基于改进模糊聚类算法的MRI脑部图像分割研究

发布时间：2018-06-15 04:27

本文选题：核磁共振图像 + 偏移场矫正模糊C-均值聚类　；参考：《沈阳建筑大学》2015年硕士论文

【摘要】：随着科技的发展,核磁共振成像(Magnetic Resonance Imaging, MRI)、X射线计算机断层扫描成像、正电子发射断层成像术和单光子发射计算机断层成像术等医学成像技术的使用越来越广泛。医生的时间精力有限,同时存在一定的主观差异性,计算机辅助诊断系统在临床上发挥的作用越来越重要。医学成像技术可以在无创伤的情况下为医护人员提供很多有用的信息。同时由于其可以多次使用,在疾病诊断、病情监控、手术效果反馈等方面起着非常重要的作用。MRI拥有无辐射性伤害、分辨率高和多参数任意角度成像等优点,因而得到了广泛应用。作为三维重建等其它医学图像处理手段的基础,脑MRI图像的精确分割对医学图像处理和医疗技术的发展具有重要意义。然而,在实际应用中,脑MRI图像中存在一定的缺陷如噪声、偏移场、部分容积效应及低对比度等。这导致脑MRI图像的精确分割成为了一个困难的问题。本文针对MRI图像的分割问题进行研究。由于脑MRI图像分割的重要性,国内外研究人员对MRI图像分割的问题提出了很多行之有效的方法。如活动轮廓模型、马尔科夫随机场、Atlas图集、模糊聚类等。其中模糊C均值算法(Fuzzy C-means, FCM)是尤为被国内外学者常用的一种方法。但是标准的FCM算法基于图像的灰度信息,而没有考虑图像的空间信息,在分割高噪声高偏移场的图像上效果不佳。偏移场矫正模糊C-均值算法(Bias-Corrected FCM,BCFCM)在FCM的基础上增加了对偏移场的估计和空间信息的使用,可以很好的解决图像偏移场对分割造成的影响。但是BCFCM算法由于没有考虑到噪声对偏移场估计的影响,所以在分割高噪声图像的时候会对偏移场估计错误,从而对分割结果造成影响。本文针对MRI脑组织分割,在图像预处理过程中提出一种快速的分割方法来去除颅骨及其附属物。此外,针对于BCFCM算法存在的缺陷,主要提出了两方面的改进方案：提出了一种基于BCFCM的改进算法,该改进算法在迭代过程中可以通过对噪声强度的估计来自适应的改变目标函数窗口的大小。通过对比实验表明,该改进方案大大增强了BCFCM算法对噪声的鲁棒性。同时,考虑到BCFCM算法在偏移场上的估计错误,提出了一种算法,通过引入高斯核函数对偏移场进行平滑处理,并通过阈值限制偏移场的估计值,可以有效地避免偏移场的错误估计对分割结果的影响。实验结果表明,改进后算法不仅可以有效准确地对脑组织分割,而且具有较强的抗噪声和处理偏移场的能力。
[Abstract]:With the development of science and technology, magnetic resonance imaging, X-ray computed tomography, positron emission tomography and single-photon emission tomography are more and more widely used in medical imaging technology. The time and energy of doctors are limited and there are some subjective differences at the same time. Computer aided diagnosis system plays a more and more important role in clinical practice. Medical imaging technology can provide a lot of useful information for health care workers without trauma. At the same time, because it can be used many times, it plays a very important role in disease diagnosis, disease monitoring, surgical effect feedback and so on. MRI has the advantages of no radiation injury, high resolution and multi-parameter arbitrary angle imaging. As a result, it has been widely used. As the basis of 3D reconstruction and other medical image processing methods, the accurate segmentation of brain MRI image is of great significance to the development of medical image processing and medical technology. However, in practical applications, there are some defects in MRI images, such as noise, offset field, partial volume effect and low contrast. This leads to the accurate segmentation of brain MRI image becomes a difficult problem. In this paper, the segmentation of MRI images is studied. Because of the importance of brain MRI image segmentation, researchers at home and abroad have put forward many effective methods for MRI image segmentation. Such as active contour model, Markov random field Atlas atlas, fuzzy clustering and so on. Fuzzy C-means algorithm (FCM) is one of the most popular methods used by scholars at home and abroad. However, the standard FCM algorithm is based on the gray level information of the image and does not consider the spatial information of the image, so it is not effective in the segmentation of the image with high noise and high migration field. The bias-corrected FCM-BCFCM) adds the estimation of migration field and the use of spatial information on the basis of FCM, which can solve the effect of image migration field on segmentation. But the BCFCM algorithm does not take into account the effect of noise on the estimation of the migration field, so the estimation of the offset field will be wrong in the segmentation of high noise images, which will have an impact on the segmentation results. In this paper, a fast segmentation method is proposed to remove skull and its appendages in the process of image preprocessing for MRI brain tissue segmentation. In addition, aiming at the defects of BCFCM algorithm, this paper mainly puts forward two improved schemes: an improved algorithm based on BCFCM is proposed. The improved algorithm can adjust the size of the object function window through the estimation of the noise intensity in the iterative process. The comparison experiments show that the improved scheme greatly enhances the robustness of BCFCM algorithm to noise. At the same time, considering the error of the BCFCM algorithm in the offset field, a new algorithm is proposed, in which the offset field is smoothed by introducing Gao Si kernel function, and the estimated value of the offset field is limited by the threshold. The effect of misestimation of offset field on segmentation results can be effectively avoided. The experimental results show that the improved algorithm can not only effectively and accurately segment the brain tissue, but also has a strong ability to resist noise and process the offset field.
【学位授予单位】：沈阳建筑大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：R445.2;TP391.41

【相似文献】