A Novel Approach for Reduction of Breast Tissue Density Effects on Normal and Abnormal Masses Classification

Breast tissue density prevents the separation of the abnormal tissue from normal tissue in mammography images due to the negative effects on diagnostic success often hiding abnormalities. In this study, significant reduction of these adverse effects is provided by proposing a complete system including also breast tissue density classification. The breast tissue density type of image, which will first be subjected to normal and abnormal tissue classification, is classified with the proposed system. For the breast tissue density classification, artificial neural network (ANN) and the multiresolution analysis, which were previously proposed in the literature, were used. At the second stage, mammography image was subjected to the classification of normal and abnormal tissue by using trained ANN with the other images which have the same type of breast tissue density according to breast tissue density classification results. Wavelet transform, ridgelet transform and contourlet transform were used at this stage in obtaining image features of mammography. In order to test the success of the proposed system, 265 pieces of region of interests belonging to MIAS database were used. At the end of the study, the highest accuracy is 95.472%, sensitivity is 0.8514, specificity is 1 and A(z) is 0.960. These results can be further increased with semi-automatic operation of the system by performing the classification of breast tissue density by the radiologist. The highest accuracy is 97.736%, sensitivity is 0.9324, specificity is 0.9948 and A(z) is 0.974 for semi-automatic system.