【作者】 郑罡；

【导师】 王惠南；

【作者基本信息】 南京航空航天大学 ， 精密仪器及机械， 2007， 博士

【摘要】 人脑MRI结构像(anatomical MRI, aMRI)能够以较高的空间分辩力反映脑部组织的解剖结构。在放射学与神经影像学研究领域,人脑aMRI数据分析的关键问题是如何精确地检测与测量各部分脑组织,可以归结为医学影像的多目标分割问题。该问题的解决对于计算机辅助诊断、组织容积定量分析、异常组织定位、解剖结构分析、外科手术规划与手术导航等具有重要的学术意义和临床应用价值。本文主要内容是:基于活动轮廓模型,提出了五种具有创新意义的塔式结构的多层次分割方法,解决了人脑二维aMRI、多信道融合的矢量量化aMRI以及三维aMRI数据的多目标分割问题。第一,针对二维aMRI数据,给出了构成塔式结构多目标分割的关键技术——背景填充技术,将该技术与活动轮廓模型中的Chan-Vese(C-V)模型相结合,提出了塔式多相水平集分割算法(简称:塔式多相C-V模型),解决了C-V模型在多目标分割以及复杂连接情况表示上的局限,适合分割目标中含有子目标的多目标图像。第二,为了减少塔式多相C-V模型得到的边缘与手工分割真值间的误差问题,提高aMRI分割的准确度,据可调填充色提出了背景填充技术的广义形式,并将该技术与C-V模型相结合,进一步提出了基于广义背景填充技术的塔式多相C-V模型,实现子目标边缘在一定范围内可调,即在少量人工干预实现人机交互式aMRI的多目标分割。第三,为解决aMRI中一些组织因驰豫时间较为接近而难以分离的问题,针对矢量量化图像提出了矢量量化塔式多相C-V模型。能够利用MRI不同的扫描序列以及不同影像仪器对于组织的不同敏感性,实现aMRI中驰豫时间较为接近的组织的有效分离。第四,利用水平集函数能够隐含表示轮廓曲面的特点,提出了基于体素的三维塔式多相C-V模型,实现三维体数据的多层次分割。第五,基于并行多相C-V模型和塔式多相C-V模型,提出了具有复杂分割树形结构的塔式并行多相C-V模型,能够结合两种多相分割模型的优点,实现复杂塔式多层次分割。实验结果表明,本文算法可以实现二维aMRI、矢量量化结构像以及三维aMRI的较精确的多目标分割,并且可以通过调节填充色实现少量人工干预下的交互式分割来进一步减小客观分割结果与主观分割结果间的误差,适合目标中含有子目标的医学影像的多目标分割要求。更多还原

【Abstract】 The anatomical MRI (aMRI) is able to represent the structure of brain with high accuracy and has become the main medical image for the analysis of human brain. In radiology and neurology, a key problem of aMRI of the brain is how to detect and to measure tissues to represent the anatomy structures of the brain which can be seen as a basic problem of multi-object segmentation for medical images. It is most important for computer-aided diagnosis, quantitative measurement of tissue volumes, localizing foci, anatomy analysis, surgical planning and for surgical navigation. In this thesis, five pyramidal multi-layer frameworks for two-dimensional (2-D) aMRI, vector-valued aMRI and for three-dimensional (3-D) aMRI segmentation were proposed based on the active contour model to deal with the multi-object segmentation of the aMRI.Firstly, a pyramidal multiphase level set framework (or named as pyramidal multiphase C-V model) was developed for multi-object segmentation of 2-D anatomical MRI by combining a key technique, named as the technique of painting background with the Chan-Vese model which was able to deal with the limitation of the C-V model in multi-object segmentation and in multiple junction representation and was especially suitable for images with multiple sub-objects.Secondly, to reduce the difference between the results obtain by the proposed pyramidal model and manual segmentation results, a generalized technique of painting background was proposed by using a variable painting color. An improved pyramidal model using the generalized technique of painting background and the C-V model was proposed which was able to adjust the boundaries of sub-objects and can be used for interactive segmentation of the aMRI.Thirdly, to partition the tissues with close relaxation time, a vector-valued pyramidal multiphase C-V model using the different sensitivities of different kinds of medical images or different MR images obtained by different serials was developed.Fourthly, a voxel-based pyramidal model was developed for 3-D aMRI segmentation by extending the level set function to three dimensional.Fifthly, by combining the simultaneous multiphase C-V model and the pyramidal model, a pyramidal simultaneous multiphase C-V model was proposed which was able to partition a given image according to a complex segmentation tree.Experimental results show that the proposed algorithms are able to detect multi-objects with high accuracy in 2-D aMRI, vector-valued aMRI and 3-D aMRI. Also, they can reduce the difference between the obtained segmentation results and the mammul segmentation results by using interactive segmentation and are suitable for multi-object segmentation of medical images whose object regions contain sub-objects.更多还原