【作者】 尤建洁；

【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， 2007， 博士

【摘要】 目前，心脏核磁共振影像技术已经成为心脏疾病临床诊断的重要辅助手段。心脏核磁共振影像分析是无创性评价心脏功能的重要方法，运用心脏核磁共振图像不仅能观察到心脏的形态结构，还可以估测心室的整体功能及局部心肌功能，使医师能对心脏的病理和生理状况做出正确的判断。在临床心脏疾病诊断中，由于左心室是全身血液循环的泵体，在心脏功能中起着重要的作用，因此是目前研究的重点。从心脏核磁共振图像序列提取左心室在收缩期的心肌运动场、位移场、应变应力，进而评定左心室整体和局部心肌功能，在临床具有重要的应用价值。本文着重对基于心脏核磁共振图像的左心室分割及心肌运动进行深入研究与探讨。从左心室的二维分割、三维表面恢复、三维运动重建和应力应变分析等方面着手，提出了相应的二维分割及三维运动分析模型，构建了运用心脏核磁共振图像序列进行左心室三维心肌功能评定的一般框架。本文的工作主要包括以下内容：(1)提出了基于模拟退火算法的简化Snake图像分割模型。该模型对传统Snake模型进行了改进，运用简化Snake的思想，使运算简单，并增加系数可变的面积项，使演化曲线不再受初始位置的限制。在求解模型时，引入了模拟退火优化算法，该算法与改进的Snake模型相结合，发挥了模拟退火算法的长处，也降低了运用Snake计算的复杂度。本文算法不仅能较好地处理弱边缘和不规则边缘区域，而且能较好地处理凹陷和拐角区域。(2)针对标记线对心脏核磁共振图像进行分割时所产生的干扰，本文提出了一种基于细节信号能量的纹理分析算法，该算法利用标记线本身的信息特征，去除了标记线对分割的影响。由于心脏核磁共振图像边缘较弱，噪声相对较强，本文应用测地线活动轮廓模型定义演化曲线，并用水平集方法求解，能够很好地提取左心室的心内膜。在提取左心室心外膜时，为处理边界断裂、缺省的现象，添加了距离约束能量项，解决了依靠梯度和纹理信息无法准确分割的问题。(3)本文利用带标记线的心脏核磁共振的时序图像，针对左心室建立了具有系数函数的物理可形变模型，对左心室在心脏收缩期的形状和三维运动进行了重建和分析。在对初始时刻左心室表面恢复的过程中，本文提出了反演系数函数成形法；在运动跟踪阶段，本文提出了一种有效的模型节点标记力计算方法，并利用薄板样条对其进行了优化。本文模型不仅能跟踪左心室质点级的运动轨迹，并能最终将左心室在心脏收缩期的运动和形变映射为系数函数的变化曲线，直观易懂，有助于临床应用。(4)为直观地分析左心室在心脏收缩期的应力分布及形变情况，本文提出了一种左心室力学形态分析方法。首先，利用带标记线的心脏核磁共振图像数据，针对左心室建立系数可变的物理可形变模型；在所建模型的基础上，利用心脏收缩期各个时刻的左心室轮廓点数据进行三维重建。其次，利用心脏收缩期各相邻时刻的标记点数据计算左心室模型标记力。最后，将模型标记力转换为收缩应力分量、切向应力分量和拉伸应力分量，并将各个应力分量用彩色云图显示。本文提出的左心室力学形态分析方法，能直观有效地反映左心室内外表面在整个心脏收缩期的应力分布及形变趋势。(5)利用时间序列的带标记线心脏核磁共振图像，提出了一种基于有限元方法的左心室运动分析模型，重建了左心室在心脏收缩期的三维位移场及应变分布。本文首先提出了一种基于Delaunay三角网的三维表面恢复算法，该算法能对表达形式多变、空间分布不一致的稀疏点进行表面恢复。然后建立左心室有限元模型，重建三维运动，获取左心室形变信息。本文提出的左心室表面恢复算法及基于有限元方法的左心室模型，能有效地获取左心室的形状信息和运动信息，并为临床诊断提供有用的参考数据。(6)提出了基于有限元方法的左心室生物力学模型。心肌是生物物质高度优化的复合材料，心肌纤维的方向是连续变化的，本文在基于有限元方法的左心室模型的基础上，进一步考虑到心肌纤维方向，将心肌的物质材料属性添加到模型中，建立了左心室生物力学模型。在该模型基础上，本文分析了左心室在局部纤维坐标系下的应力应变分量，给出了各个分量的彩色云图，这对临床应用具有重要的意义。更多还原

