Classification of breast tumors in digitized mammograms

An automatic computer-aided diagnosis algorithm to classify malignant and benign tumors in digitized mammograms was developed. Images selected from the publicly available data set "Digital Database for Screening Mammography" were used for evaluation of this algorithm. A new segmentation algorithm was used to segment the lesion. To represent spiculation we used features previously developed for mass detection. Features based on margin sharpness, location and contrast were also extracted. A k-nearest neighbor algorithm was used for feature selection. The average test results corresponded to A(z) value of 0.84 and 0.79 in a case and in an image based classification scheme respectively.