Computational Intelligence Techniques for Classification in Microarray Analysis - Computational Intelligence in Healthcare: Advanced Methodologies - page 292

Section 5 presented the case study considered in this report, which classified 91
CLL leukemia patients into groups. The aim of the case study was to identify the
probes that allow classifying the CLL leukemia patients into subgroups. In an initial
test, data from 91 patients, where previous classification was not taken into account,
were used in the system. The pre-processing phase began with 54.675 probes and the
RMA was applied to obtain the luminescence values for each of the probes and to
homogenize the values from different chips. After the pre-processing phase, the filter-
ing process was applied, notably reducing the probes to 541, without increasing the
error rate.
Once the filtering was executed, it was still difficult to extract knowledge from the
data. Figure 2 shows the 91 individuals in a bar graph, where the bars are divided in
541 probes with amplitude proportional to their value. The upper part of Figure 2
shows the classification obtained for each of the individuals, and the bottom of Figure
2 shows the parallel coordinates that represent 561 coordinates and 91 lines for each
of the individuals.
Fig. 2. Values for the probes obtained from the individuals.
The reuse phase begins when the probes have been filtered, and generates the
meshes for the groups as well as the distribution of the individuals along the space.
The mesh closest to the new case was then selected and the classification was made.
To evaluate the proposed model, the system classified 90 individuals together with a
new individual, and the results obtained were compared to the previous existing clas-
sifications. This process was repeated for each of the 91 individuals considered for the
experiment and the results obtained demonstrate that 82 of the 91 individuals were
successfully classified.
In the revise phase, the CBR system extracted the knowledge obtained during the
classification process, as shown in Figure 3. Figure 3 presents the decision rules
obtained in the revision phase that are applied to extract knowledge from the classifi-
cation carried out in the previous phase.