The Research Of Graph-based Segmentation Method For Breast Lesions In 3D Ultrasound Images

Breast cancer is one of the most common tumors in women which cause serious damages to women’s health, therefore, the early diagnosis and treatment for breast cancer can not be ignored. Because of its advantages of non-invasive, real-time performance, low cost, and convenient for the patient, Ultrasound imaging is one of the non-invasive diagnosis methods and it has become the chief method for the diagnosis of breast tumors. Influenced by the difference of theoretical basis, experience and practice, if clinicians only use the ultrasound images to analyze breast tumor, it would lead to the diagnosis with high subjectivity and uncertainty, and even cause misdiagnosis and missed diagnosis. With the rapid development of computer technology and medical image processing technology, computer-aided diagnosis has been developed, it can provide more objective scientific diagnosis and stable, accurate diagnostic rate through a combination of medical imaging and computer technology. But ultrasound image has the inherent defects of high inherent speckle noise, poor contrast, which increase the difficulty of the computer-aided diagnosis system. This paper devotes to research a graph-based segmentation method, which can segment the breast tumors in three-dimensional ultrasound images in high precision, and lay a foundation for subsequent illness diagnosis and treatment.Image segmentation is a critical and essential step of computer-aided diagnosis system. The segmentation technology of traditional two-dimensional ultrasound image can only display the information of a proflie of human tissues, and loss some spatial information, this requires clinical doctors have solid clinical knowledge as well as the sense of space position, hence the result of diagnosis always has the doctor’s subjective judgment. The segmentation technology of three-dimensional ultrasound image can solve this problem effectively, the three dimensional medical ultrasound images can expand the overall vision in a certain extent, show the three-dimensional contour of the diseased tissue. A segmentation method for extracting objects of interest(OOI) in 3D breast ultrasound images is convenient to the subsequent diagnosis and treatment of breast tumor. This paper is devoted to the study which aims to achieve more accurate results for breast tumors in 3D ultrasound images by extending the classic two-dimensional graph-based segmentation method to 3D space. In this paper, by taking into account the smoothness and distributions of the subregions(denoted by subgraphs in the graph-based segmentation) to be merged, we make further modifications to the PRCP(Pairwise Region Comparison Predicate) of RGB algorithm and get more accurate segmentation results for breast tumors in 3D ultrasound images.The main contribution of this article is apply the graph-based segmentation method which based on the principle of MST(Minimum Spanning Tree) to the segmentation problem of breast tumors in three-dimensional ultrasound images. Based on the characteristics of image segmentation, we make further modifications to the PRCP of RGB algorithm to get the results of segmentation more reasonable and efficient. We test the proposed method using 10 groups 3D breast US data, in which 2 are diagnosed to be benign and the remaining are malignant. The experimental results demonstrate that, whether for benign or malignant tumors, the graph-based segmentation method outperforms the traditional Snake based and FCM clustering methods on the segmentation accuracy rate.