心脏核磁共振成像技术(Magnetic Resonance Imaging，MRI)作为一种非介入的成像方法，已经成为心脏疾病临床诊断的重要辅助手段。心脏核磁共振图像不仅能够提供心脏的形态结构，还可以用来分析心脏的运动和形变，定性或定量地估计心室的整体及局部心肌功能，为心脏疾病的有效诊断治疗提供科学依据。左心室作为躯体供氧的动力枢纽，它的运动情况能够反映心脏疾病的表象和诱因，是目前研究的重点。本文采用图像处理、计算机视觉、模式识别等理论和方法，对左心室自动定位、分割、三维运动与应变分析等关键问题进行了分析和研究，主要研究工作和贡献如下： (1)提出一种基于图像序列相邻帧运动差分图像的左心室自动定位方法。心脏MRI能够提供高分辨率高品质的图像，图像序列中同一成像层面相邻帧运动差分图像保留了具有较大运动位移的左心室的位置和形状信息；针对左心室近似为圆形的特点，在差分图像中采用Hough变换自动定位左心室质心和半径以及左心室兴趣区域。该方法大大缩小了目标研究范围，同时所得半径靠近左心室内膜，可以作为分割模型的初始化条件，为实现心脏MR图像的自动分割奠定了基础。 (2)提出一种基于径向立动轮廓模型的左心室内膜分割方法。该方法与传统主动轮廓模型在二维图像平面内寻找能量极小点的思路不同，它在极坐标变换图像中沿一维径向方向搜索目标边界，模型能量项由二维简化至一维，提高模型计算效率；针对变换图像中左心室的形状特点，引入相应的形状约束能量项以及区域一致性约束，提高算法的鲁棒性。实验结果表明，该方法能够高效准确地分割左心室内膜。 (3)提出一种基于改进梯度矢量流的径向B-snake模型。在径向主动轮廊模型中引入梯度方向信息，分别构造提取内膜和外膜的边缘图，计算相应的外力场，在一次snake演化过程中同时分割左心室内、外膜；使用B样条简洁表达目标轮廓，简化内能的同时提高模型抗噪能力；对于左心室内膜的分割，采用基于GVF方向图的二阶段曲线演化策略较好地克服图像噪声、伪影以及乳突肌的影响；在分割外膜时，以内膜分割结果作为初始化，在外膜力场的作用下再次激活曲线运动收敛于外膜。实验结果表明，该方法对初始轮廓位置鲁棒，能够同时分割左心室内膜和外膜。 (4)提出了一种战于非均匀有理B样条(NonUniform Rational B-Spline，NURBS)体模型与三维调相技术(Harmonic Phase，HARP)的左心空运动分析方法。该方法融合MR短轴图像和长轴图像中网格标记模式提供的运动信息，将HARP方法从二维柘展到三维；采用NURBS体模型拟合左心室的复杂拓扑结构，该模型在三维HARP提供的运动信息作用下发生形变，并对相对稀疏位移场进行差值；将各时帧估计得到的NURBS体模型在时间上进行平滑，重建左心室四维连续时变运动模型。实验结来衣明，该方法能够快速有效地估计左心室的三维运动。 关键词：心脏核磁共振图像；医学图像分割、左心室运动分析；主动轮廓模型；梯度矢量流

 Cardiac magnetic resonance imaging (MRI), as a noninvasive technique, has been an essential supplementary means in the clinical diagnosis of heart disease. Cardiac MR images could provide the anatomicial configuration to analyze motion and deformation of heart and assess global and regional myocardial function for effective diagnosis and treatment of heart disease. Left ventricle (LV) is the pump of the blood circulation of the whole body and has receiced a significant attention in medical image processing due to its motion could reflect the appearance and reason of heart disease. This dissertation focuses on the localization, segmentation, 3D motion reconstruction and strain analysis of the left ventricle based on the theories and methodologies of image processing, computer vision and pattern recognition. The main contributions include: (1)An automatic left ventricle localization using the intensity difference image computed between two consecutive frames in temporal image sequences is proposed. Cardiac MR images have excellent spatial resolution and superb soft tissue contrast. Intensity difference image between two consecutive frames could preserve the position and the shape of LV which has large motion displacement. According to the circle-shape characteristic, left ventricle centroid, radius and region of interest (ROI) could be located using temporal intensity difference along with Hough transform. The radius near the endocardium could be used to initialize the snake. This method reduces the object extent largely and lays the foundation of segmentation from cardiac MR images. (2)Left ventricular endocardium segmentation from cardiac MR images based on radial snake model is presented. Different from the traditional snake moving through 2D image plane to minimize the energy functional, the model evolves to the object boundary along 1D radial direction in polar transformation image, simplifying the energy from 2D to 1D and improving the computational efficiency. Shape constraint and homogeneous region constraint are adopted to enhance the robustness of the model on account of the shape characteristic of LV in the transform image. Experiment results demonstrate the effectiveness of the proposed method for endocardium extraction. (3)Improved gradient vector flow (GVF) based left ventricle segmentation from cardiac MR images using radial B-Snake model is described. Introducing gradient directional information into the radial snake model, external force field of endocardium and epicardium computed by their edge maps will attract the contour to endocardial and epicardial boundaries successively in a single snake evolution. Concise B-spline representation of contour replaces the internal energy and enhances the denoising ability of model. During endocardium extraction, GVF directional map based two-phase curve evolution strategy is adopted to improve the ability of overwhelming noise, artifacts and papillary muscles. Then taking the resultant endocardium as an initialization, epicardium external force reactivates the snake forward to epicardial contour again. Experiment results demonstrate the robustness to the initialization and the effectiveness of segmenting the endocardium and epicardium simultaneously. (4)Cardiac motion estimation from tagged MRI using NURBS volumetric model and 3D-ⅡARP is proposed. HARP is extended from 2D to 3D mathematically, making full use of information afforded by SA and LA images each with a grid tagging pattern to obtain 3D motion displacements. The NURBS model represents anatomy of IV compactly and displacement information that 3D-HARP provides drives the model to deform. Also the model interpolates the known sparse displacements to get a 3D dense motion field reflecting the natural continuity and smoothness of the three-dimensional tissue deformations. After estimating the motion at each phase, we smooth the NURBS models temporally to achieve a 4D continuous time-varying representation of LV motion. Experimental results on in vivo data show that the proposed strategy could estimate 3D motion of LV rapidly and effectively benefiting from both HARP and NURBS model. Keywords: Cardiac Magnetic Resonance Images; Medical Image Segmentation; Left Ventricle Motion Analysis; Snake Model; Gradient Vector Flow