Segmentation of breast cancer lesion in digitized mammogram images

Segmentation or abnormality detection is an essential step in mammographic computer-aided diagnosis (CAD) systems. This paper presents a novel computerized method to automatically detect mass lesions (i.e. detect suspicious locations containing abnormalities inside the breast area) on digitized mammogram images. In particular, we implement an enhanced version of the region growing algorithm for segmentation of mass lesions that can be implemented in a complete CAD system. The proposed algorithm uses region growing technique with a novel automatic threshold estimation method to detect and segment mass lesions. The proposed algorithm detects masses by analyzing a single view of the breast (i. e. Medio-Lateral oblique (MLO) view or Cranio-Caudal (CC) view). The performance of the proposed algorithm was evaluated using two mammogram databases from two different hospitals. The matching percentage of the segmented regions obtained by the proposed algorithm is 83% with respect to the ground truth (i.e. reference determined by an expert radiologist). The proposed algorithm showed promising performance when compared with other commonly used segmentation techniques.