Detecting and Monitoring of Cancer at the Cellular Level - Vibrational Spectroscopy for Tissue Analysis - page 169

Table 6.3
Comparative Analysis of Quantitative Measurements

Mathematical
Diagnosis Type of Samples Peak 1 Position Peak 2 Position Ratio of Heights

S 878-882 934-939 1.04-2.42 <2

R 881-886 936-945 5.94-13.69 >6

Ratio of Areas

S 878-882 934-939 1.72-3.21 <3

R 881-886 936-945 6.21-20.59 >6

The results of peak height ratio calculations show that there are significant
quantitative differences between R and S samples. In spite of minor differences
in the position of the peaks, the main dissimilarities are in the ratios. Whilst
almost all of the height ratios of S samples are less than 2, this number is more
than 6 for R samples. Another difference is the method of distribution of the
quantities. The ratios of R spectra are very widely distributed, ranging from
5.94 to 13.69. On the other hand, S numbers are all in a relatively narrow range
of 1.04 to 2.42. Peak areas, in a very similar way, illustrate the quantitative dis-
similarities. Whilst almost all the ratios of S samples are less than 3, this num-
ber is more than 6 for R samples. The method of distribution of the quantities is
also different. The ratios of R spectra are very widely distributed, ranging from
6.21 to 20.59. On the other hand, S numbers are all in a relatively narrow range
of 1.72 to 3.21. These results demonstrate that in proteins of resistant cell lines,
there are much more ρ(CH 2 ) and ν(C-C) vibrations in comparison with ρ(CH 3 )
and ν(C - C) vibrations. Table 6.3 summarizes these findings. Similar results
can be obtained if peak areas are measured instead of peak height.

Statistical analysis

The results of cellular studies might not be particularly useful if there is no
statistical data to support them. As a result, statistical analysis is used to
provide supportive data to spectral findings.
There has been much interest recently in using vibrational spectroscopy
and computer-based pattern recognition techniques to discriminate between
different types of tissue and tumours [22,23]. These techniques could poten-
tially complement traditional histopathological diagnosis by providing a
rapid and objective analysis of tissue samples. Here, we demonstrate the
ability of PLS-DA, a comparatively simple classification algorithm, to distin-
guish between the R and S cell lines.
PLS-DA involves two steps. In the first step, a PLS decomposition of the
spectral and class data is performed, reducing the large number of wave-
length variables to a small number of latent variables called scores [24]. I n t he
second, these scores are used as descriptor variables in a standard linear
discriminant analysis, or LDA [25]. This approach is generally believed to