Image Processing Reference
In-Depth Information
classification rate and total cost are listed in Tables 6.1 and 6.2 which also provide results
for a conventional fuzzy rule-based classifier as described in Section 6.3.1. We note that the
performance of the conventional method is constant as it does not consider the weight of
training patterns. Table 6.1 shows the performance for the training patterns and Table 6.2
that of the test dataset. From there we can see that in two of the three cases there is a
clear improvement, both in terms of overall cost (the main aim of the proposed classifier) as
well as in classification performance. For both the benign-focussed and class-proportional
cases the cost is more than halved as compared to a standard fuzzy classifier. In turn the
classification rate is improved from 89.98 to 92.97 and 94.38 respectively.
Classificationrate(%) Cost
Proposed Conventional Proposed Conventional
Benignfocussed 92.26 89.28
2.74
6.04
Malignantfocussed 72.37 89.28 7.89
3.12
Class-proportional 93.86 89.28
2.88
6.05
TABLE 6.1
Experimental results for training patterns on breast cancer diagnosis.
Classificationrate(%) Cost
Proposed Conventional Proposed Conventional
Benignfocussed 92.97 89.98
24
57
Malignantfocussed 72.58 89.98 78
28.5
Class-proportional 94.38 89.98
26.31
57
TABLE 6.2
Experimental results for test patterns on breast cancer diagnosis.
The experimental results for the weighted classifier with integrated learning described in
Section 6.3.3 are presented in Table 6.3. Except for the malignant focussed case the total
cost of the weighted classifier with learning is now lower in all cases as compared to classifier
without learning strategy.
Classificationrate(%) Cost
Withlearning Conventional Withlearning Conventional
Benignfocussed
89.99
89.98
45
57
Malignantfocussed 88.06
89.98
37
28.5
Class-proportional
91.74
89.98
39.65
57
TABLE 6.3
Classification results on breast cancer dataset after learning.
6.4.2 Breast cancer classification based on thermograms
The second application for computer-aided diagnosis of breast cancer is related to thermal
medical imaging. This method uses a camera with sensitivities in the infrared to provide
a picture of the temperature distribution of the human body or parts thereof. It is a non-
 
Search WWH ::




Custom Search