Multivariate calibration with basis functions derived from optical filters.

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemometrics

Background:

• Multivariate calibration models are essential for extracting information from spectral data.
• Near-infrared spectroscopy (NIRS) is a powerful tool for quantitative analysis.
• Existing calibration methods like partial least-squares (PLS) regression can face challenges with long-term stability.

Purpose of the Study:

• To develop and evaluate a novel calibration methodology using Gaussian basis functions for NIRS.
• To compare the performance of Gaussian basis function models against conventional PLS models.
• To assess the long-term stability and predictive accuracy of the developed models.

Main Methods:

• Construction of multivariate calibration models using Gaussian basis functions.
• Optimization of basis functions via genetic algorithms.
• Application of models to determine glucose levels in synthetic biological matrices using NIRS.
• Validation against external prediction data and comparison with PLS regression.

Main Results:

• Gaussian basis function models outperformed PLS models in quantitative analysis of glucose.
• The developed models demonstrated enhanced calibration stability over extended periods (up to 4 months).
• The methodology offers a direct implementation pathway for calibration models in spectrometer hardware.

Conclusions:

• Gaussian basis functions provide a robust and stable alternative for NIRS multivariate calibration.
• This approach enhances the reliability of quantitative spectral analysis, particularly for biological samples.
• The direct hardware implementation potential offers significant advantages for real-time spectroscopic measurements.