The Research On Ultrasonic Image Segmentation Method In High Intensity Focused Ultrasound

High intensity focused ultrasound, as a new ultrasound image guidance radiation treatment, has been widely used for the therapy of benign and malignant tumors tis-sues and organs. HIFU has some advantages such as local noninvasive treatment, less postoperative side effects and short postoperative recovery period etc. As one of the key technologies of HIFU, the precision guidance of lesion areas in the ultrasonic im-ages directly affects the efficiency of HIFU preoperative planning and the accuracy of intraoperative focus the target. Until now, this work mainly relies on the clinical doc-tor manually draw the lesion areas in the ultrasonic image, and then guide the HIFU radiation therapy, which greatly reduces the efficiency of the preoperative plan and treatment effect. Due to the inherent characteristics of ultrasonic imaging such as low signal-to-noise ratio and inhomogeneous intensity distribution, the accurate segmenta-tion for the lesion areas is still a difficult task. In addition, in the process of ultrasonic imaging, the deformation of the soft organs caused by patient's breathing or external force makes the geometric shapes more complex in the images. These factors cause that achieving accurate ultrasonic image segmentation is a challenging issue in HIFU ther-apy. In this paper, the ultrasonic image segmentation algorithms are deeply researched in the background of HIFU therapy, and the some innovative achievements have been achieved.A MRI statistical deformable model based Ultrasound image segmentation method is proposed. In this method, we utilize the MRI's advantage that has more clear imaging for the soft tissues than ultrasonic images, introduce the physical and mechanical model and statistical analysis to detailed analysis the target's deformable law in the MRI in the pre-operation and post-operation, and then build the 2-D shape deformation model. This model is integrated into the framework of active shape model segmentation in order to improve the robustness of the deformable model resists the blurred boundary. For realizing the automatic segmentation, the method also uses saliency detection to initializing the location of deformation model, consequently more enhances the efficiency of segmentation.A novel dynamic statistical shape model of the ultrasonic image segmentation method is proposed. This method utilizes the characteristic of nonlinear dynamic model to study the variable low of target shapes and construct the statistical shape prior. And then, according to object intensity changes from the inside and outside of object boundary object boundary appearance model is developed, which drives the deformable model to search the object boundary. The experimental results show that using the properties of nonlinear dynamics to make the priori knowledge of target shape can well model the details of the complex shape, which is benefit to narrow the search scope of the deformation model and improve the accuracy of segmentation.A shapes similarity based ultrasonic image segmentation method is proposed. The most characteristic of this method is to use the object shapes similarity among the pa-tient's HIPU ultrasonic image sequence to construct the shapes prior, this shapes prior need not large training samples and reflect the segmenting object shape information in the greatest extent? and then this shapes prior as a constraint term is integrated into active contour model to segment the ultrasonic image sequence. The experimental results show that the shapes similarity based active contour can effectively resist the boundary leak and blurred edge in the ultrasonic images.A sparse local features and shape similarity based the active contour segmentation method is proposed. This method introduces the sparse representation theory to model the object boundary and background local appearance features respectively, and uses the local appearance features to build an object boundary feature complete model. This feature complete model is integrated into active contour segmentation framework to enhance the robustness of active contour to resist the fuzzy edge or weak edges. In addition, the method introduces the sparse subspace clustering theory to model the shapes prior among the ultrasonic image sequence and uses the L1,2 norm to measure the shapes similarity between the adjacent image among the image sequence. The experimental results show that the method can improve the robustness of active contour to resist the weak boundary or weak edge.