Multimodal multiplex Raman spectroscopy optimized for in vivo chemometrics

Scott T. McCain, Michael E. Gehm, Y. Wang, Nikos Pitsianis, David J. BradyProceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE 2006 · DOI

TL;DR

This paper introduces a novel multimodal multiplex Raman spectroscopy system that uses multiple excitation wavelengths to significantly improve the detection of Raman signals in the presence of strong fluorescence, a common challenge in biological samples. By leveraging the shift-variance of Raman signals and employing chemometric analysis (specifically PLS regression), they achieved a more robust and accurate estimation of ethanol concentration in a tissue phantom compared to traditional single-wavelength methods.

What Problem Does This Paper Address?

Raman spectroscopy is a powerful tool for chemical analysis due to its specificity, offering a unique fingerprint for different molecules. This makes it highly attractive for in vivo (within living organisms) applications, such as monitoring blood glucose or other analytes. However, biological samples are notoriously difficult for Raman spectroscopy. They often exhibit strong autofluorescence and intense scattering, which can easily overwhelm the weak Raman signal. This makes it challenging to accurately identify and quantify specific molecules, especially at low concentrations. Existing methods often require high excitation powers and long acquisition times, and the resulting models can be unreliable over time or across different users. The authors aim to overcome these limitations by developing a more robust Raman spectroscopy system that can better distinguish Raman signals from interfering fluorescence.

Methods

The core innovation is a multimodal multiplex Raman spectrometer. Instead of using a single laser, this system employs eight different laser wavelengths (ranging from 800-808 nm) simultaneously. This is coupled with a coded aperture spectroscopy approach, which allows for a large collection area. The key principle behind using multiple wavelengths is the shift-variance of the Raman signal. Raman peaks shift slightly with changes in excitation frequency, while fluorescence and stray light are largely unaffected. By collecting spectra at multiple excitation wavelengths, the system can then shift and sum these spectra. This process causes the Raman peaks to align and reinforce each other, while the non-Raman signals (fluorescence, scattering) become spread out and effectively reduced.

To analyze the complex spectral data, the researchers utilized Partial Least Squares (PLS) regression, a standard chemometric technique for dimensionality reduction and quantitative analysis. They also developed a sophisticated spectral pre-processing pipeline. This involved using a known Raman spectrum (acetaminophen) to calibrate the spectrometer and accurately remap pixel locations to wavenumber values. This ensured that spectra from different excitation lasers were properly aligned before being summed.

Datasets

The primary dataset used for evaluation was generated from tissue phantoms made of Intra-Lipid (mimicking biological tissue's scattering and fluorescence properties) mixed with varying concentrations of ethanol. Samples were prepared at 7 concentrations between 0.05% and 2% by weight (8.7-348 mmol/L), plus a pure Intra-Lipid sample. Each sample was measured with all 8 lasers, with 3 exposures of 60 seconds each, resulting in a dataset of 192 spectra.

For comparison, the authors also utilized a previous dataset from a single-wavelength excitation experiment under similar sample conditions but with higher laser power and longer exposure times.

Key Insights and Ideas

  • Exploiting Raman Shift-Variance is Key: The fundamental insight is that the slight shift in Raman peaks with excitation wavelength, combined with chemometric analysis, can be a powerful tool to separate Raman signals from broadband fluorescence.
  • Multiplexing Enhances Robustness: Using multiple excitation wavelengths simultaneously doesn't just increase signal-to-noise; it fundamentally makes the system more resilient to interferences and fluctuations, leading to more reliable quantitative models.
  • Fewer Components, Better Model: The multimodal approach required significantly fewer principal components in the PLS regression model to achieve optimal accuracy. This suggests that the resulting models are more parsimonious and likely more robust and generalizable.
  • Improved Low-Concentration Performance: The multimodal system showed a marked improvement in accuracy at lower ethanol concentrations, where single-wavelength methods struggled significantly.

Key Results

The multimodal multiplex Raman spectroscopy system demonstrated a significant improvement in performance for ethanol estimation in a tissue phantom:

  • Reduced Error: The Root-Mean-Squared-Error-Cross-Validation (RMSECV) for the multimodal system was 9.2 mmol/L. This is a substantial improvement compared to the single-wavelength system, which achieved an RMSECV of 39.1 mmol/L using equivalent total energy.
  • Fewer Model Components: The multimodal system required only 2 principal components in the PLS regression model to achieve its minimum error, whereas the single-wavelength system needed 7 components.
  • Visual Confirmation: Regression vectors derived from the multimodal system showed Raman peaks that more closely matched the pure ethanol spectrum, indicating the model was indeed learning the correct spectral features. Scatter plots also highlighted the superior accuracy of the multimodal system, particularly at lower concentrations.

Limitations and Future Work

  • Limitations: The paper notes that the model formation and validation were performed on the same dataset due to a limited amount of data. They used a contiguous block cross-validation to mitigate this, but a truly independent validation set would be ideal. The current system is also described as a prototype.
  • Future Work: The authors suggest that this technique holds promise for improving in vivo molecular estimation. They envision further advancements by exploring more complex excitation patterns and processing techniques to filter individual molecular signatures beyond just Raman features, potentially leading to even more sophisticated chemometric analysis.

Summary updated

Abstract

We have designed and constructed a multimodal multiplex Raman spectrometer which uses multi-wavelength excitation to better detect signals in the presence of fluorescence by taking advantage of the shift-variance of the Raman signal with respect to excitation frequency. Coupled with partial-least-squares (PLS) regression, the technique applied to ethanol estimation in a tissue phantom achieves root-mean-squared-cross-validation errors (RMSCVE) of 9.2 mmol/L with a model formed with 2 principal components, compared to a single wavelength data set with equivalent energy where 7 principal components were used to achieve an RMSCVE of 39.1 mmol/L.
BibTeX
@inproceedings{mccain2006multimodal,
  author = {Scott T. McCain and Michael E. Gehm and Y. Wang and Nikos Pitsianis and David J. Brady},
  title = {{Multimodal multiplex Raman spectroscopy optimized for in vivo chemometrics}},
  booktitle = {Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE},
  year = {2006},
  doi = {10.1117/12.646755},
}
Figure