基于窄带水平集的带标记线左心室核磁共振图像的自动分割  

An Auto Segmentation Method of gagged Left Ventricle Magnetic Resonance Images Based on Narrow Band Level Set

作　　者：陈强[1] 周则明[1] 屈颖歌[1] 王平安[2] 夏德深[1] 

机构地区：[1]南京理工大学计算机系,南京210094 [2]香港中文大学计算机科学与工程系

出　　处：《模式识别与人工智能》2005年第1期107-113,共7页Pattern Recognition and Artificial Intelligence

基　　金：香港特区政府研究资助局资助项目(No.CUHK 4180/01E;CUHK 1/00C)

摘　　要：提出了一种全自动分割带标记线左心室核磁共振图像的方法。它主要由两部分组成:先采用支持向量机(SVM)进行左心室定位,从而给出初始轮廓线;然后用改进的窄带水平集(Level Set)方法演化曲线得到最终分割结果。该方法改进了窄带生成方法,减少了窄带生成时间。针对带标记线左心室核磁共振图像的成像特点,引入了块像素变差和灰度相似性的思想对水平集方法的速度项进行了改进。实验结果表明,该方法能全自动、快速、准确地实现左心室的分割。An auto segmentation method is presented for tagged left ventricle MR images. It mainly consists of two parts: First, this method localizes the left ventricl based on SVM and gives the initial contour. Second, it evolves the initial curve based on the improved narrow band level set, and achieves the final segmentation. This method improves the generating method of narrow band and decreases the generating time. Focusing on the imaging characters of tagged left ventricle MR images, we introduce the block--pixel variation (BPV) and the intensity comparability to improve the speed term of level set. The results demonstrate that the method advanced in this paper can automatically, quickly and accurately achieve the segmentation of the left ventricle.

关 键 词：水平集 窄带 支持向量机 核磁共振图像分割 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]