Segmentation Of Ultrasound Image And Its Application Based On Grouping Bandlet

In recent years, the number of cancer cases and deaths from cancer are increasing. For instance, breast cancer has become one of the leading mortality causes among women in the word, especially in Europe and North America. However, if detected at early stage, most cancers can be cured by economic and simple treatments. Accurate lesion segmentation is critical in cancer early detection and diagnosis systems and it has attracted continuous research efforts.Medical ultrasound imaging is widely used in medical diagnosis and treatment of tumor lesions due to its unique characteristic of noninvasiveness, real-time, easy repetition, high sensitivity, and cheapness. It has high potential in real-time monitoring, quantitative analysis, and treatment scheduling and so on. However, due to the intrinsic speckle noises and the issue-related textures, the segmentation of tumor lesions in ultrasound images remains a worldwide difficult problem. Based on the above facts, this paper does some researches on the ultrasonic imaging mechanism and ultrasound image segmentation to improve current methods or propose new algorithms followed by both theoretical and experimental verification.In this paper, a comprehensive overview on segmentation of tumor lesions of ultrasound images is presented firstly, followed by an introduction to the principle of ultrasonic imaging and medical ultrasound treatments. Based on the mentioned above, a novel approach based on adaptive nonlinear coherent diffusion (ANCD) is adopted for reduction and coherence enhancement of ultrasound images. The limitation of the original nonlinear coherent diffusion model (NCD), such as the sensitiveness to the parameters, can be successfully overcome by employing the unsupervised classification algorithm that makes NCD model become more practical and robust. Experimental results show that ANCD only has strong ability in de-noising, but also can keep the edge characteristic of ultrasonic images perfectly well. Compared with the popular de-noising methods, such as SRAD, the proposed model has better de-noising effects.In the phase of extracting lesion, combining DWPF Grouping Bandlet proposed in this thesis, the improved stochastic neighbor embedding (ISNE), and SVM, and so on, a novel lesion extraction of breast tumor in ultrasound images is approached, which is effective and excellent extraction method. Applications to clinical ultrasound images with fibroadenoma or fibrocystic lesions contaminated by speckle noise, and ultrasound images with complex cyst of the breast are performed, respectively. Experimental results show that the proposed method is robust to speckle noise and demonstrates superior performance on the effectiveness when compared to the state-of-the-art approaches to breast lesion extraction for medical ultrasound images.