电子学报 ›› 2017, Vol. 45 ›› Issue (9): 2202-2209.DOI: 10.3969/j.issn.0372-2112.2017.09.022

• 学术论文 • 上一篇    下一篇

基于多参数配准模型的脑核磁影像分割算法

张万1, 刘刚1, 朱凯1,2, 廖恒旭1   

  1. 1. 上海电力学院自动化工程学院, 上海 200090;
    2. 国网浙江省电力公司金华供电公司, 浙江金华 321000
  • 收稿日期:2016-06-20 修回日期:2016-11-02 出版日期:2017-09-25
    • 通讯作者:
    • 刘刚
    • 作者简介:
    • 张万,男,1990年3月出生于湖北省孝感市.现为上海电力学院自动化工程学院硕士研究生.主要研究方向为机器视觉与图像处理.E-mail:zwspark@163.com;朱凯,男,1988年7月生于河南省永城市.现为国网浙江省电力公司金华供电公司员工,2016年毕业于上海电力学院.主要研究方向为机器学习与图像处理.E-mail:zhukai0729@163.com;廖恒旭,男,1993年2月出生于上海市.现为上海电力学院自动化工程学院硕士研究生.主要研究方向为深度学习与图像处理.E-mail:liaohengxu@126.com
    • 基金资助:
    • 国家自然科学基金 (No.61203224); 上海市教育委员会创新项目 (No.13YZ101)

Multi-parameter Registration Model for Brain MR Image Segmentation Based on Label Fusion

ZHANG Wan1, LIU Gang1, ZHU Kai1,2, LIAO Heng-xu1   

  1. 1. College of Automation Engineering, Shanghai University of Electric Power, Shanghai 200090, China;
    2. Jinhua Power Supply Company, State Grid Corporation, Jinhua, Zhejiang 321000, China
  • Received:2016-06-20 Revised:2016-11-02 Online:2017-09-25 Published:2017-09-25
    • Supported by:
    • National Natural Science Foundation of China (No.61203224); Innovation Program of Shanghai Municipal Education Commission (No.13YZ101)

摘要: 配准技术在基于多图谱的分割方法中能有效地将医学图谱的先验知识融入分割过程,再结合以高效的标记融合算法,最终实现精确地自动分割.针对图谱配准的较大误差及其对标记融合的重要影响,本文建立了一种新的概率图模型框架并以此提出了基于多参数配准模型的分割算法,将此方法与高效的标记融合算法相结合,可以提高目标图像中特定组织区域的分割精度,更使其在少量图谱分割的情形下具有重要应用.首先,使用多种配准参数对所有目标图像进行配准;然后,分别采用不同的算法对配准图像进行灰度融合和标记融合,实现训练图像的重构过程;最后,利用高效的标记融合算法对重构后的图像进行融合得到最终精确的分割结果.实验结果表明该方法均优于本文其他分割算法,能够有效提升脑部组织分割精度.

关键词: 图像分割, 图像配准, 标记融合, 多参数配准模型, 脑核磁影像

Abstract: Registration technology can effectively integrate the prior knowledge of medical atlases into the segmentation process,and then combine with the efficient label fusion algorithm to obtain the segmentation results accurately and automatically.Aimed at the large error in registration of target image and its great influence on label fusion,a framework of probabilistic graphical model is established and the idea of multi-parameter registration model is proposed.Combined with an efficient algorithm on label fusion,this framework can improve the segmentation accuracy of specific tissue regions on target image,which has important application value in segmentation with a few available atlases.After the multi-parameter registration and the reconstruction process of training sets on target images,the final segmentation results are obtained by an efficient fusion algorithm.According to the experiment which was conducted on the brain magnetic resonance image segmentation with different segmentation methods,the proposed framework can effectively improve the accuracy of segmentation.

Key words: image segmentation, image registration, label fusion, multi-parameter registration model, brain magnetic resonance images

中图分类号: