# A Novel Spectral Barcoding and Classification Approach for Complex Biological Samples Using Multiexcitation Raman Spectroscopy (MX-Raman)

## TL;DR

This paper introduces a new method using multiexcitation Raman spectroscopy to improve the classification of complex biological samples, particularly in neurodegenerative diseases.

## Contribution

The novel MX-Raman methodology enhances label-free classification accuracy by combining multiple excitation wavelengths and polarization states.

## Key findings

- MX-Raman achieved 96.7% classification accuracy compared to 78.5-85.6% with single-excitation Raman.
- Combining laser polarizations slightly improved accuracy without a second laser.
- Spectral barcodes with minimal disease-specific features improved clustering and classification.

## Abstract

We report the development
and application of a novel spectral barcoding
approach that exploits our multiexcitation (MX) Raman spectroscopy-based
methodology for improved label-free detection and classification of
complex biological samples. To develop our improved MX-Raman methodology,
we utilized post-mortem brain tissue from several neurodegenerative
diseases (NDDs) that have considerable clinical overlap. For improving
our methodology we used three sources of spectral information arising
from distinct physical phenomena to assess which was most important
for NDD classification. Spectral measurements utilized combinations
of data from multiple, distinct excitation laser wavelengths and polarization
states to differentially probe molecular vibrations and autofluorescence
signals. We demonstrate that the more informative MX-Raman (532 nm–785
nm) spectra are classified with 96.7% accuracy on average, compared
to conventional single-excitation Raman spectroscopy that resulted
in 78.5% accuracy (532 nm) or 85.6% accuracy (785 nm) using linear
discriminant analysis (LDA) on 5 NDD classes. By combining information
from distinct laser polarizations we observed a nonsignificant increase
in classification accuracy without the need of a second laser (785
nm–785 nm polarized), whereas combining Raman spectra with
autofluorescence signals did not increase classification accuracy.
Finally, by filtering out spectral features that were redundant for
classification or not descriptive of disease class, we engineered
spectral barcodes consisting of a minimal subset of highly disease-specific
MX-Raman features that improved the unsupervised and cross-validated
clustering of MX-Raman spectra. The results demonstrate that increasing
spectral information content using our optical MX-Raman methodology
enables enhanced identification and distinction of complex biological
samples but only when that information is independent and descriptive
of class. The future translation of such technology to biofluids could
support diagnosis and stratification of patients living with dementia
and potentially other clinical conditions such as cancer and infectious
disease.