基于多图谱最优标签融合的脑MR图像分割-浙江工贸职业技术学院.PDFVIP

第17卷第2 期 浙江工贸职业技术学院学报 Vol. 17 No.2 2017 年06 月 JOURNALOFZHEJIANGINDUSTRYTRADEVOCATIONALCOLLEGE Jun. 2017 Doi: 10.3969/j.issn. 1672-0105.2017.02.008 基于多图谱最优标签融合的脑MR图像分割* 程有娥 (浙江工贸职业技术学院，浙江温州325003) 摘 要：脑MR 图像分割已经成为医学相关领域研究的热点和难点，是辅助相关疾病临床诊断和医学研究的不可 或缺的一项关键技术。目前，比较有效的一类方法是基于多图谱的脑MR 图像分割方法，该类方法涉及到的一个关键 步骤是多图谱标签的融合问题。本文针对传统的图谱标签融合方法所有图谱标签均参与待标记体素的标签融合，而没 有进行图谱标签选择的问题，提出了一种新的基于图谱图像灰度信息与图谱图像标签信息相结合的图谱标签选择方 法。通过该方法可以从各个图谱中选择合适的最优图谱标签参与标签融合，排除了不合适的图谱标签对标签融合结果 的影响，从而可以提高脑MR 图像的分割精度。通过脑MR 图像的分割实验，并与常用的图谱标签融合方法进行了实 验对比，表明了本文提出的图谱标签融合方法的有效性。 关键词：标签选择；标签融合；多图谱；图像块 中图分类号：TP391.41 文献标识码：A 文章编号：1672-0105(2017)02-0030-05 Brain MR Image Segmentation Based on Multi-Atlas Optimal Label Fusion CHENG You-e （Zhejiang Industry Trade Vocational College, Wenzhou, 325003, China） Abstract: Brain MR image segmentation has become a hot and difficult point in the field of medical research, and it is one of the key technologies for clinical diagnosis and medical research. At present, one of the most effective methods is based on the multi-at- las MR image segmentation, and multi-atlas label fusion is a key step of this kind method. In this paper, we focus on the traditional la- bel fusion methods that all of the atlas labels are involved in the label fusion process, but there is no choice of the atlas label. In this pa- per, a new label selection method based on the gray level information and the label information of the atlas is presented. This method can select the optimal atlas label from all of the atlas, excluding the impact of labels that not suitable for label fusion, which can im- prove the accuracy of the segmentation of brain MR images. Through the experiment of brain MR image segmentation, and the com-