Unsupervised textural segmentation of SonoElastographic breast images

Abstract

Breast Cancer, the most common malignancy in women and the second leading cause of death for women all over the world. By earlier detection of cancer, the better treatment can be provided. SonoElastography, a new medical imaging technique reduces un necessary biopsies compared to mammography and conventional ultrasonography. The diagnosis and treatment of the cancer rely on segmentation of SonoElastographic images. Texture features are widely used in classification problems, i. e. mainly for diagnostic purposes where the Region Of Interest (ROI) is delineated manual ly. It has not yet been considered for SonoElastographic segmenta tion. SonoElastographic images of 15 patients taken using Siemens Acuson Antares are considered for experimentation. The images contain both benign and malignant tumors. From the experimental procedure it is proposed that the combination of texture features, Local Binary Pattern (LBP), Contrast and Variance are best suited for segmentation of SonoElastographic breast images. The images are first enhanced using sticks filter to remove noise, to improve contrast, and emphasize tumor boundary. Then extract the features to segment the breast image. The resultant images undergo some post-processing steps to remove the spurious spots. The segmented image is thinned to mark the tumor boundary. The results are then quantified with the help of an expert radiologist. The proposed work can be used for further diagnostic process, to decide if the segmented tumor is benign or malignant.