节点文献

基于4D-CT数据的心脏重构方法研究

Research of Heart Reconstruction Approach Based on 4D-CT Datasets

【作者】 王兴家

【导师】 冯焕清;

【作者基本信息】 中国科学技术大学 , 生物医学工程, 2011, 博士

【摘要】 心血管疾病是全世界发病率和死亡率最高的疾病,而且发病急、具有很强的隐蔽性,因此实现这类疾病的初期诊断有很重要的临床意义。十来年前问世的多层螺旋CT(MSCT)通过多排探测器技术显著提高了成像速度和扫描精度,已被广泛应用于心脏检查,诊断和评价冠心病;尤其是伴随着MSCT出现的心电同步扫描技术,能提供含有时间信息的四维心脏CT成像数据,这为观察心脏的动态特性提供了一种有效的手段。这些新技术的诞生,推动着医学图像处理和分析领域的研究步伐,摆在我们面前的任务是如何从这些海量图像数据中提取出心脏的病理信息和功能状态,以极大发挥这些高端成像设备在心血管疾病预防和诊断中的作用。因此,基于MSCT数据集的心脏分割与功能分析成为了当前国际上医学图像处理领域中一个新的研究热点。本论文根据从64层MSCT采集的3D心脏数据集和从256层MSCT采集的4D(3D+t)心脏数据集,研究了无训练数据集的自动和半自动心脏分割算法以及基于训练数据集的多心相分割算法,建立了心脏左心室的四维分割模型,并在此基础上对心脏的全局静态参数和动态参数进行了初步分析。本文主要的研究内容及其创新点包括:1.针对MSCT心脏数据集特点,通过分析MSCT心脏数据集分割的难点,完成了基于多心相数据集心脏分割的整体框架设计,主要包括对数据集的轴向翻转、灰度映射、噪声去除等预处理操作,根据层片的结构连续性设计的左心室自动定位算法以及左心室的分割算法等,为基于模型的参数提取和功能分析打下了坚实基础。2.在对基于边缘和区域水平集模型进行深入分析的基础上,针对水平集模型对初始位置敏感及边缘泄露的问题,本文结合心脏解剖结构和MSCT层片特点,提出了不需要训练数据集的无监督改进耦合水平集(ICLS)全自动分割模型。在用左心室自动定位算法准确确定水平集曲线的初始轮廓位置的基础上,通过对边缘检测函数的改进以及结合左心室形状对水平集函数进行耦合,形成了完整的ICLS心脏分割模型。该模型较好地抑制了曲线的演化泄露,避免了局部极值对曲线演化的影响,可以获得较为理想的内、外心膜边缘。3.对外心膜边缘模糊甚至缺失的心脏层片,ICLS模型严重依赖演化的结束条件和水平集参数设置,为此本文提出了人工交互的IFLW(Improved Fuzzy connectedness-based Live-Wire)半自动分割模型,通过可以自动选取种子点的融合骨架信息的模糊连接度函数,来表示心膜边缘模糊、缺失层片的区域模糊特性,并在极坐标下将模糊连接度嵌入联合矩阵代价函数,利用改进Livewire的IFLW模型实现了心脏数据的半自动分割。与传统Livewire算法相比,IFLW模型只需要较少特征点就可以接近甚至达到手工分割的精度,在降低人工干预的同时增加了算法的鲁棒性,特别是对模糊边缘层片分割的准确性。4.为减少人工干预和充分利用IFLW模型精确分割的优点,本文提出了基于IFLW分割训练样本集的改进形状统计模型分割算法。通过IFLW对样本层片的分割结果构建数据集的形状模型和局部灰度模型以及对定位算法和模型搜索算法的改进,实现了基于改进多阶形状统计模型的三维MSCT数据集分割。此外在4D多心相MSCT数据集中,通过引入基于信息量的Demons非刚性配准方法,获得从已分割三维数据集层片向待分割心相三维数据集对应层片的形变场,然后通过对分割结果进行形变获得对应层片的分割结果,从而实现了MSCT四维数据集的分割。5.在上述工作的基础上,初步进行了心脏静态和动态参数提取和功能分析的研究。最后,作者特别感谢国家自然科学基金(60771007)和安徽省自然科学基金(2006KJ097A)对本文工作的资助。

