Knowledeg-Based Automatical Segmentation Of Breast Ultrasound Images

Breast cancer is the second leading cause of female cancer death and seriously endangers the health of women. Ultrasound imaging method has non-radioactive, non-invasive and highly accurate identification of benign and malignant tumors, etc., has been widely used as an alternative method of X-ray in early diagnosis of breast cancer. In order to assist doctors improving the accuracy and objectivity of ultrasonic detection of breast cancer and reducing the rate of misdiagnosis of malignant tumors, many computer-aided diagnosis (CAD) methods have been widely used in breast cancer diagnostic process.In CAD systems of breast ultrasound, breast tumor segmentation is one of the most important and is also one of the most difficult tasks. In order to improve the objectivity and accuracy of CAD systems, and lessen operator intervention, automatic segmentation method is the current hotspot. Automatically and accurately locating region of interest and efficiently and accurately segmenting tumors are two difficult problems in automatic segmentation algorithm.In this paper, we study the anatomical structure of the breast tissues and their ultrasound echo characteristics for dealing with the important problems in automatic segmentation. We establish the location model and the appearance and space connectivity model. We proposed an automatic tumor detection algorithm and an automatic segmentation algorithm into which prior knowledge based constraints have been integrated under the maximum a posteriori probability with markov random field(MAP-MRF) segmentation framework. The main contents include the following two aspects:First of all, several current methods of automatic locating region of interest(ROI) are dependent on certain constraints which are hard and not flexible, and can not effectively distinguish the tumor from the low gray area. To solve these problems, we propose a fully automatic segmentation methods based on the constraints from breast anatomical structure and location of the tumor. This method can adaptively and automatically obtain the global reference point according to image intensity distribution and locate the seed point using Mean Shift Algorithm. Combining the automatic thresholding with the obtained seed point, we can locate the ROI.Secondly, this paper presents a fully automatic segmentation method based on the automatic locating ROI method under MAP-MRF segmentation framework. We chose gray level distribution and location prior and established the location model and appearance and fuzzy connectedness model. Experiments show that our automatic segmentation method can effectively eliminate interference effects of the fat tissue and artifacts, can accurately locate the tumor boundary.