基于MR图像的阿尔茨海默病和轻度认知障碍患者海马三维纹理分析

摘要

阿尔茨海默病（Alzheimer disease，AD）是老年痴呆症中最常见的类型，有研究表明，轻度认知障碍（mildcognitive impairment，MCI）可能是AD的前驱阶段．研究样本包括21例早期AD患者（AD组）、22例MCI患者（MCI组）及20例健康对照者（normal control，NC）的磁共振（magnetic resonance，MR）脑图像，采用灰度共生矩阵和游程长矩阵方法对图像中海马结构进行三维纹理分析，比较纹理特征——能量、熵、游程长不均匀度因子和灰度不均匀度因子在3组之间的差异，并分析这些纹理参数与临床广泛应用的简易智能状态检查量表评分之间的相关性，同时利用支持向量机方法对样本进行分类识别．研究结果显示，以上纹理参数在3组之间存在显著差异，且纹理参数与临床简易智能状态检查量表评分具有相关性，对AD组与NC组的分类正确率可达85．37％（敏感性和特异性分别为90．48％、80％）．表明三维纹理分析反映出了早期AD及MCI患者海马结构的病理变化，有助于AD的早期诊断．%Alzheimer＇s disease （AD） is the most common type of dementia. Mild cognitive impairment （MCI） is considered as a preclinical stage of AD. We studied the MR images of 21 early AD patients, 22 MCI patients and 20 normal controls with 3D texture analysis. Texture features of the hippoeampus were extracted from the gray level co-occurrence matrix and run length matrix. These parameters including energy （ENG）, entropy （ETP）, grey level nonuniformity （GLN） and run length nonuniformity （RLN） were significantly different among three groups. Meanwhile, the correlations between parameters and MMSE were calculated, which are widely used in clinical applications. The groups were distinguished based on support vector machines （SVM）. Results show that the texture features above are significantly different among three groups, and they are correlated with MMSE scores. The classification accuracy for AD and NC groups is 85.37% （ sensitivity and specificity are 90.48% and 80% , respectively）. 3D texture analysis can reflect the pathological changes of hippocampus in patients with early AD, and can be helpful for early diagnosis of AD.