Enhancing the Accuracy of Non-Invasive Blood Glucose Monitoring Using Deep Learning Methods

Abstract

Accurate blood glucose monitoring is crucial for diabetes management. Traditional invasive methods remain the gold standard due to their reliability; however, the demand for non-invasive alternatives has surged due to patient comfort and the necessity for continuous monitoring. This study bridges the accuracy gap between non-invasive and invasive blood glucose measurements using deep learning algorithms. An infrared-based sensor was employed to capture voltage variations correlated with blood glucose levels, collecting data from over 110 participants. Initially, a polynomial regression model achieved an accuracy of 83.5%. After expanding the dataset and incorporating additional biometric features (such as age, BMI, blood pressure, and family history), the enhanced deep neural network (DNN) model was optimized through hyper parameter tuning, significantly improving prediction accuracy to 96.85%. These results highlight the superiority of deep learning over traditional regression methods in refining non-invasive glucose measurements. The substantial reduction in measurement discrepancies suggests promising clinical applications. Beyond its technical contributions, this research aligns with the United Nations Sustainable Development Goals (SDGs), particularly in promoting good health and well-being and fostering innovation in healthcare technologies. By enhancing the accuracy of non-invasive glucose monitoring, this study advances the potential for cost-effective, patient-friendly diabetes management solutions, improving accessibility and reducing reliance on invasive procedures.