节点文献

脑磁共振图像的白质结构提取—分割算法及其评价

Brain Magnetic Resonance Images Segmentation and Its Evaluation

【作者】 周振宇

【导师】 韦钰; 阮宗才;

【作者基本信息】 东南大学 , 生物医学工程, 2006, 硕士

【摘要】 磁共振(MR)成像技术是一种有效的研究人脑的非侵害性途径。由于其自身的优势,MRI被越来越多的应用于医学、神经科学、心理学、认知科学等方面的研究。我们对临床采集的孤独障碍患儿的核磁共振大脑结构扫描数据进行了图像分割,有效提取了大脑的白质结构,为研究白质组织在该病患儿发育中的变化奠定了重要基础。本文主要针对大脑白质结构提取的问题,对医学图像分割方法从理论研究的角度,简单阐述各类分割算法,重点研究了基于分水岭算法和小波变换的分割算法,并将改进的基于小波变换的多分辨率阈值分割算法应用于临床数据,取得了较为满意的结果。我们提出了一种新的基于分水岭算法和小波变换的分割算法。基于分水岭算法,我们有效地去除MR图像中的脑壳部分,利用区域合并技术克服分水岭算法固有的过分割问题。我们利用形态学算法对图像求梯度作为分水岭算法的输入图像,在分水岭算法之后,根据灰度相似性或者通过参数的选择对分割的小区域进行合并,最终将图像中脑组织有效提取出来。实验证明,这种方法具有较理想的结果,并具有一定的鲁棒性。然后,在将二维离散小波变换应用于图像分解的基础上,分析了二维小波变换在大脑MR图像分割中的应用,并提出了一种新的基于小波变换的分割算法。该方法首先利用一个滑动窗口取出局部图像,对局部图像进行小波多尺度变换,从较大的尺度系数到较小的尺度系数逐步定位出局部图像的初步灰度阈值;滑动窗口以一定的规则遍历整幅图像,重复上述过程;最后利用最大隶属原则判断象素的灰质、白质和脑脊液的划分,提取白质结构。实验结果表明,该方法有较好的分割结果。我们对该算法应用于临床数据的分割结果与前人的算法分割结果进行了比较,说明了我们提出的算法更适合于临床采集的孤独障碍患儿大脑MR图像数据。作为对比,我们对功能磁共振图像分析软件SPM2中的灰质、白质分割算法进行了研究,在无法获得先验参数的条件下实现了该算法,但分割效果不明显;使用C语言,借助图像分割与配准工具包ITK实现阈值水平集算法。我们将改进的基于小波变换的多分辨率阈值分割算法和SPM2分割结果、水平集方法分割结果进行了对比,说明我们的新算法不仅分割效果好,而且比其它两种算法更适合于弥散张量成像数据。这种算法可以很好地应用在临床采集的孤独障碍儿童大脑MR图像白质结构提取和分析中。对于这些方法的分割结果,我们基于可视化工具包VTK采用传统的等值面抽取算法实现了白质结构的三维可视化,更加直观观察大脑白质结构提取的效果。在程序的编写开发过程中,系统中的部分成熟算法使用了医学图像处理领域比较先进流行的开发工具包:图像分割和配准开发包ITK和可视化开发包VTK。提高了代码的执行效率,可读性。标准开放库的使用也使得今后对于系统功能的扩充成为可能。

【Abstract】 Magnetic resonance imaging is an effective and noninvasive approach to observe human brain. Now Magnetic resonance imaging is widely applied in many fields, such as medicine, neuroscience, psychology and cognitive science because of its advantages. This paper focused on the brain magnetic resonance images, which is one of the key problems in medical image processing.A novel segmentation method based on watershed transform and wavelets transform is presented for white matter in thin sliced single-channel brain magnetic resonance scans. The original image is smoothed by using anisotropic filter and then the morphologic grad image is computed, which is the input image over-segmented by the watershed algorithm. After the segmentation, the small regions are incorporated into their neighbor regions according to the comparability of the two regions. The experiment shows that the method can lead a perfect result, and has some robustness. Finally, the brain MR image is segmented automatically by using the multicontext wavelets-based thresholding method. In this method, the wavelet multiscale transform of local image gray histogram is done and the gray threshold is gradually found out from large scale coefficients to small scale coefficients. Image segmentation is independently performed in each local image to calculate the degree of membership of a pixel to each tissue class. A strategy is adopted to integrate the intersected outcomes from different local images. The result of the experiment indicates that the algorithm can obtain segmentation result fast and accurately.Some traditional methods of image segmentation are used and compared with the improved segmentation method based on watershed transform and wavelets transform. They are individually based on probabilistic modeling of intensity distributions which is the basic idea of Statistical Parametric Mapping 2 (SPM2) and Level Set filter of Insight Segmentation and Registration Toolkit (ITK). The result of the experiment indicates that our method not only outperforms other traditional segmentation methods in classifying brain MR images but also more suitable to diffusion tensor imaging data. As to the Visualization, We use a mature method named Marching Cube and it provides comparatively satisfied result.Developing the whole project, we use some popular toolkits in the medical imaging field: Insight Segmentation and Registration Toolkit (ITK) and Visualization Toolkit (VTK). Code is highly organized and is extendable.

  • 【网络出版投稿人】 东南大学
  • 【网络出版年期】2007年 04期
  • 【分类号】R318
  • 【被引频次】2
  • 【下载频次】391
节点文献中: 

本文链接的文献网络图示:

本文的引文网络