基于K-Means及K-NN的磁共振颅脑图像分割初探  

The head MRI segmentation based on K-Means and K-Nearest Neighbor rule

作　　者：王臣[1] 叶春涛[1] 

机构地区：[1]复旦大学附属华东医院放射科MRI室,200040

出　　处：《上海医学影像》2011年第1期9-11,14,共4页Shanghai Medical Imaging

基　　金：复旦大学附属华东医院资助(项目编号H-459)

摘　　要：目的对K-均值聚类(K-Means)和K-最近邻规则(K-NN)方法在MR颅脑图像分割中的应用进行研究,分析二者优缺点并提出改进。方法利用K-Means算法和K-NN算法对脑组织进行分割。为了减少人脑的劳动使算法自动进行,提出使用K-Means方法提取K-NN方法的训练样本。结果 K-Means与K-NN及改良后的K-NN算法(KMN)能很好地从大脑结构中分割白质(white matter,WM)、灰质(grey matter,GM)和脑脊液(cerebrospinal fluid,CSF)。结论实验表明K-Means与K-NN能快速有效地分割脑组织,改进后的K-NN方法减少了人工参与,并能获得较好的分割效果。Objective To study the application of K-Means and K-NN method in the MRI brain image segmentation.To analysis and improve them.Methods This method had three main steps.Firstly,segment the MRI image using K-Means.Secondly,segment the MRI image using K-NN.Finally,proposed the improved K-NN method that using K-Means.Results Using our methods to segment 10 group MRI head clinical data,we got good results that gray matter,white matter and cerebrospinal fluid.Conclusion The experimental results shows that these methods can automatically and quickly segment the head tissue.And the modified K-NN method can be independent of manual work and obtain good segmentation results.

关 键 词：磁共振脑图像 医学图像分割 K-均值聚类 K-最近邻规则 

分 类 号：R445[医药卫生—影像医学与核医学]