Segmentation Of Ultrasound Image Based On GVF Snake Model

In the process of cardiovascular ultrasonic diagnosis, the morphological character of heart are important information for disease detection, such as cell wall thickness, transverse diameter, range and direction of motion, area, anomalous structure of anatomize, and so on. Those information are used by most doctors for the heart disease diagnoses, or provide evidence for anatomic structure and function evaluate of cardiac. As the basic factor of computer assisted diagnosis and analytic system, image segmentation can provide information for image analysis system and computer assisted diagnosis. For the complexity of human organs and tissue structure, in ultrasonic imaging, it will be represented as the difference of acoustic impedance, as well as the reflection, refraction and dispersion of the beam of ultrasonic light, followed by the interfererence of sound wave info. All of the above badly reduce the robust of segmentation algorithm.This thesis aimed at the common initialization problem of parametric snake model under noise condition, research and improve the external force field of GVF snake. According to analysis of the external force field of GVF, it conclude the reason for contour error location and noise disturbing effect. Base on above discussion, MCVF gradient vector flow combine with mean curvature flow and anisotropic diffusion are proposed in this thesis. The infection of local minimum and the sensitivity of initialization position have been reduced by tangential diffusion, adaptive diffusion weight parameter and subsection function control. It has larger capture range under the speckle noise condition of ultrasound image, enhance the robustness of GVF model.Finite difference method, the traditional deformation method for GVF snake model, is difficult to realize under VC platform. Based on the improved GVF external force field, internal force constructing and energy searching are discussed in the thesis. A new form of internal force which match with the external force has been generated, and a global energy searching deformation algorithm has been proposed. According to three kinds of extern-inner maximum energy point criterion, the maximum energy point is identified as split point between objects, based on above principle, the topologic function of GVF snake model is discussed in the paper as well.This thesis realized the segment algorithm of improved GVF snake model with Matlab and VC. According to the experiment on artifact images and ultrasound images, the GVF snake model improved by this paper is less sensitive to noise, based on global energy searching and larger initialization range, which is proved to be effective for ultrasound image segmentation.