波谱学杂志 ›› 2019, Vol. 36 ›› Issue (3): 268-277.doi: 10.11938/cjmr20182686

• 研究论文 • 上一篇    下一篇

基于脑部磁共振图像三维局部模式变换特征提取进行阿尔茨海默病病程预测分类

孙京文, 闫士举, 韩勇森, 宋成利   

  1. 上海理工大学 医疗器械与食品学院, 上海 200093
  • 收稿日期:2018-10-15 发布日期:2018-12-12
  • 通讯作者: 闫士举 E-mail:yanshiju@usst.edu.cn

Classifying the Course of Alzheimer's Disease with Brain MR Images and a Method Based on Three-Dimensional Local Pattern Transformation

SUN Jing-wen, YAN Shi-ju, HAN Yong-sen, SONG Cheng-li   

  1. School of Medical Instrument and Food Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China
  • Received:2018-10-15 Published:2018-12-12

摘要: 本文提出一种三维局部模式变换提取进行纹理特征并与常规特征相融合的方法,基于脑部磁共振图像,对认知功能正常的健康人体(CN)、轻度认知障碍(MCI)患者和阿尔茨海默病(AD)患者进行预测分类.首先对46例CN对照组、61例MCI患者和25例AD患者的脑部磁共振图像提取感兴趣区域,然后提取双侧海马体组织、灰质和白质的三维局部模式变换纹理特征和常规特征,并将两类特征融合,使用支持向量机分类算法进行分类.结果显示利用本方法,基于双侧海马体组织对AD组和CN组进行分类的准确率为88.73%、敏感度为78.00%、特异度为95.7%、受试者工作特征(ROC)曲线下面积(AUC)为0.886 5;基于灰质的准确率为85.92%、敏感度为80.00%、特异度为86.6%、AUC为0.854 3.这证明基于海马体磁共振图像,利用本文提出的改进三维局部模式变换提取的纹理特征进行阿尔茨海默病病程分类效果较好,融合常规特征后更可提高分类预测的精度.

关键词: 脑部磁共振图像, 纹理特征, 三维局部模式变换, 阿尔茨海默病, 分类

Abstract: A classification method was developed to differentiate cognitive normal controls (CN), mild cognitive impairment (MCI) patients and Alzheimer's disease (AD) patients from the magnetic resonance (MR) image data. Three-dimensional (3D) local pattern transformation was used in the proposed method to obtain texture features, which were then fused with the conventional image features for the classification purposes. Region of interests (i.e., bilateral hippocampus, gray matter and white matter) were selected from the MR images of 46 CN, 61 MCI patients and 25 AD patients, from which the 3D local pattern transformation texture features and conventional image features were extracted, fused and used for classification with the support vector machine. It was demonstrated that the accuracy, sensitivity, specificity and area under the curve (AUC) were 88.73%, 78.00%, 95.7% and 0.886 5, respectively, for the fused texture feature/conventional image features in bilateral hippocampus of CN controls and AD patients. In comparison, the fused features in the gray matter gave an accuracy, sensitivity, specificity and AUC of 85.92%, 80.00%, 86.6% and 0.854 3, respectively. It is concluded that the texture features extracted from 3D local pattern transform in hippocampus could be used in conjunction with the conventional image features for better classification of the course of Alzheimer's disease.

Key words: brain magnetic resonance image, texture feature, three-dimensional local pattern transform, Alzheimer's disease, classification

中图分类号: