摘要
丘脑、海马体、伏隔核、尾状核等关键脑结构的位置、体积、形态等的变化与多种脑部疾病息息相关,对其精准分割是进行相关定量分析的前提.然而在磁共振图像中这些结构对比度不高、边界模糊,传统方法只是利用了标号图像,没有考虑到待分割图像、先验信息等,因无法实现准确分割.文中将多图谱配准与活动轮廓模型相结合,提出了一种新的多图谱活动轮廓模型框架,有效地利用了图谱的先验信息和待分割图像的灰度信息,将多图谱的形状先验项引入到活动轮廓模型中,并在融合标号图像的过程中利用活动轮廓模型校正配准引起的误差,可以得到光滑、准确的分割结果.该框架包含3个部分:第1部分为图谱先验项,利用配准的局部相似性作为权重融合多个图谱的信息;第2部分为数据项,利用待分割图像的局部信息,可以校正配准中的误差;第3部分为平滑项,用于保证曲线在演化过程中的平滑.大量的实验表明了该方法的有效性和准确性.
Various brain diseases are closely related with the position,size and shape changes of key brain structures,such as hippocampus,accumbens,caudatum,and thalamus.It requires the accurate segmentation of these key brain structures to measure their changes.However,the traditional segmentation is not accurate enough because these structures have no clear contrasted boundaries in MR images.In this paper,we propose a novel segmentation framework that combined the multi-atlas registration and active contour model.We utilize the prior information of atlases and the gray information of target image.The shape prior information be incorporated into the active contour model modeling to improve the segmentation performance.the active contour model will correct the label errors in the label fusion procedure.The framework consists of three terms.First one is the atlas prior term,which uses the local similarity measure as weight to incorporate the atlas information.The second one is data term,which uses the local information of target image to correct the registration errors.The third one is smooth term,which was used to ensure the smoothness of the evolution curve.Experiments results demonstrate the efficacy and accuracy of the proposed method.
出处
《计算机学报》
EI
CSCD
北大核心
2016年第7期1490-1500,共11页
Chinese Journal of Computers
基金
广东省自然科学基金(2014A030313316
2016A030313574)
广州珠江新星专项(2012J2200041)资助
