage to human organs. Conventional invasive ways for monitoring glucose are painful, expensive, and risk infection. This paper introduces a non-invasive system to monitor blood glucose levels based on photoplethysmography (PPG) using near-infrared LED and a photodiode integrated with a machine learning regression. The PPG circuit consists of an amplifier with controllable gain, a 3.4 HZ active LPF, and 0.5 Hz passive HPF for DC bias rejection. A preprocessing stage involves an Asymmetric Least Squares (ALS) technique for baseline drift remove. A Gaussian filter is used for reduction of noise and motion artifacts cancellation, the signal is normalized using the min-max technique. Key features are extracted from the pre-processed signal, mean and variance are calculated for each feature extracted. Finally, all features are fed into the regression model for final estimation of blood Glucose.