Breast Ultrasound Image Segmentation Based On Homogeneity Histogram

Breast cancer continues to be a significant public health problem in the World. Early detection and therapy is the key to improving breast cancer prognosis at present. Currently, the supplemental use of sonography in screening for breast cancer is an area of active research, and has proven to be a valuable adjunct to mammography in the detection and classification of breast lesions. Being an assistant to doctors, computer-aided diagnosis (CAD) applied to early breast cancer detection is a novel method in the treatment field, which greatly improves image qualities as well as brings diagnosis enhancement.Image segmentation is the most difficult task in image processing. It's a key point in computer-aided diagnosis, and determines the quality of the final result of analysis. This paper analyzed and compared several ultrasound image segmentation methods in breast cancer diagnosis and made conclusions for their advantages and disadvantages. Then a new method of breast ultrasound image segmentation based on homogeneity histogram is proposed. Texture as well as edge features are used in the computation of homogeneity, so both global and local messages are considered, which can't be achieved by previous algorithms. The image is divided into homogeneity subset and the non-homogeneity subset according to the threshold computed from the maximum entropy principle. The two subsets are segmented separately thereafter.Radio experts affirmed the processed result, and the difference assessment experiment demonstrated that the proposed approach could generate valid nodule discrimination from breast sonographies, with valuable TP, FP and FN data. It will be a better assistant for radiologists in the diagnosis of breast cancer.