Optimal Wavelength Bands for Remote Blood Glucose Estimation Using Facial NIR Spectroscopic Images Measured at...

Area of Science:

• Biomedical Optics
• Spectroscopy
• Medical Imaging

Background:

• Traditional blood glucose monitoring is invasive and inconvenient.
• Non-invasive near-infrared (NIR) spectroscopy presents a promising alternative.
• Facial NIR imaging offers a novel approach for unobtrusive monitoring.

Purpose of the Study:

• To explore a novel non-invasive method for blood glucose estimation using spatial features from facial NIR images.
• To identify an optimal narrowband wavelength for improved accuracy and simpler hardware.
• To assess the feasibility of a compact NIR system for real-world applications.

Main Methods:

• Facial NIR spectroscopic images were acquired using discrete light sources (800-1650 nm).
• Spatial features were extracted using independent component analysis (ICA) from single-frame images.
• Narrowband wavelengths were investigated to optimize estimation accuracy and reduce noise.

Main Results:

• The optimal wavelength for blood glucose estimation was found to be 1200 nm.
• Spatial features, particularly along the side of the nose, showed strong correlation with glucose levels.
• While RMSE was higher than contact methods, the trade-off allows for unobtrusive monitoring.

Conclusions:

• 1200 nm is a viable wavelength for non-contact blood glucose estimation.
• The developed method supports the feasibility of compact, low-cost NIR systems for non-invasive glucose monitoring.
• Potential applications include telehealth, smart homes, and public health settings.