Models and methods for quantitative analysis of surface-enhanced Raman spectra

Area of Science:

• Spectroscopy and Imaging
• Biomedical Engineering
• Chemometrics

Background:

• Surface-enhanced Raman spectroscopy (SERS) shows promise for in vivo molecular imaging.
• Quantitative analysis of SERS data is crucial but current methods lack clear understanding.
• Existing methods include classical least squares and various multivariate calibration models.

Purpose of the Study:

• To elucidate the theoretical foundations of commonly used quantitative analysis models for SERS.
• To demonstrate the suitability of latent variable regression (LVR) models for SERS quantitative analysis.
• To compare the performance and underlying principles of different LVR methods.

Main Methods:

• Theoretical analysis of direct classical least squares, full spectrum, and selected multivariate calibration models.
• Comparative analysis of LVR methods including principal component regression, reduced-rank regression, partial least square regression (PLSR), canonical correlation regression, and robust canonical analysis.
• Evaluation of model performance using SERS datasets and diverse criteria.

Main Results:

• LVR models are theoretically better suited for Raman spectra quantitative analysis.
• Partial least square regression (PLSR) is identified as a hybrid model combining multivariate calibration and feature extraction.
• PLSR effectively relates nanotag concentrations to spectrum intensity by optimizing latent variables.

Conclusions:

• A deeper understanding of LVR models, particularly PLSR, is provided for SERS quantitative analysis.
• PLSR's unique approach explains its superior performance in determining nanotag concentrations from SERS data.
• The study validates the effectiveness of various models through empirical testing on SERS datasets.