【Abstract】 Cardiovascular diseases (CVDs) characterized by acute onset and great hidden nature, with the highest incidence and mortality rate, are the number one cause of death all over the world. The early quantitative diagnosis and accurate evaluation of CVDS have an important clinical significance. The occurence of Multi-Spiral CT (MSCT) detector based on multiple-detector technique, has significantly promoted imaging speed and scanning accuracy, which has been widely used in cardiac examination and diagnosis, the evaluation of coronary heart disease. Associated The Appearance of the ECG-synchronization scanning technique, the 4D cardiac CT imaging datasets included the temporal Information would be obtained from MSCT, which provide an effective means of clinical observation for dynamic characteristics of heart.As these new technology have been promoting the development of the medical image processing and analysis, the crucial task ahead of us is how to extract the heart morphology, function and cardiac Pathological Information from the massive imaging data, in order to make full use of the high-end imaging equipment in the field of CVDs early quantitative prevention and diagnosis . Hence the heart extraction and function analysis based on MSCT have become a hot issue in the medical image processing area in recent years.This paper based on 3D cardiac datasets from 64-slices MSCT and 4D cardiac datasets (3D+t) from 256-slices MSCT, have explored the automatic and semi-automatic cardiac segmentation without training sample set, the cardiac multi -phases segmentation with training sample set, and buildup of the 4D segmentation model of left ventricle(LV). Based on the segmentation model, this paper analyzes the cardiac global static parameters and dynamic parameters preliminarily. The main research work and contributions of this dissertation can be summarized as follows:1. According to the characteristic of cardiac MSCT datasets, with the analysis of difficulties of heart segmentation for MSCT imaging, the design of the integral segmentation framework based on multi-phase cardiac datasets have been accomplished, which mainly include pre-processing operation such as cardiac axis’s transformation for the datasets, intensity transformation, image denosing and so on, and automatic localization algorithm, automatic and semi-automatic LV segmentation algorithm, etc. the construction of segmentation models will lay a solid foundation for the further heart parameter extraction and function analysis .2. With the deep analysis of the edge- and region-based levelset model, according to the drawbacks of edge leakage and sensitivity to initial position, we proposed unsupervised improved coupled levelset (ICLS) automatic LV segmentation algorithm without training datasets,by combination of cardiac anatomy and MSCT slice characteristic. On the basis of the automatic location algorithm to determine the initial contour position of the level set curve accurately, through the improvement of edge detection function and coupled level set function with LV shape, a complete heart segmentation ICLS model is built. The model could inhibit edge leakage, and avoid the local minimum during the curve evolution, and by this model an ideal epicardium and endocardium could be obtained automatically.3. With regard to the LV segmentation of cardiac slices with edge blur and lack, ICLS model relies heavily on the termination conditions of the evolution and parameters of the level set model. This paper proposes a manual-interaction IFLW (Improved Fuzzy connectedness-based Live-Wire) semi-automatic LV segmentation model. By generating the seed points automatically, a new Fuzzy connectedness function incorporating myocardium position information is introduced to represent the region fuzzy features of the slices with edge blur and loss. The fuzzy connectedness of myocardium is then embedded into the joint cost matrix in polar coordinates, and with the improved Livewire algorithm (IFLW), the heart extraction from datasets is achieved semi-automatically. Compared with traditional Livewire algorithm, IFLW model requires much fewer feature points and the segmentation results are close to the accuracy of manual segmentation. IFLW model reduces extra manual intervention while increasing the robustness and accuracy of the LV segmentation algorithm.4. To reduce the manual intervention level and make full use of the advantages of IFLW accurate segmentation model, an improved shape statistical models segmentation algorithm based on IFLW model to segment the sample sets is proposed. On the basis of the segmentation results of the chosen slices by IFLW, The shape model and local intensity model are constructed. Through the new localization algorithm of the initial shape model and improved shape model search algorithm, the 3D segmentation of MSCT datasets based on improved multi-scale shape statistical model is proposed. In addition, by the Demons registration algorithm based on mutual Information in the cardiac multi-phase 4D MSCT datasets, the deformation field from the slice of the segmented 3D datasets to the corresponding slice of the 3D dataset to be segmented could be obtained, which is used to deform the segmented results of the segmented 3D datasets, to achieve the segmentation result of the 3D dataset. By the means of the non-rigid registration and shape statistical models, the whole MSCT 4D data set segmentation results are obtained.5. On the basis of the above work, a preliminary function analysis for cardiac static and dynamic parameters was carried out for the further research.Finally, special thanks to the State of the National Natural Science Foundation of China (60771007) and Anhui Natural Science Foundation of China (2006KJ097A) for funding this work.

节点文献中: