Page 207 - Vibrational Spectroscopic Imaging for Biomedical Applications

W idefield Raman Imaging of Cells and T issues 183

Raman images, which uses ordinary least squares to fit a set of basis
spectra to the data. The basis spectra are acquired from the major
morphological features found in a set of representative samples in
situ using a Raman confocal microscope. The technique of morpho-
logical modeling was compared with other chemometric techniques
such as PCA, multivariate curve resolution (MCR) and ED. Point
mapping was used to obtain Raman spectral images of biological
samples in the study. They found that each technique has advantages
and disadvantages. PCA and MCR are excellent techniques when
nothing or little is known of the samples beforehand. When some
information about the sample is known, ED is useful. From morpho-
logical modeling, it is possible to obtain structure and chemical
information about subcellular features in a biological sample, although
the sample under study must be well understood.
Omberg et al. used principal factor analysis in their study of
77
tumorigenic and nontumorigenic cells. Factor analysis (FA) is a sta-
tistical method used to explain variability among observed variables
in terms of fewer unobserved variables called factors. FA differs from
PCA in the fact that PCA is used to find optimal ways of combining
variables into a small number of subsets, while FA may be used to
identify the structure underlying such variables and to estimate
scores to measure latent factors themselves. FA was used to evaluate
the Raman dispersive spectra obtained from two different rat fibro-
blast cell lines: M1 and MR1. Both cell lines were transfected with the
c-myc gene which causes immortality; however, the MR1 line was fur-
ther transfected with the T24Ha-ras oncogene which causes tumor
formation. Two approaches were used in the study. First, constrained
principal factor analysis provided a measure of the relative contribu-
tion of protein, lipid, DNA, RNA, and buffer to the Raman spectra in
each cell line. When using constrained FA to analyze the spectra of
the two cell lines, the increased intensities in specific band assign-
ments were interpreted as increased protein and lipid concentrations
in the tumorigenic cells (MR1). Second, FA of the raw spectral data
was used to demonstrate that Raman spectra can be used to differen-
tiate the two cell lines. Similar to constrained FA analysis, the spectral
differences between M1 and MR1 cells were also attributed to
increased protein and lipid relative to DNA; additionally, the ratio of
scores of the appropriate factors shows a clear distinction between
the M1 and MR1 spectra.
Nijssen et al. used K-means clustering analysis (KCA) in the
evaluation of basal cell carcinoma (BCC) from surrounding tissue by
78
Raman spectroscopy. Raman pseudo-color images were constructed
from the spectra obtained from each sample by PCA, and then
the PCA scores were used as input for KCA. KCA was used to
find groups of spectra with similar spectral characteristics (clusters).
The main steps of KCA are as follows: (1) the number of clusters is
set by the user; (2) for each cluster a spectrum is randomly chosen