用于脑部核磁共振图像分割的具有抗噪能力的BCFCM算法

脑部核磁共振成像(MRI)是脑疾病临床诊断的重要手段,而脑组织的准确分割则是其中一个重要的环节。然而MRI图像中普遍存在的噪声和偏移场给脑组织的准确分割造成了很大的困难。在MRI图像分割算法中,偏移场矫正模糊C-均值算法(BCFCM)在模糊C-均值聚类算法(FCM)的基础上增加了对偏移场的估计和空间信息的使用,可以很好地消除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,因此对高噪声图像的分割效果欠佳。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,提出基于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地改变目标函数窗口的大小。同时,该算法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,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,改进后的算法不仅可以有效准确地 分割脑组织,而且具有较强的抗噪声和处理偏移场的能力。

Anti-noise BCFCM Algorithm for Brain MRI Segmentation

Magnetic resonance imaging (MRI) of brain is an important tool for the clinical diagnosis of brain diseases.The accurate segmentation for brain tissues is one of the important parts.However,it is difficult to acquire the accurate segmentation results because of the noise and intensity inhomogeneities in MRI.Among the MRI segmentation me-thods,Bias-Corrected FCM (BCFCM) algorithm based on Fuzzy C-Means (FCM) algorithm utilizes the spatial information and estimation of intensity inhomogeneities which can deal with the problem caused by intensity inhomogeneities.Because the BCFCM algorithm fails to consider the high level noise when estimating intensity inhomogeneities,the segmentation results are not accurate enough.For the MRI of brain tissue segmentation,this paper proposed a fast segmentation method to remove the brain skull and its appendages during the image preprocessing.In addition,we proposed an improved algorithm based on the BCFCM algorithm.The improved BCFCM algorithm can automatically change the size of window in the objective function by estimating the noise level in the iterative processing.Besides,the Gaussian kernel in the object function was utilized to smooth the intensity inhomogeneities,and the estimation value of intensity inhomogeneities was limited by using an experimental threshold which can effectively avoid the incorrect estimation of intensity inhomogeneities in the segmentation results.The experimental results show that the proposed algorithm can not only effectively and accurately segment the brain tissues,but also deal with high level noise and intensity inhomogeneities.