The Research And Application Of Breast Ultrasound Image Processing

The development of computer science and medical image processing provide a scientific basis for clinical diagnosis and biomedical research. Breast disease is the most common disease of women. Ultrasound examination became a routine examination in the diagnosis and treatment of breast diseases because of its simple operation, no trauma, and timely examination results and so on. The distinguishing and diagnosis of breast ultrasound image mainly rely on the experience of doctor. It may cause some error because of no quantized indicator. Many researchers have proposed computer-aided diagnosis system for breast ultrasound image to increase the accuracy and efficiency of the diagnosis.But most of these systems have encapsulated some common algorithms or detected or classified the breast tumor. Seldom are they used to diagnose other diseases. This paper discusses the technologies that are mainly used in these systems, image enhancement(including image denoising, gray level transform and edge enhancement), image segment and image recognization (including feature extractation and classifification algorithms). This paper also proposes a breast Ultrasound image processing system. It is developed in matlab, including a file processing module and an image processing module. The file processing module includes image reading, writing and reverting. The image processing module includes image denoising, gray level transform, edge enhancement as well as two segmentation algorithms for cyclomastopathy image, the local segmentation and global segmentation.In local segmentation, the user should firstly chooses a part of the lesion area. Then the system uses Otsu threshold method to get an approximate contour of the lesion. At last, the system performs the local binary fitting method to obtain a more accurate lesion area. In global segmentation, the system firstly applies the simple linear iterative clustering algorithm to divide the image into small blocks. Color and position, texture and SIFT features are extracted. The system utilizes the Adaboost scheme for feature selection and classification to find the breast tissue. After detection, a result correction is developed to reduce the misclassification rate. Then, a boundary extraction scheme implemented by the dynamic programming is performed to find the specific boundaries of the breast tissue. After that, the system uses an active contour model with local statistics information to segment the lesion of cyclomastopathy. When the lesion area is found, the system can analyze the lesion, such as echo and area.We have tested 56 local regions and 56 ultrasound images. The accuracy of local segmentation is 85.3%. In global segmentation, the accurate of classification of breast region after result correction id 90.1% and the segmentation is 80.1%.