NON- INVASIVE GLUCOSE MONITORING DEVICE USING PHOTOPLETHYSMOGRAPHY SIGNAL AND MACHINE LEARNING MODEL
Keywords:
diabetes mellitus, photoplethysmography, feature extraction, signal processing machine learning, extra trees.Abstract
Diabetes mellitus is a metabolic condition that is permanent and that is why the blood sugar level is under constant monitoring. Conventional surveillance is based on invasive finger-prick tests that may be painful and may decrease patient compliance. In order to deal with this problem, this paper suggests an estimation system of glucose that will be non-invasive and work with photoplethysmography (PPG) signals and machine learning. The system employs a MAX30105 optical sensor to record PPG signals through the fingertip and ESP32 microcontroller to take data, process it, and analyze it in real-time. The synchronized PPG signals along with reference glucose values of a standard glucometer were acquired as a dataset. The PPG waveform was processed to obtain relevant features and was sent through a machine learning model to predict glucose. It has been demonstrated that a low-cost, portable, and non-invasive glucose monitoring system can be developed using PPG signal analysis and machine learning together and be used in wearable healthcare systems.
