Infrared and Raman Spectroscopy and Imaging of Breast Cancer - Vibrational Spectroscopy for Tissue Analysis - page 129

parameters of lipids/acylglyceride, proteins, and collagen together helps in
classifying three different nuclear grades (high, intermediate, and low) more
precisely and consistently. All of these results have been backed by our pub-
lished review works in Applied Spectroscopy Reviews [54,55].
Analysis of Cancerous Tissues Using FTIR Rapid Scan Imaging
The basic question is whether FTIR-MS (micro-spectroscopy) imaging with a
Focal Plane Array (FPA) camera can be useful in cancer diagnosis and grad-
ing by analysis of tissue samples, e.g., biopsies or excised tumours. Its poten-
tial advantages are that it requires little sample preparation and is therefore
rapid; that a single image can contain contrasts that would normally take
several different types of stains to obtain; and that it can be implemented
more objectively than a standard histological analysis.
There appear to have been a few tens of studies dealing with FTIR-MS
imaging of cancer-related samples published in the last few years. Several
reviews have been published recently [56-58].
FTIR-MS imaging produces three-dimensional data: an intensity quan-
tity (e.g., absorbance or reflectance) as a function of two spatial coordinates
(x and y) and wavelength (λ). This type of data is sometimes called a hyper-
spectral image . Since an entire infrared spectrum is recorded for every pixel,
hyperspectral images potentially contain a lot more information than can be
obtained with a visible-light microscope.
Most of the studies published to date analyse only one or a few samples,
and seek to distinguish between types of tissue in the imaged area, with the
aim of revealing information about the physical structure of the sample. The
infrared image is processed to give a false-colour image in which chemi-
cally similar cells are represented by the same colour. To put it another way,
the hyperspectral image is reduced in dimensionality in an optimal way so
that it can be interpreted. The dimensionality reduction can be achieved in
a purely unsupervised (or nearly so) manner [59] by hierarchical or K-means
clustering algorithms. However, attaching meaning to the false-colour
images thus produced generally requires comparison of the image with a
visible-light image of the same sample that has been examined by a patholo-
gist. Potentially, this information could be obtained by comparison of the IR
spectra with reference spectra, but this comparison can be difficult because
the different kinds of tissue have remarkably similar spectra.
These studies with small sample sizes have shown that FTIR-MS imag-
ing can produce information equivalent to several different types of optical
examination [60]. However, a much larger set of samples is necessary to show
that the method can be useful for diagnosis. A few studies have used large
sets of data, e.g., 57 or 20 samples [61-64]. These studies were all by the same