基于MR图像纹理特征的阿尔茨海默病分类模型

摘要

Objective To establish an early classification and prediction model of Atzheimer's disease (AD) based on texture parameters of hippocampus of MR image.Methods The data were collected from ADNI database of National Institute on Aging,NIH.Magnetic resonance (MR) brain images were collected and extracted based on left,right and bilateral hippocampal images.Region growing algorithm and Contourlet transformation were used to extract texture features.Combined with the basic information of the research objects and texture features,Gaussian process classification method was used to establish a diagnosis model for AD patients and healthy control subjects,and a predictive model from mild cognitive impairment (MCI) into AD.The sensitivity,specificity and area under the ROC curve were evaluated.Results A total of 420 research objects were included in the study.The sensitivity,specificity and area under the ROC curve of the bilateral hippocampal images in the diagnosis model for AD patients and heahhv control subjects were 92.7％,87.1％,and 0.922,respectively,which were higher than those based on the left or right hippocampal area model.The sensitivity was 82.4％ and the area under the ROC curve was 0.836 in AD early prediction model based on MCI data.Conclusions Contourlet texture of the hippocampus can be used to construct the predictive model to identify the early stage of AD,which helps to monitor the progression of MCI to AD,providing evidence for the prevention and treatment of AD.%目的 基于MR图像,提取脑部海马区域纹理特征参数建立阿尔茨海默病(Alzheimer disease,AD)的早期分类预测模型.方法 研究数据来源于美国国立老年研究所ADNI数据库,收集研究对象的磁共振(magnetic resonance,MR)脑图像,分别基于左、右和双侧海马图像,通过区域增长法和Contourlet变换提取纹理特征参数,结合研究对象的基本信息作为特征变量采用高斯过程分类方法建立AD患者和健康对照的诊断模型以及轻度认知障碍(mild cognitive impairment,MCI)患者转变为AD的预测模型,并评价模型的灵敏度、特异度以及ROC曲线下面积.结果 研究共纳入420例研究对象.基于AD和健康对照两组构建的分类模型,双侧海马区的灵敏度、特异度以及ROC曲线下面积分别为92.7％、87.1％和0.922,均大于基于左侧或右侧海马区图像建立的模型.基于MCI数据建立的AD早期预测模型中,灵敏度最高为82.4％,ROC曲线下面积最高为0.836.结论 基于脑部海马区的Contourlet纹理特征构建预测模型,可以识别AD早期的病变情况,这将有助于早期监测MCI进展为AD,为减缓和治疗AD发病提供依据.