Background: Breast cancer is diagnosed as the most leading and dangerous cancer in women
all over the world. Recent patents have shown that breast cancer is the second leading cause of death
worldwide, among other cancers. This paper presents a development of breast cancer diagnosis system in
digital mammograms using multiresolution technique.
Methods: The proposed method uses the Ripplet Transform (RT) which holds potential properties for
feature extraction. The input mammogram image is first decomposed with the ripplet transform at different
scales. Then the statistical features are extracted from each scale that is used as a feature vector. The
Support Vector Machine (SVM) classifier is used as a classifier to distinguish between normal and abnormal
and to classify the abnormality between benign and malignant images.
Results: In this paper, the application of multiresolution based ripplet transform in feature extraction for
mammogram classification has been demonstrated. A comparative analysis is performed with the features
extracted from Gray Level Co-occurrence Matrix (GLCM), Discrete Wavelet Transform (DWT) and
Curvelet Transform (CT). The experimental results demonstrate that the ripplet transform based feature
extraction is an efficient and promising tool for the successful classification of digital mammograms. The
average classification rate achieved for normal and abnormal is 94.41% with curvelet features, 92.68%
with wavelet features, 91.75% with GLCM features. RT exploits the multiscale property along with high
degree of directionality so it achieves relatively higher average classification rate of 95.08%. The average
classification rate obtained abnormalities between benign and malignant through ripplet transform coefficients
is 95.56%. Also the average classification rate obtained is 94.17% for curvelet, 93.61% for wavelet
and 90.28% for GLCM.
Conclusion: In this paper, the advantages of ripplet transform in mammogram analysis are exploited and
a new model is proposed using a ripplet transform for feature extraction and classification of mammogram
images. The statistical features extracted from the ripplet transform coefficients are employed in the
SVM to classify the ROI into normal and abnormal and to differentiate abnormality between benign and
malignant. When compared to other multiresolution transforms, ripplet offers improvement by representing
the images with singularities along smooth curves. The experimental results show that the proposed
method using the ripplet transform coefficients achieves relatively improved classification rate than the
other multiresolution feature extraction methods.