A novel approach for automatic detection of abnormalities in mammograms

This paper proposes a novel approach for the development of a computer aided decision system to automatically detect abnormalities in mammograms. In this method preprocessing of images is done by enhancing the contrast of the intensity image by transforming the values using contrast-limited adaptive histogram equalization (CLAHE). Then, Mathematical morphology is used for the extraction of abnormalities which are located on a non-uniform background. After performing the thresholding of the images by the extended maxima transformation, feature extraction is focused on the extraction of both statistical and textural features of the objects. Finally the extracted objects are classified using Naïve Bayes Classifier and abnormalities are detected. This forms a basic step in the automatic detection system of abnormality in breast images and thus increases in the sensitivity of breast cancer detecting algorithms. The accuracy of the method has been verified with the ground truth given in the data base and it is as high as 82.40%. The algorithm is validated by considering the mini - MIAS´s data base, a benchmark data supplied by American Society of Radiology UK. With the help of this data base, the feasibility of the proposed method is demonstrated.