A novel non-invasive glucose monitoring system is developed with combining dual-wavelength PPG signals (1480 nm and 1640 nm) and machine learning. Real-time signal quality evaluation achieved 95.6% accuracy, and the XGBoost model predicted glucose levels with RMSE 10.932 mg/dL as presented in Appendix Fig. 4. Compared to the FDA requirement that at least 95% of values fall within Zones A and B of the Clarke Error Grid, the proposed method achieved 98.2%.

This technology uniquely combines dual long wavelengths PPG signals at 1480 nm and 1640 nm with a deep learning-based quality assessment and XGBoost glucose prediction model. It achieves superior accuracy in non-invasive glucose monitoring by exploiting glucose-specific absorption peaks, enabling continuous, real-time, and reliable blood glucose estimation.

This technology offers a practical, non-invasive, wearable glucose monitoring solution for diabetic patients, improving compliance and enabling continuous metabolic health tracking. It can be integrated into smartwatches or portable health devices, supporting personalized diabetes management and early intervention, potentially reducing healthcare costs and improving patient outcomes.