【作者】 段先华；

【导师】 夏德深；

【作者基本信息】 南京理工大学 ， 模式识别与智能系统， 2005， 博士

【摘要】 医学图像处理的主要研究方向有图像分割、图像配准、结构分析、运动分析等，而其中医学图像分割的研究具有重要意义。由于医学图像具有复杂性和多样性，成像质量受到医学影像设备成像技术的影响，医学图像具有局部边缘模糊、存在噪声、区域灰度不一致性等特点，使得医学图像分割更加困难。目前，人们在继续关注对医学图像进行自动分割的同时，对交互式分割方法的研究也成了医学图像分割的研究重点。本文对交互式live wrie、Intelligent Scissors模型、参数活动轮廓模型、几何活动轮廓模型进行了分析；对可形变模型在心脏核磁共振图像中的分割进行了研究；对光流技术理论模型进行了分析，并用光流技术对心脏运动进行评估方面进行了研究。 分析了交互式图像分割的两种方法，即lire wire方法及Intelligent Scissors(IS)方法，并指出了这两种方法在分割图像时存在运算速度慢、操作复杂的缺陷。 传统的Snake模型分割图像时，要求初始轮廓线设置在感兴趣区域的边界附近，曲线在变形过程中难以分割深度凹陷区域。本文在分析参数活动轮廓模型及其改进模型的基础上，提出了基于距离均衡化的自适应动态轮廓模型。通过顶点到其邻点的平均距离作为Snake模型的弯曲力，通过图像自身的特征构造的膨胀力，使得该模型具有自适应能力，同时提出了距离均衡化的概念；对改进后模型的稳定性、光滑性及处理畸变物体的凹陷区域进行了深入分析。用该模型分割心脏MRI图像时，取得了较好的效果。 根据心脏MR图像的特点，提出了先对心脏MR图像进行K均值聚类，把K均值聚类后的图像作为特征图像，在特征上用Song and Chan提出的快速分割方法进行粗分割，再用粗分割的曲线作为水平集的初始曲线，在心脏MR图像上用Chan和Vese方法进行细分割；同时对Chan和Vese提出的简化Mumford-Shah模型进行了改进，使得每次迭代中对能量函数的优化更具有全局性，大幅度减少迭代次数，并对Song和Chan快速算法中扫描图像的区域也进行了改进，两方面的改进提高了计算速度和分割效果。分割实验证明，该方法能够快速、准确地分割心脏MRI图像。 针对左心室外轮廓类似椭圆的特点，提出了基于先验形状信息的水平集模型。该模型在Chan-Vese模型的基础上增加椭圆形状约束项，来控制曲线的演化，将水平集的演化曲线作为对轮廓新的位置预测，并用椭圆对预测结果进行修正，把预测结果和修正结果分别作为新的水平集曲线和形状信息。实验表明，这种将椭圆形状约束引入水平集模型的先验形状信息的水平集模型对心脏外轮廓的分割是有效的，非常适合分割象心脏这一类医学图像。 在分析带标记线的心脏MRI特点的基础上，提出了利用Gabor滤波和先验形状的水平集模型相结合对带标记线的心脏MR图像进行分割的方法。该方法通过对带标记更多还原

【Abstract】 There are image segmentation and matching, structure and motion analysis ect, in main research directions of medical image managing. It has important significance of research into medical image segmentation among them. Because of the complexity and variability of medical images, the image definition will be affected by the existing medical equipment techniques. Apart from that, medical image segmentation becomes even more difficult due to regional illegibility, noise, and different intensity. Currently, medical image automatic segmentation draws continuous attention. Meanwhile, researchers begin to focus on exploring interactive segmentation methods. Analyzing from interactive live wire, Intelligent Scissors model, the parameter active contour model, and geometric active contour model, this topic studies the segmentation of the cardiac MR image by adopting deformable models, and optical flow technology theory and its application in cardiac movement estimation.It analyses two methods of interactive image segmentation, that is, methods of live wire and Intelligent Scissors (IS). Then points out their defects of slow calculating speed and complicated operating while image is segmented.When the traditional parametric active contour model (Snake) is applied to segment the image, an initial contour must be set near the boundary of ROI and the model cannot segment deeply concave regions accurately. On the basis of analyzing the Snake model and improved model, the adaptive active contour model based on distance equalization is proposed. This new model, through the average distance between the vertex and the line of its neighborhood dots as the flexure of Snake model, the inflation is defined through from characteristics of image itself, make the model has its adaptive ability. Meanwhile, the concept of distance equalization also has been proposed. The stability and smoothness of the improved model and the management of deformed concave regions have been deeply analyzed. The segmentation experiments demonstrate the effectiveness of improved model for the cardiac MRI.Based on the characteristics of cardiac MR image, the topic proposes a faster segmentation process. A rough segmentation is made by using Song and Chan method to the feature images, which are the result of k-Means cluster to cardiac MR image. Then it is followed by a further segmentation of the initial level set curve that was got by the first step, based on Chan and Vese method. At the same time, improvement of the simplified Mumford-Shah model proposed by Chan and Vese is made to achieve a more global optimization of energy functions and accelerated evolution process, and a better outline of the scanned regions, which combine to improve the calculation speed and segmentation effectiveness. Experiments prove the rapidness and accuracy of the method in its application in segmentation of cardiac MRI.In respect of the ellipse-like outline of the left ventricle, ellipse shape restriction is adopted based on Chan-Vese model to control evolvement of curve. The evolvement curve of level set being regarded as the new forecast of contour location, ellipse restriction is introduced to modify the forecast result. Accordingly, the results of the forecast and modification are taken respectively as new curve of the level set and shape information.更多还原