Wavelength selection-based nonlinear calibration for transcutaneous blood glucose sensing using Raman spectroscopy

Area of Science:

• Biomedical Optics
• Spectroscopic Diagnostics
• Computational Chemistry

Background:

• Raman spectroscopy offers noninvasive, real-time diagnostics for biological samples.
• Clinical translation is hindered by calibration model fragility and system bulkiness.
• Linear models struggle with spurious correlations, overfitting, and nonlinearities from interferences.

Purpose of the Study:

• To enhance the robustness and clinical applicability of Raman spectroscopic calibration models.
• To address limitations of linear models by incorporating wavelength selection and nonlinear regression.
• To facilitate the development of miniaturized Raman systems for disease diagnosis.

Main Methods:

• Implemented residue error plot-based wavelength selection to identify informative spectral regions.
• Utilized nonlinear support vector regression (SVR) to model complex spectral effects like turbidity and temperature.
• Validated methods using glucose detection in tissue phantoms and clinical human subject data.

Main Results:

• Wavelength selection with SVR achieved prediction accuracy comparable to linear full spectrum analysis, even with reduced data.
• Selected wavelength subsets maintained prediction accuracy on clinical datasets.
• Reduced spectral data acquisition time and potential for calibration maintenance and transfer.

Conclusions:

• Combined wavelength selection and SVR improve Raman spectroscopic calibration model robustness.
• The approach supports accurate diagnostics with reduced data, enabling faster acquisition.
• This methodology paves the way for miniaturized, clinically viable Raman diagnostic systems.