Chemistry Reference
In-Depth Information
+
0.2
Colon
Gastric
Esophagus
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
Class
Figure 21.5
Box plot of initial feasibility for classification of blood samples' IR spectra according to the cancer
or healthy situations.
feasibility of classification between spectral data. In this process two classes are formed consisting of a
similar number in a group (e.g., 15 spectra of normal cases in class A and 15 spectra of malignant cases in
class B). Then a box plot is obtained by ANOVA, and by comparing the box of each class and also the
variance values the possibility for classification is concluded. Figure 21.5 shows the cumulative box plot for
all three cancer cases. More distance between the mean values of the boxes of each class will ensure the
researchers to better differentiate the ability of the chemometric models.
Usually a size reductive process such as PCA is also needed to provide the most informative set of data to
be classified. It is mentionable that PCA itself has also been reported as a classification strategy. Of course in
case of cancer diagnosis, according to the complexity of spectral features and huge amount of variations, the
simple PCA would not be able to produce any noticeable result for detection of healthy or malignant cases.
Score plots obtained by performing PCA on IR spectra of the organs' tissue samples which are shown in
Figure 21.6 confirm the lack of differentiation. Linear discriminant analysis of spectral data obtained from
samples can also be effective in reliable diagnostic classification of tissue samples. Application of LDA for
this aim has been reported as a very effective strategy for diagnosis of cancer. As shown in Figure 21.7, LDA
of IR spectral data from colon, gastric and esophagus tissue samples confirm this fact. Considering IR
spectroscopic analysis of tissue sample as a useful biodiagnostic approach, one needs to examine the statistical
parameters which are very important for clinical decision making. Two of the most famous are sensitivity and
specificity which are dealing with method capabilities as:
Sensitivity:
probable infection of the sample to the illness pattern while the analysis response is positive:
Number of True Negatives
Number of True Negatives
+
Number of False Positives
Specificity:
probable negative response for the case which is healthy:
Number of True Negatives
Number of True Negatives
+
Number of False Positives
