Chemometric Methods for Biomedical Raman Spectroscopy and Imaging - Emerging Raman Applications and Techniques in Biomedical and Pharmaceutical Fields - page 175

8
Chemometric Methods for Biomedical Raman
Spectroscopy and Imaging
Rohith K. Reddy and Rohit Bhargava
Abstract The vibrational spectrum is a quantitative measure of a sample's molecu-
lar composition. Hence, classical chemometric methods, especially regression-based,
have focused on exact mapping between identity and sample composition. While
this approach works well for molecular identifications and scientific investigations,
problems of biomedical interest often involve complex mixtures of stochastically
varying compositions and complex spatial distributions of molecules contributing
to the recorded signals. Hence, the challenge often is not to predict the identity of
materials but to determine chemical markers that help rapidly detect species (e.g.
impurities, conformations, strains of bacteria) in large areas or indicate changes in
function in complex tissue (e.g. cancer or tissue engineering). Hence, the rate of data
analysis has to be rapid, has to be robust with respect to stochastic variance and
the provided information is usually related to biomedical context and not to molec-
ular compositions. The emergence of imaging techniques and clinical applications
are spurring growth in this area. In this chapter, we discuss chemometric methods
that are useful in this milieu. We first review methods for data pre-processing with
a focus on the key challenges facing a spectroscopist. Next, we survey some of the
well known, widely used pattern classification techniques under the framework of
supervised and unsupervised classification. We discuss the applicability, advantages
and drawbacks of each of these techniques and help the reader not only gain use-
ful insights into the techniques themselves but also acquire an understating of the
underlying ideas and principles. We conclude by providing examples of the coupled
use of chemometric and statistical tools to develop robust classification protocols for
prostate and breast tissue pathology. We specifically focus on the critical factors and
pitfalls at each step in converting spectral data sets into hi-fidelity images useful for
decision making.
8.1 Introduction
Spectroscopic data can broadly be used to understand molecular structure,
to relate function to molecular content, and to understand evolution of struc-
ture and properties with both time and/or stimulus [1]. Generalizing these
activities, one can summarize that much effort in biomedical spectroscopy is