【Abstract】 Currently, the technique of the cardiac Magnetic Resonance (MR) imaging has becomean important assistant measure in the clinical diagnosis of heart diseases. The analysis ofthe cardiac MR imaging is an important approach to measure the heart functionnon-invasively. Through the cardiac MR images, the physicians not only can observe thestructure of the heart, but also can estimate the global function and local myocardiumfunction of the ventricles. So the physicians can make the right estimation of the pathologyand physiology of the heart.In the clinical diagnosis of heart diseases, as the Left Ventricle (LV) is the pump of theblood circulation of the whole body, it leads an important role in the heart function. So theLV is the focus in the current research. From the sequence of the cardiac MR images, themotion, displacements and strain-stress of the myocardium of the LV during systole will beextracted. Then the global function and local myocardium function of the LV can bemeasured, which is significant in the clinical diagnosis. This thesis based on the cardiacMR images, will focus on the segmentation of the LV and the myocardium motion analysis.From the aspects of the 2D segmentation of the LV, 3D surface restoration, 3D motionreconstruction and strain-stress analysis, we have proposed the corresponding 2Dsegmentation models and 3D motion analysis models, building a general framework ofevaluating the 3D myocardium function of the LV from the cardiac MR images. Our workmainly includes the following parts:(1) A simulated annealing (SA) algorithm based simplified Snake model for imagesegmentation is proposed. This proposed model improves the traditional Snake model,introducing the idea of simplified Snake to make the computation easy. Also an area energyterm with variable coefficients is added to make the evolving curve not influenced by theinitial position. The SA optimization algorithm is used to solve the improved Snake model.This idea exerts the characteristics of the SA algorithm, also keeps the low computationcomplexity by applying Snake. Our model not only can deal with the weak edges andirregular edges, but also can deal with concave region and corner region.(2) Aiming to the disturbance from the tagged lines when segmenting the tagged cardiacMR images, a texture analysis method is proposed based on the detail signal energy whichapplies the signal feature provided by the tagged lines themselves. Therefore it caneffectively remove the influence from the tagged lines. In the cardiac MR images, theedges are weak and noises are a little stronger. So the geodesic active contour model isapplied to define the evolving curve, and it is solved by the level set method, thereby theinner edges of the LV are well extracted. When extracting the outer edges of the LV, adistance constraint energy term is added in order to solve the problem of the edges’ breakor default. Depending on this energy term, the feature edges which cannot be directlyextracted by gradient or texture information can be well extracted.(3) By applying the tagged cardiac MR images, we build a physics-based deformablemodel with parameter functions to reconstruct the shape and 3D motion of the LV duringsystole. In the process of the surface restoration of the LV of the initial frame, we propose aretro-deducting parameter functions method to restore the mathematical surface model of the LV. And during the process of the motion tracking, we propose an effective method tocalculate the tagged forces of the model. Meanwhile these forces are optimized usingThin-Plate splines. The proposed model not only can track any material point of the LV,and also almost all the motions and deformations of the LV during systole can be reflectedon the changing curves of parameter functions. These curves are very intuitive and easy tobe understood, which will be very helpful for clinical applications.(4) In order to intuitively analyze the deformations of the LV during systole, amechanical modality analysis method of the LV is proposed. First, applying the taggedcardiac MR images data, we build a physics-based deformable model with parameterfunctions of the LV, and then based on the model built we apply the boundary data sets ofeach frame during systole to reconstruct the 3D shape of the LV. Second, we compute thetagged forces of the LV model during systole by using the tagged data sets of the adjacenttwo frames. Lastly, the tagged forces of the model will be transformed to the contractivestress component, the tangent stress component and the vertical stress component.Moreover, these three stress components will be described in color cloud pictures. So, themechanical modality analysis method of the LV proposed in this thesis will intuitively andeffectively reflect the stress distribution and deformation trends of the inner and outersurfaces of the LV during systole.(5) Based on the sequence of the tagged cardiac MR images, a Finite Element Method(FEM) based motion analysis model of the LV is proposed, reconstructing the 3Ddisplacements and strain distribution of the LV during systole. First, we proposed a kind of3D surface restoration algorithm based on Delaunay triangulation. This algorithm canrestore the surface from the sparse data points of un-uniform spatial distribution ordifferent formats. Then the FEM-based LV model is constructed, reconstructing the 3Dmotion and capturing the deformation information of the LV. The proposed surfacerestoration algorithm and the FEM-based LV model can effectively capture the shape andmotion information of the LV, which provide the useful reference data in the clinicaldiagnosis.(6) A FEM-based biomechanical model of the LV is proposed. The myocardium is thehighly optimized composite material in the biologic material. The myocardium fiberorientations of the LV vary in a continuous manner. Based on the FEM-based model of theLV, the myocardium fiber orientation is further considered, namely the material property ofthe myocardium is added to the model. Thus the biomechanical model of the LV is built.Based on this model, we analyze the strain-stress of the LV under local fiber coordinates.Also each component is shown in the color cloud picture, which is meaningful for theclinical diagnosis.更多还原