The Research Of Computer-aided Diagnosis System Of Breast Tumor Based On Ultrasonic Image

Breast cancer is the most prevalent cancer among women. The fatality ratio is keeping rising in these years. Sonography has been widely used for diagnosis of breast cancer because of its non-invasive and low cost for the patients. However, it heavily depends on operator's experience, which leads to a high false positive predictive value. That means large number of unnecessary biopsies, which are painful and economical burden to the patients.Computer-aided diagnosis of breast cancer can reduce breast biopsies and improve breast cancer diagnosis accuracy and objectivity. And it also can largely reduce the doctors'work. The current computer-aided diagnosis system extracting the edge of breast tumor mainly manual or semi-manually, which burden on the work of doctors, and the effect of edge detection influenced by doctor's subjective factors. This research focused on automatically extracting the edge of the tumor and determining an optimal feature vector to improve breast cancer diagnosis accuracy and objectivity and reduce breast biopsies.Firstly, analyze the characteristics of 125 cases of breast tumor ultrasound graphic which had been pathological diagnosed. Based on the characteristics of medical ultrasound images of breast tumor, an adaptive weighted median filter is used to suppress the speckle noise in medical ultrasonic image. An improved histogram equalization algorithm to enhance the image contrast is proposed. And then extract the breast tumor initial region automatically by segmentation algorithm based on region growing. Morphology filtering and connected labeling are used to extract the exact outline of tumor. In the end, 122 sample's accurate lesion areas have been extracted. Finally, a novel approach based on the outline of breast tumor's fractal dimension is proposed to feature extraction. And then seven contour features of the tumors are extracted from the regions of interest of the ultrasonic images. Fractal dimension, circularity, boundary roughness and namely length-width ratio are selected to create an optimal feature vector through comparing the classification distance of all the features. Those features are input to a three-layer back propagation (BP) artificial neural network to distinguish the benign and malignant cases. The result is evident that the proposed method improves the accuracy of malignant and benign tumor diagnosis and achieves the purpose of reducing the rate of biopsy.