A New Image Mining Approach for Detecting Micro-Calcification in Digital Mammograms

Figures & data

Figure 1. Structure of the proposed technique.

Figure 2. (a) The original mammogram image, (b) cropped image and (c) image filtered with CLAHE.

Table 1. Initializes CLAHE.

Feature name	Value
Block size	127
Histogram bins	256
Maximum slope	3

Figure 3. (a) The histogram chart of the original mammogram image. (b) The histogram chart of the CLAHE image.

Table 2. Features identified by GLCM on MIAS images.

Feature no.	Feature name	Feature values	Feature no.	Feature name	Feature values
1	Autocorrelation	24.9451	12	Sum of squares	24.9392
2	Contrast	0.25088	13	Sum average	9.72831
3	Correlation	0.91109	14	Sum variance	63.2808
4	Cluster Prominence	80.3496	15	Sum entropy	2.11979
5	Cluster Shade	2.04246	16	Difference variance	0.5088
6	Dissimilarity	0.25019	17	Difference entropy	0.56471
7	Energy	0.13979	18	Information measure of correlation1	−0.55016
8	Entropy	2.29565	19	Information measure of correlation2	0.90836
9	Inverse difference (INV) is homom	0.87502	20	Inverse difference normalized	0.97221
10	Homogeneity	0.87497	21	Inverse difference moment normalized	0.99614
11	Maximum probability	0.23528

Table 3. Variables from Table 2 that had the greatest impact on the diagnosis and classification of a cancer image.

Feature no	Feature name	Feature values
1	Cluster Prominence	80.3496
2	Cluster Shade	−2.04246
3	Sum average	9.72831
4	Sum variance	63.2808
5	Autocorrelation	24.9451
6	Homogeneity	0.8502
7	Sum of squares	24.9392
8	Homogeneity	0.87502
9	Inverse difference normalized (INN)	0.97221
10	Dissimilarity	0.25019

Figure 4. Cloud level with maximum separating border along with separating borders, for classification of the data samples into two different classes.

Figure 5. Variables from Table 2 that had the greatest impact on the diagnosis and classification.

Table 4. Results obtained by our method.

Classify	Samples	Accuracy (%)
Correctly Classified	147	97.35
Incorrectly Classified	4	2.65

Table 5. Confusion matrix.

Actual	Predicted class Normal Cancer Benign
Benign	49	0	0
Cancer	4	48	0
Normal	0	0	50

Figure 6. The new method proposed compared with other similar methods.