Infrared and Raman Spectroscopy and Imaging of Breast Cancer - Vibrational Spectroscopy for Tissue Analysis - page 155

linkage) to generate false-colour images, which were compared with the con-
ventional microscopic images.
A special issue of Vibrational Spectroscopy devoted to the conference
Shedding Light on Disease: Optical Diagnostics for the New Millennium
included several papers dealing with infrared microspectroscopic imaging
(including those described in the next three paragraphs).
Beleites et al. [61] used FTIR microspectroscopic imaging to distinguish
among different types of gliomas (brain tumours). Ten-micron slices (area
~0.5 cm 2 ) were cut from frozen samples and mounted and dried on CaF 2
windows. Parallel slices were mounted and stained on slides for histological
evaluation. They used linear discriminant analysis (on the whole dataset of
59 samples) in conjunction with a genetic algorithm for wavelength selection.
The model was assessed by cross-validation. The classification for each sam-
ple was obtained by aggregating the predictions for its individual spectra.
Their method was able to distinguish between cancer and noncancer with
~95% accuracy, but the prediction of the grade of cancer was much worse.
Romeo and Diem [68] used a different sampling method. To reduce costs,
they used reflection-absorption spectroscopy with thin samples (6µm)
mounted on silver-coated glass microscope slides (low-e slides—cheaper
than IR windows). This approach can result in reflectance artefacts (super-
imposed dispersion-shaped bands in the spectra), particularly around the
edges of the sample where it may not adhere as well to the slide. They pro-
pose an interesting (and very simple) correction method that appears to
work quite well [69].
Ó Faoláin et al. [70] discuss the effects of tissue processing. They compared
spectra from fresh frozen samples to those from wax-block archived samples
using both Raman and FTIR spectroscopy (not imaging). They list sample
types as: fresh, frozen, air-dried, formalin-fixed, and de-waxed formalin-
fixed paraffin preserved. They found that there are changes in the spectra
depending on the treatment: bands due to formalin/wax were evident even
after standard de-waxing procedures. In addition, there were more subtle
changes in the spectra, for which they provided some partial explanations.
Amharref et al. [60] studied rat brain tumours with FTIR microspectro-
scopic imaging. Cryosectioned samples were mounted on ZnSe windows.
First derivative and SNV preprocessing was used, and K-means clustering.
They were able to distinguish between several different kinds of brain tissue
based on the infrared spectra. In fact, the infrared map contained information
equivalent to several separate samples stained with different methods, but
they do not present any cross-sample results as described by Beleites et al. [61].
Fabian et al. [72] used IR microspectroscopy to distinguish between benign
(fibro-adenoma, FA) and malignant (DCIS) breast tumours, as well as adipose
and connective tissues. They had a large set of spectra from 22 patients, and
spectra were classified histologically to provide the Y-block information. An
ANN was used for classification, and worked fairly well. Initially, there was
some misclassification among the tumour types. The ANN was replaced with