Highly Sensitive Perovskite Photoplethysmography Sensor for Blood Glucose Sensing Using Machine Learning Techniques

Accurate non-invasive monitoring of blood glucose (BG) is a challenging issue in the therapy of diabetes. Here near-infrared (NIR) photoplethysmography (PPG) sensor based on a vapor-deposited mixed tin-lead hybrid perovskite photodetector is developed. The device shows a high detectivity of 5.32 × 10¹² Jones and a large linear dynamic range (LDR) of 204 dB under NIR light, guaranteeing accurate extraction of eleven features from the PPG signal. By a combination of machine learning, accurate prediction of blood glucose level with mean absolute relative difference (MARD) as small as 2.48% is realized. The self-powered PPG sensor also works for real-time outdoor healthcare monitors using sunlight as a light source. The potential for early diabetes diagnoses by the perovskite PPG sensor is demonstrated.