The Research Of Image Segmentation Based On Deformable Model

Medical image segmentation plays an important role in medical imageprocessing and a wide range of application, particularly in diagnosing diseases andevaluating the effects of therapy. Deformable model based segmentation turned out tobe one of the most successful techniques for incorporating prior knowledge, which iswidely used in medical image analysis. This paper first introduced the principle of thedeformation model and its application, and then described the active shape model indetails.Statistical shape modeling is a critical factor affecting the performance ofdeformable model based segmentation methods for organ shape extraction. Organshape plays an important role in clinical diagnosis, surgical planning and treatment.However, statistical shape modeling of the deformable based segmentation algorithmsin most existing works did not consider two limitations: first, most of these statisticalshape modeling were population-based shape modeling, and did not take thespecificity of the patient; second, shape modeling is completed in the original imagespace containing “noise” effect, so it did not reflect the intrinsic similarity betweenthe target shape and the training shapes. This paper proposed a patient-specific shapemodeling to deal with these two problems in a unified framework. Manifold learning,nonlinear dimensionality reduction algorithms, can reflect their intrinsic structure.The proposed method measures the intrinsic similarity between the target shape andthe training shapes on an embedded manifold by manifold learning technique. Thedata obtained from the real world may not always meet the manifold assumption. weexhibited the structure-constrained target-oriented shape prior estimation algorithmwith considering manifold assumption. With this approach, shapes in the training setcan be selected according to their intrinsic similarity to the target image. With moreaccurate shape guidance, an optimized search is performed by a deformable model tominimize an energy functional for image segmentation, which is efficiently achieved by using dynamic programming.Our method has been validated on2D prostate localization and3D prostatesegmentation in MRI scans. Compared to other existing methods, our proposedmethod exhibits better performance in both studies.