基于空间特征联合核稀疏表示的脑肿瘤提取方法

摘要

To improve the brain tumor extraction accuracy by muhisequence MR image,the spatialfeature combining kernel sparse representation (KSR) was proposed to connect spatial structure information and intensity feature information for brain tumor extraction.The sub-dictionary of each label was built,and the neighboring filtering kernel based on KSR was applied to extract brain tumor from MSMR images.The spatial information and the intensity feature information were combined in the proposed method to improve the accuracy of brain tumor extraction.The clinical and simulation data from MICCAI BraTS dataset were divided by the proposed method.The results show that compared with sparse representation method,the proposed brain tumor extraction method based on spatial-feature kernel sparse representation can increase the brain tumor extraction accuracy by 5％ ～ 6％ due to the introducing of spatial structure information.%为了提高融合多序列MR图像应用于脑肿瘤提取时分割区域的准确性,基于核稀疏表示分类方法,联合多序列MR图像中的空间结构和灰度特征信息,提出一种空间特征联合的脑肿瘤核稀疏表示分类方法.首先构建各个类别的子字典,再用邻域滤波核稀疏表示方法对多序列脑MR图像进行分类,该邻域滤波核可以有效地将灰度特征与空间结构结合起来提高脑肿瘤提取的准确性.对国际数据库MICCAI BraTS提供的临床和仿真数据进行分割.结果表明:与稀疏表示分类方法相比,所提出的基于空间特征联合核稀疏表示的脑肿瘤提取方法由于增加了空间结构信息,所得的提取准确率提高了5％～6％.