改进K均值与模糊置信度的脑部MRI分割  

Brain MRI image segmentation based on improved K-means clustering and fuzzy confidence method

作　　者：刘小明[1,2] 喻杰[1,2] 刘俊[1,2] 梅明[1,2] 

机构地区：[1]武汉科技大学计算机科学与技术学院,湖北武汉430065 [2]武汉科技大学湖北省智能信息处理与实时工业系统重点实验室,湖北武汉430065

出　　处：《计算机工程与设计》2015年第3期710-715,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目(31201121);湖北省教育厅重点基金项目(D20131101);湖北省重点实验室基金项目(znss2013A006)

摘　　要：针对脑部磁共振图像(MRI)的灰度分布特性,提出一种结合灰度距离加权K-means聚类与模糊置信度的混合医学图像分割方法。采用改进的灰度加权K-means聚类方法对MRI图像进行训练分类得到粗略分类结果,运用基于支持向量数据域描述(SVDD)的模糊置信度方法对每个类精细分割,得到脑部各组织的输出图像。该算法分割时逐渐增大目标模糊置信度门限,通过对模糊置信度的动态优化来逼近最佳分割结果。在脑部MRI图像上的实验结果表明,该方法在处理图像灰度分布不均匀、存在孤立点、细化轮廓等问题时具有较高的准确度和鲁棒性。Aiming at grayscale distribution characteristics of the brain MRI image,a medical image segmentation method combining intensity distance weighted K-means clustering and fuzzy confidence method was proposed.Firstly,an MRI image was classified with the improved gray weighted K-means clustering method to obtain a rough segmentation result.Then,the fuzzy confidence method based on support vector domain description（SVDD）was used to further refine each class.Finally,the output image was obtained according to the segmentation of brain image organs.The segmentation algorithm gradually increased target fuzzy confidence threshold,based on the dynamic optimization of fuzzy confidence degree,to achieve optimal segmentation results.Experimental results on MRI brain images show the robustness and accuracy of the method when dealing with the uneven gray distribution of target,isolated points and thin contours.

关 键 词：图像分割 K-MEANS聚类 支持向量数据域描述 模糊置信度 磁共振图像 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]