The application of Raman spectroscopy to studies of disease states results from an interesting confluence of spectroscopists
and medical researchers. Vibrational spectroscopy has the potential to provide a wealth of information about biological materials,
and spectroscopists who have established collaborations with medical researchers have been developing the field. The challenge
is to determine how to extract the information from the data. There are two paths to achieve this goal. On the one hand, spectroscopists
like to "interpret" spectra. A given molecule will have characteristic bands (1–3), and these spectra can be modified by many
environmental factors. However, any biological system will have so many molecules, each of which has in principle hundreds
of vibrational degrees of freedom, in all types of environments, that one can argue that spectral interpretations of the disease
state will be nearly impossible. From the opposite approach, one can collect lots of data and subject it to multivariate (statistical)
analysis; these tools enable the identification of spectral trends that are difficult to perceive visually (4,5). This column
will consider what progress is being made in the field and why it is worth the effort to pursue.
As an example of work that followed the first method of investigation, we cite the work of Michael Feld (6), who used "bulk"
spectra of atherosclerotic and normal tissues (elastic laminae, collagen fibers, smooth muscle, fat cells, foam cells, necrotic
core, cholesterol crystals, β-carotene crystals, and Ca mineralization) as basis spectra for linear least squares minimization
modeling of morphological structures of coronary artery tissue spectra. He succeeded in classifying these structures as nonatherosclerotic,
calcified plaque, or noncalcified atheromatous plaque. Subsequent prediction of 68 coronary artery samples produced a 94%
correct classification (64 samples).
While it is tantalizing to be able to assign specific molecular species to biochemical, physiological, and disease states,
there is a real risk of overinterpreting the results. As an example, Diem cites the mistaken association of glycogen as a
marker for cervical cancer (page 6 of reference 5).
On the other hand, mapping unstained histological sections is a way of collecting large heterogeneous data sets. Without any
a priori assumptions about the origins of the spectra, the data sets can be subjected to numerous unsupervised and supervised multivariate
algorithms and the results can be correlated with pathological conditions (7,8).
In this installment of "Molecular Spectroscopy Workbench," I will give an overview of the development of Raman mapping for
disease diagnosis. Certainly vibrational spectroscopy is not expected to be intuitive to medical practitioners, so why would
one want to hurdle the barrier to merge the two fields? For one thing, there must be a need. Why do medical diagnosticians
need another tool? The answer is obvious when one realizes that diagnosis is quite often not black versus white. For example,
in the case of atherosclerosis, it is nearly impossible to predict when deposits can be classified as vulnerable plaque, which
means they are susceptible to rupture resulting in a heart attack. In the case of cancer, there is often a continuous transformation
from healthy tissue to a premalignant state (hyperplasia) to carcinoma in situ to metastatic–invasive cancer. The problem
for the pathologist is to be able to recognize cancerous regions, identify their origins (when a metastasis provides the initial
diagnosis), and define their boundaries. But because early states of the disease often do not have identifiable histological
features, that identification is not always straightforward. And it is not unusual for two pathologists examining the same
tissue to draw different conclusions. Note that the consequences for both false positives and false negatives are substantial.
A false negative means that disease goes untreated. A false positive can result in physical disfiguration as well as emotional
trauma. Any technique that can improve the results of the diagnostic process will be a valuable tool for the diagnostician.