earticle

영어

Breast cancer is the leading cause of death among the women worldwide. Early detection and proper treatment may reduce the mortality rate and ultrasound imaging is used as a complimentary with the mammography to detect and diagnose the tumor. During the inspection of thick and dense breast tissue ultrasound imaging gives better tissue characterization than mammography. Detection and demarcation of tumor is acquired through segmentation process, but it is challenging due to some inherent artifacts (speckle, attenuation) of US image. Here in this paper, new segmentation algorithm has been proposed with three stages. It has also discarded the overhead of preprocessing (image enhancement, and noise removal techniques) used by traditional and other algorithms discussed in the literature. In the first stage of the proposed algorithm, probability image and its histogram equalized image are obtained from the original US image. In the second stage, clustering process is implemented on probability and histogram equalized image separately using KMCG, KFCG and augmented KMCG/KFCG codebook generation algorithms. Further, these clusters are merged sequentially one-by-one. In the third stage, post processing has been implemented on the selected merged cluster to obtain resultant segmented image. Complexity analysis is done for all codebook generation algorithms used for clustering and the results are compared. KFCG is found to be the fastest algorithm amongst all. Eventually in consultation with the expert radiologist all segmentation results are compared with each other and best results are displayed with red border.