北京生物医学工程

___________基于LBP的改进Random_Walks算法在颅脑磁共振影像分割中的应用_________

Brain tissue segmentation in MRI with improved Random Walks based on local binary patterns

作者：刘伟童同黄煜峰冯焕清

单位：           中国科学技术大学电子科学与技术系(合肥230027)    

关键词：随机游走；局部二值模式；先验概率；脑组织；分割

分类号：

出版年·卷·期（页码）：2013·32·3（237-242）

摘要：

目的由于颅脑结构复杂且颅脑磁共振影像易受噪声、磁场不均匀性、部分容积效应等因素的影响，

精确的脑组织分割方法仍需深入研究。方法本文提出一种基于Random Walks的改进算法以提高脑白质、

脑灰质及脑脊液分割的准确性。通过引入局部二值模式(local binary patterns，LBP)改进了传统Random

Walks权重函数的构造，在反映相邻像素灰度变化信息的同时包含了局部图像的纹理信息，有利于合并同

质区域并增强边缘轮廓的识别。本文还使用了灰度先验概率模型减少Random Walks种子点交互的次数。结

果实验结果表明基于LBP的改进算法在多种不同水平的噪声及不均匀场作用下，能够有效识别磁共振影像

中脑组织区域的边缘轮廓，并对噪声有良好的鲁棒性。结论基于LBP的改进Random Walks算法可精确分割

颅脑磁共振影像。

Objective Accurate segmentation of brain tissue in MRI is an essential step and

remains a challenging problem because of acquisition noise，non-uniformities in the MR

magnetic field，partial volume effects and the complex anatomy structure of the brain.

Methods In this paper，we presented a novel approach based on Random Walks (RW) to extract

white matter (WM)，gray matter (GM) and cerebrospinal fluid (CSF). To overcome the

shortcomings of Random Walks，we introduced the concept of local binary patterns (LBP) into

Random Walks to construct a new weighting function. The new weighting function not only

reflected the changing information of adjacent-pixel’s gray value，but also contained the

texture information of local image，which could strengthen the ability of RW to identify

homogeneous pixels and edges. We also achieved a better performance with prior probability

model. Results Experiment results were analyzed against different levels of noise and bias

field，and the proposed method performed better discriminative power of identifying the

brain tissue boundary. Conclusions This improved Random Walks based on LBP segments brain

tissue images accurately.

参考文献：

服务与反馈：

【文章下载】【加入收藏】

提示：您还未登录，请登录！点此登录