Research On Key Technologies Of Medical Image Segmentation Based On Active Contour Model

The development of medical imaging technique has extremely accelerated the computer aided diagnosis, which is based on digital imaging processing. Medical image segmentation is the procedure of acquiring the regions of interest in the picture, which is the foundation of further processing in computer aided diagnosis system, including feature extraction, three-dimensional reconstruction, disease detection, quantitative analysis, and so on.Recently, active contour model was studied and applied more and more in medical image segmentation, because of its diversified format, flexible structure and superior performance. In this dissertation, the application of active contour model on medical image segmentation was studied intensively, and the main contents and innovations are as follows:(1) In view of the traditional lung parenchyma segmentation algorithm’s inaccurate result for pathological lung CT images, this dissertation presented a novel pathological lung parenchyma segmentation algorithm based on anatomical physiology knowledge and Snake model. This algorithm used ribs’edge as initial contour, and took full advantage of the Snake model’s features, which were contour’s continuity and insensitivity to hollow cavity. Snake model was improved by adding a parameter to represent the location of ribs, which can drag the contour to ribs. This algorithm overcame the lung CT image’s abnomalities caused by pathology, such as discretion, fissure, hollow cavity, and so on.(2) In view of the lung segmentation of serial CT image, this dissertation presented an interactive lung segmentation algorithm, which is named LW-Snake model. This algorithm improved Snake model and Live-Wire model, and took full advantage of lung contours’slow variance in adjacent CT images layer and operators’professional knowledge. First, the key slices of lung parenchyma were manually selected in serial CT images, and then the lung’s contours in key slices were drawn by Live-Wire model, and the lung’s contours in other slices were acquired by contour interpolation, finally the lung’s accurate contours were segmented in all slices by Snake model and manual modification. This algorithm overcame the disadvantages of manual segmentation for serial medical images, which were overloaded, time-consuming, and unrepeatable.(3) In view of the brain white matter fibers segmentation from diffusion tensor mage, this dissertation presented an algorithm based on Riemannian manifold. First, construct a3×3symmetric positive definite covariant tensor for each voxel using diffusion tensor image, by which tensor field was constructed to illuminate brain white matter. Second, regard the tensor field as Riemannian manifold, and the fluid motion in the tensor field was expressed by Navier-Stoke equation, so the problem of brain white matter fibers between two voxels can be transformed into the computation of smallest distance between two points in Riemannian manifold. Finally, distances between two points in Riemannian manifold, which were the brain white matter fibers, can be expressed by geodesic, whose numerical solution was based on Level-Set algorithm. Compared with the conventional brain white matter fibers segmentation, this algorithm’s accuracy and robustness were greatly improved.(4) In view of the segmentation of cerebrospinal fluid, white matter and gray matter, this dissertation discarded the conventional ideas of segmenting these structures from regular MR images, and presented a novel segmentation algorithm based on the parameters from diffusion tensor image. First, diffusion and anisotropy parameters from diffusion tensor images were computed to achieve the parameter image channels. Then, using expectation maximization model, WM/non-WM regions in anisotropy parameter channels and CSF/non-CSF regions in diffusion parameter channels were calculated. Finally, using improved simultaneous truth and performance level estimation model, all parameter image channels’segmentation results were fused to achieve the final brain structures segmentation result. The most advanced point of this algorithm was components segmentation first, whole fusion latter. Compared with conventional brain structures segmentation algorithm, this algorithm’s sensitivity and specificity were greatly improved.