基于二维集合经验模式分解的距离正则化水平集磁共振图像分割  

Distance regularized level set evolution in magnetic resonance image segmention based on bi-dimensional ensemble empirical mode decomposition

作　　者：范虹[1] 韦文瑾 朱艳春[2] 

机构地区：[1]陕西师范大学计算机科学学院,西安710062 [2]中国科学院深圳先进技术研究院生物医学与健康工程研究所,深圳518055

出　　处：《物理学报》2016年第16期268-277,共10页Acta Physica Sinica

基　　金：陕西省自然科学基金(批准号:2014JM2-6115);陕西省科学技术研究发展计划(批准号:2012K06-36);国家自然科学基金(批准号:41271518)资助的课题~~

摘　　要：针对现有磁共振(MR)图像分割算法大多直接在原图像上进行处理,分割效果受噪声影响较大的问题,本文引入二维集合经验模式分解(BEEMD)算法,提高距离正则化水平集(DRLSE)方法对MR图像的分割精度.算法中首先使用BEEMD将待分割MR图像分解为多个二维固有模式函数(BIMF),通过对各BIMF赋予不同加权系数重构待分割图像,从而增强分割目标;然后在DRLSE的边界指示函数中添加部分BIMF分量,恢复因高斯平滑被模糊的目标轮廓,并使用DRLSE方法对重构图像进行分割.通过对仿真图像和临床MR图像分割验证,表明本文算法具有较高的分割精度和鲁棒性,能有效实现对临床MR图像的分割.Original image is directly processed by the existing image segmentation algorithms, which is easily affected by noise. A bi-dimensional ensemble empirical mode decomposition （BEEMD） method is introduced to improve the accuracy of MR image segmentation by distance regularized level set （DRLSE） method. The BEEMD method is the extension of one-dimensional noise assisted data analysis from ensemble empirical mode decomposition （EEMD）. The key points of BEEMD are as follows. four-neighborhood optimization is used to find extermum; three-spline interpolation is used to obtain the envelope;amplitude standard of added white noise is restricted;a certain time of integration is used to avoid modality aliasing problem. The main steps of the proposed method are as follows. Firstly, the MR image is decomposed into a number of two-dimensional intrinsic mode functions （BIMF） by BEEMD method;different weighting coefficients are endued to BIMF for image reconstruction to enhance the segmentation target. Secondly, part of BIMF components are added into edge indicator function of DRLSE to recover the blurring boundary caused by Gauss smooth operation. Then DRLSE is used to segment the reconstructed MR image. High accuracy and robustness of proposed algorithm are obtained in both simulations and clinical MR images. However, compared with DRLSE, the proposed method is complex and time consuming because using BEEMD for preprocessing the segmentation image.

关 键 词：磁共振图像 距离正则化水平集 二维集合经验模式分解 固有模式函数 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]