摘要
脑肿瘤分割在疾病的辅助诊断、治疗方案规划以及手术导航中扮演着重要的角色.为了分割病变的肿瘤,提出了一种基于卷积神经网络特征提取的由2个串联的阶段组成的分割方法,提取完整的特征并使用分类器分类.在训练阶段,训练了Convolution Neural Network(CNN)学习从图像空间到肿瘤标记空间的映射.在测试阶段,使用从CNN得到的标记输出,随同测试的灰度图像送到一个Support Vector Machine(SVM)分类器中,以得到精确的分割.实验结果证明,该方法能自适应脑肿瘤的差异性,分割准确率最高达到93%.
Brain tumor segmentation plays an important role in assisting diagnosis,planning treatment and surgical navigation.In this paper,a segmentation method based on convolution neural network feature extraction to segment the lesion of the tumor is proposed.Our algorithm is composed of two cascaded stages.In the first stage,we train CNN to learn the mapping from the image space to the tumor label space.During the testing phase,we use the predicted label output from CNN and send it along with the testing image to an SVM classifier for accurate segmentation.The experimental results show that the method can adapt to the difference of brain tumor,and the maximum segmentation accuracy is up to 93%.
作者
谢铭超
韩旭
栾帅
李芳
王春兴
XIE Ming-chao;HAN Xu;LUAN Shuai;LI Fang;WANG Chun-xing(Department of Physics And Electronics Shandong Normal University,250000,Jinan;Qingdao E-Government and Information Resources Management Office,266000,Qingdao,Shandong,PRC)
出处
《曲阜师范大学学报（自然科学版）》
CAS
2018年第2期49-53,共5页
Journal of Qufu Normal University(Natural Science)
基金
国家自然科学基金(11674198)
关键词
卷积神经网络
支持向量机
特征提取
脑肿瘤分割
convolutional neural networks
support vector machine
feature extraction
brain tumor segmentation
