Non Invasive Glucose Level Prediction Leveraging PPG Signal from Smartphone

The increasing prevalence of diabetes necessitates innovative and approachable blood glucose monitoring techniques to ensure effective disease management and prevent complications. The traditional methods, despite being accurate, are often uncomfortable and inconvenient, which lowers patient adherence. In order to overcome this difficulty, we suggest a non-invasive glucose monitoring system that makes use of Photoplethysmography (PPG) signals obtained from index finger video recordings made with smartphones. Employing a unique methodology, we extract distinctive features from both temporal and frequency domains of PPG signals acquired through noninvasive means. The heart of our approach lies in regression techniques within machine learning, allowing us to forecast actual blood glucose levels rather than categorizing them. Rigorous evaluation on an extensive dataset of 60 subjects attests to the efficacy of our system, achieving a notable MAE of 0.9 in predicting blood glucose concentrations. This underscores the promising potential of our approach for daily and clinical applications in non-invasive blood glucose testing, emphasizing its role in accurate and continuous glucose level monitoring without resorting to invasive methods.