【作者】 陈刚；

【导师】 顾力栩；

【作者基本信息】 上海交通大学 ， 计算机应用， 2008， 硕士

【摘要】 医学图像分割是医学图像处理与分析的一个重要领域,同时也是计算机辅助诊断与治疗的基础。所谓图像分割就是根据某种均匀性(或一致性)的原则将图像分成若干个有意义的部分,使得每部分都符合某种一致性的要求。图像分割在医学图像处理中具有十分重要的意义,比如三维可视化,计算机辅助外科手术以及放射治疗等医学应用都假设已经对图像进行了精确分割。在本文研究的肝脏灌注自动分析中,第一步就是对肝脏进行精确分割。基于可变形模型的分割算法已经渐渐发展为医学图像分割领域最为活跃和成功的领域之一。可变形模型的基本思想是建立模型的能量函数,在模型内部控制力和外部图像力的共同作用下使曲线或曲面演化,并使该能量函数达到最小化,从而收敛到待分割区域的边缘。基于可变形模型分割算法的一个突出的优点是图像数据、初始形状和目标轮廓统一于一致的数学模型中。其中,基于形状先验的可变形模型通过并入待分割区域的形状先验知识到可变形模型中,大大提高了分割的精度,成为当前可变形分割模型研究的热点。本文主要研究了基于平均形状先验可变形分割模型的原理、数值实现和实验。水平集方法(Level Set Methods)是可变形模型的数学基础,也是一种在医学图像处理上应用非常广泛的分割算法。水平集方法的优点在于它的迭代演化过程不依赖具体参数,演化曲线或曲面可以隐式地表达为高维函数的零水平集,因此可以自动处理感兴趣区域的拓扑结构变化。但是水平集方法的缺点在于经常出现欠分割、过分割和溢出等问题。因此,通过对水平集算法进行改进来解决欠分割、过分割和溢出等问题就成为一项非常重要的课题。为了解决这些问题,许多学者已经做了大量研究。本文主要在多初始化和改进速度函数的定义上来改进水平集分割算法。本文的主要工作和创新点如下:1.介绍常用的医学图像分割算法的研究现状及其优缺点。重点讨论了可变形模型中的主动轮廓模型和基于水平集的分割算法。2.提出多初始化水平集分割算法。针对传统水平集分割算法经常出现的欠分割、过分割和溢出等问题进行研究。本文在改进速度函数的基础上,提出了多初始化水平集分割算法,基本上解决了这些问题。3.提出基于平均形状先验的可变形模型。本文提出基于平均形状先验的可变形模型分割算法,及其数值实现。4.实现腹腔MR(Magnetic Resonance)图像肝脏的分割。本文用多初始化水平集算法从充满噪声并且边缘不明显的腹腔MR图像分割出肝脏,并且和其它分割方法的实验结果进行对比和评价。5.实现肝脏灌注的自动测量。本文用改进的Chamfer Matching算法自动跟踪MR图像序列中每一幅图像的肝脏灌注点,然后计算该灌注点的灌注强度并且自动画出灌注曲线,供医生诊断参考。更多还原

【Abstract】 Medical image segmentation is a fundamental problem in medical image processing and analysis, and is the basis of computer aided diagnosis and treatment as well. The general segmentation problem is the process of partitioning an image or data-set into a number of uniformity or homogeneous segments. Image segmentation is important for medical image analysis, for example, 3D Visualization, Computer Aided Operation, and Radiology Treatment are all assume that Region of Interesting (ROI) are well segmented. Also automate liver perfusion analysis’s first step is liver segmentation. Perfusion analysis is based on the segmentation result.Deformable model segmentation method has generally become one of the most active and successful section in medical image segmentation research. The basis idea of deformable model is to construct an energy function for the model and let the curve or surface evolve under the model’s inner control force and outside image force. When the energy function reaches its minimum, the evolving curve or surface reach its target region. The advantage of the deformable mode is image data, initial contour and target contour are included by one uniform mathematical mode. One of the most popular methods is deformable model with prior shape information. By incorporating the prior shape information, it can improve the accuracy of deformable model. Here, we propose a deformable model with mean shape prior information, and its numerical realization.Level Set Methods is the mathematical foundation of deformable model, and it is very popular in medical image segmentation. Its advantage is that it is independent of detail parameters during the evolving process. The evolving curve or surface can be represented as the zero level set of a higher dimensional function, which can deal with the topological change of region of interesting automate. But Level Set Methods often will produce under-segmentation, over-segmentation and leakage problems. How to solve these problems becomes a big challenge for researchers. To solve these problems, we use multiply initialization for level set and improve the speed function’s definition.The main works are described as follows:1. Overview of the popular medical image segmentation methods, which emphasis on deformable model and Level Set Methods.2. Propose a multiple initialization Level Set Methods. We use the multiple initializations and an improved speed function to solve the under-segmentation, over-segmentation and leakage problems.3. Propose a deformable segmentation model with mean shape prior information. In this thesis, we propose a deformable segmentation model with a mean shape prior and its numerical realization.4. Realize liver segmentation in abdomen MR images. We use the multiple initialization Level Set Methods to segment the liver in the noisy and low contrast abdomen MR images, and the segmentation results are compared with other segmentation methods.5. Realize automate liver perfusion analysis. We use Chamfer Match to trace the perfusion position in each slice of a whole abdomen MR series, and draw the perfusion curve for radiologist.更多还原