Automated breast masses segmentation in digitized mammograms

Abstract

In this paper, an automated segmentation method is proposed. The method is applied to the segmentation of breast masses in digitized mammograms and it operates on the whole mammograms instead of manually selected regions. Pixels with local maximum gray levels are flagged as seeds, from which many candidate objects are grown using modified region-growing technique. Following which false positive (FP) reduction using decision tree is applied to discard the normal tissue regions. A total of 40 mammograms from mammographic image analysis society (MIAS) are analyzed. 36 masses are correctly segmented by the proposed method, resulting in 90% true positive rate at 1.3 FPs per image.