Segmentation of suspicious clustered microcalcifications on digital mammograms: using fuzzy logic and wavelet coefficients

We proposed an automated segmentation of suspicious clustered microcalcifications on digital mammograms. The algorithm consists three main processing steps for this purpose. In the first step, the improvement of the in microcalcifications appearance by using the "a trous wavelet" transform which could enhance the high-frequency content of breast images were performed. In the second step, individual microcalcifications were segmented using wavelet histogram analysis on overlapping subplanes. Then, the extracted histogram features for each subplane used as an input to a fuzzy rule-based classifier to identify subimages containing microcalcifications. In the third step, subtractive clustering was applied to assign individual microcalcifications to the closest cluster. Finally, features of each cluster were used as input to another fuzzy rule-based classifier to identify suspicious clusters. The results of the applied algorithm for 47 images containing 16 benign and 31 malignant biopsy cases showed a sensitivity of 87% and the average of U. 5 false positive clusters per image.