基于磁共振图像海马三维纹理特征的阿尔茨海默病及健康对照的分类研究  

Classification Studies in Patients with Alzheimer's Disease and Normal Control Group Based on Three-dimensional Texture Features of Hippocampus Magnetic Resonance Images

作　　者：于鲁 夏翃[2] 刘卫芳[2] 

机构地区：[1]首都医科大学,北京100069 [2]首都医科大学生物医学工程学院,北京100069

出　　处：《生物医学工程学杂志》2016年第6期1090-1094,共5页Journal of Biomedical Engineering

基　　金：北京市教育委员会科技发展计划项目资助(SQKM201210025009)

摘　　要：利用磁共振(MR)图像对阿尔茨海默病(AD)和健康对照(NC)进行分类识别,比较双侧海马在分类识别中的意义。选取AD患者和NC各25人,采用灰度共生矩阵和游程长矩阵提取每位受试者的海马部位的三维纹理特征。通过筛选得到组间存在显著差异的纹理特征参量,对主成分分析、线性判别分析和非线性判别分析3种方法得到的识别结果进行比较。利用反向传播(BP)神经网络建立识别模型,对AD和NC进行分类识别,采用相关性分析比较双侧海马纹理参数与简明智力状态检查(MMSE)评分的相关性。结果显示使用神经网络模型的非线性判别分析的分类识别正确率最高,右侧海马分类识别的正确率均高于左侧。两侧海马的纹理特征与MMSE评分均具有相关性且右侧海马的相关性系数均大于左侧。利用三维纹理特征的神经网络模型可分类识别AD组和NC组,并且采用右侧海马进行分类识别可能更有利于AD的诊断。This study aims to explore the diagnosis in patients with Alzheimer＇s disease （AD） based on magnetic resonance （MR） images, and to compare the differences of bilateral hippocampus in classification and recognition. MR images were obtained from 25 AD patients and 25 normal controls （NC） respectively. Three-dimensional texture features were extracted from bilateral hippocampus of each subject. The texture features that existed significant differences between AD and NC were used as the features in a classification procedure. Back propagation （BP） neural network model was built to classify AD patients from healthy controls. The classification accuracy of three methods, which were principal components analysis, linear discriminant analysis and non-linear discriminant analysis, was obtained and compared. The correlations between bilateral hippocampal texture parameters and Mini-Mental State Examination （MMSE） scores were calculated. The classification accuracy of nonlinear discriminant analysis with a neural network model was the highest, and the classification accuracy of right hippocampus was higher than that of the left. The bilateral hippocampal texture features were correlated to MMSE scores, and the relative of right hippocampus was higher than that of the left. The neural network model with three-dimensional texture features could recognize AD patients and NC, and right hippocampus might be more helpful to AD diagnosis.

关 键 词：阿尔茨海默病 三维纹理分析 海马 分类识别 

分 类 号：R749.16[医药卫生—神经病学与精神病学] R445.2[医药卫生—临床医学]