【作者】 尚岩峰；

【导师】 杨新；

【作者基本信息】 上海交通大学 ， 模式识别与智能系统， 2010， 博士

【摘要】 医学图像分割是辅助诊断、量化分析、手术规划导航等较高层次临床应用的基础。但由于疾病的多样性,造成医学图像背景分布极其复杂、医学图像中病患器官目标自身形状变化有相当的随机性、以及成像过程中各种噪声干扰和超大的数据量,均增加了传统方法对医学图像中目标分割的难度。若要在这些复杂条件下准确、自动、有效地提取特定医学目标,要着重以下两方面工作。第一,选择合适的分割模型。主动轮廓模型作为一种有效的分割模型,在医学图像中,有了很多成功的应用,这也正是本文研究的目标。第二,充分利用特定目标的先验知识。根据待提取目标的不同特征,设计最有效的分割模型。本文正是在主动轮廓框架下,针对特定的医学目标,从一般医学器官,到血管,到肝脏各组织,到心脏各组织等医学目标,设计出多种不同的主动轮廓模型,包括区域竞争主动轮廓、血管主动轮廓、多主动轮廓和基于先验形状的主动轮廓等模型,对这些医学目标进行有效提取,并在准确分割的基础上,对部分组织器官进行自动量化分析。本文首先对医学图像分割的研究背景进行了回顾,在基于先验知识目标分割的一般框架下,按原始图像、分割指标函数和先验知识三要素对基于先验知识图像分割的最新进展进行了综述,侧重于先验知识描述和其与分割准则函数的结合。对于本文的研究重点目标主动轮廓模型,围绕Snake模型和水平集模型以及二者本质上的统一性,结合一些利用先验知识对特定目标提取的具体实例进行详细介绍。通过对基于先验知识目标分割的有关问题进行的讨论分析,认为最优的主动轮廓模型,是充分利用目标的先验知识,由这些先验知识指导主动轮廓演化从而提取特定医学目标的模型,并对医学图像和主动轮廓的发展方向进行了展望。本文介绍了一种通用性的分割模型:基于区域竞争的主动轮廓模型。这一基于目标灰度统计概率和水平集的主动轮廓分割模型,把能量函数表示为在目标区域内对象素点属于目标的概率的积分,并在水平集框架下对能量函数最小化,得到分割的迭代方程。同时,通过附加的速度约束项,使得主动轮廓越过目标边缘的时候降低速度,大大提高分割的收敛性和准确度。通过大量冠状动脉和二尖瓣分割试验,表明该模型的有效性。针对血管树结构的复杂性,研究了血管的提取模型:血管主动轮廓模型。这一模型,充分利用一切和血管有关的信息,先验灰度分布、区域信息、多尺度血管矢量场和曲率,能量方程最小化,得到包括三个主要速度项的迭代方程:基于区域竞争和先验灰度的主动轮廓、血管矢量场、多曲率策略。该模型可以自动地对整个血管树进行提取,不需要太多的预处理和后处理,是一种快速、准确、健壮和自动的血管提取模型。肝脏作为人体最重要的器官之一,若对CT图像中的肝脏自动检测、分类和分割出肝脏的解剖和病理结构,包括肿瘤、肝脏和血管,得到具体病例的肝脏模型,在临床上是意义重大的。本文通过多种主动轮廓模型来对CT图像肝脏有关的组织结构进行提取,实现肝脏的模型化。其中,本文采用基于混合高斯分布的主动轮廓模型用于肝脏和肿瘤的提取。该模型的主动轮廓在灰度图像的混合高斯分布模型的驱动下,可以灵活机动地从复杂背景下提取出多重目标。对于肝脏内部的血管树状结构,本文采用增强的血管主动轮廓模型。最后,针对超声心动图中的瓣膜类疾病,本文采用一种把先验区域和形状知识融入到几何主动轮廓模型的目标分割方法。各个层次的先验知识表示成速度场直接指导水平集演化到理想轮廓。先验区域约束水平集的演化范围,先验形状驱使水平集向特定形状收敛。在三维超声图像的应用表明该方法大大减少了手工干预,提高了分割效率和精度。同时,在二尖瓣自动分割的基础上,对二尖瓣有关的参数进行了量化,包括瓣膜开口面积、瓣环面积、左心室容积和瓣叶张开角度。更多还原

【Abstract】 Medical image segmentation is a precondition of aided diagnosis, quantitative analysis and surgical planning in clinical application. Due to the wide variety of shapes, the complexity of the topology, the presence of noise in a complicated background, and the diversity of imaging techniques, it is a very difficult task to extract medical object automatically and accurately. In order to overcome these difficulties, there are two topics could be done. One is to choose an optimized model. As an efficient segmentation tool in medical image, active contour model has made a great success in medical application. The other is to make full use of object’s prior knowledge and embed them into an active contour model. In this paper, several specific active contour models are proposed according to specific clinical application, which are Region Competition Based Active Contour Model, Vascular Active Contour Model, Multi Active Contour Models for liver modeling, and Region and Shape prior base Geodesic Active Contour Models. All of them are results of combination of certain prior knowledge and active contour. On the base of these segmentation models, some quantitative analysis is made on certain organs and diseases.Firstly, we review the recent advances in prior-based image segmentation and analyze them according to a general framework which consists of three parts: original image features, segmentation models and prior knowledge. The presentation of prior knowledge and how to embed it in a segmentation model are emphasized. Then, as the most important segmentation model, active contour models are reviewed in detail, which include Snake and level set model. At last, some points, such as characteristic of prior knowledge, challenges of active contour and promising research directions are presented. Secondly, a probabilistic and level set model for three-dimensional medical object extraction is proposed, which is called region competition based active contour. The algorithms are derived by minimizing a region based probabilistic energy function and implemented in a level set framework. An additional speed-controlling term makes the active contour quickly convergent to the actual contour on strong edges, whereas a probabilistic model makes the active contour performing well for weak edges. Prior knowledge about the initial contour and the probabilistic distribution contributes to more efficient extraction. The developed model has been applied to a variety of medical images, from CTA and MRA of the coronary to rotationally scanned and real-time three-dimensional echocardiography images of the mitral valve. As the results show, the algorithm is fast, convergent, adapted to a broad range of medical objects and produces satisfactory results.Thirdly, a novel active contour model is proposed for vessel tree segmentation which makes full use of all available vascular information. Firstly, we introduce a region competition based active contour exploiting prior intensity distribution information to segment thick vessels robustly and accurately. Secondly, we define a vector field, resulting from the eigen analysis of the Hessian matrix of image intensity and used by the active contour to evolve into the thin vessels. The vector field is also specified in a multi-scale framework. Finally, a vascular smoothened term takes a strategy combined with minimal principal curvature and mean curvature, which make it smoothes surface without changing the shape of the vessel tree. The developed model has been applied to liver vessel tree, coronary artery and lung vessel extraction. Some comparisons are made between Geodesic Active Contour, CURVES, C-V and our model. The experiments show that the model is fast, accurate, robust and suited for an automatic procedure in vessel tree extraction.Fourthly, in order to extract the liver, its tumors and vessels, we developed an active contour model with an embedded classifier, based on a Gaussian mixture model fitted to the intensity distribution of the medical image. The difference between the maximum membership of the intensities belonging to the classes of the object and those of the background, is included as an extra speed propagation term in the active contour model. An additional speed controlling term slows down the evolution of the active contour when it approaches an edge, making it quickly convergent to the ideal object. The developed model has been applied to liver segmentation. Some comparisons are made between Geodesic Active Contour, C-V and our model. As the experiments show, our model is accurate, flexible and suited to extract objects surrounded by a complicated background.At last, we present an automated 3D echocardiography image protocol for quantitative analysis of mitral valve. A region and shape prior of the cardiac valve is presented in form of speed field and incorporate it into image segmentation within level set framework. Region prior constrains the zero level set evolving in certain region and shape prior pulls the curve to the ideal contour. On the base of automatic mitral detection method, we quantitate some important geometrical parameters of mitral valve in an automatic way, which include metal valve orifice area, annular area, LV volume, and leaflet opening angle. Some comparisons with Snake method and manual parameterization results show the validity of our methods.更多还原