中文摘要：

为了及早检测轻度认知障碍（MildCognitiveImpairment，MCI），降低阿尔茨海默病的患病率，文章提出一种基于径向基神经网络及核支持向量机（RadialBasisFunction-kernelSupportVectorMachine，RBF-kS—VM）的MCI检测系统，该系统首先读取三维磁共振脑图像并预处理，然后通过主成分分析（PrincipalCompo—nentAnalysis，pCA）降低特征维数，采用RBF核支持向量机作为分类模型，RBF的参数通过优化选择。实验数据采用OASIS公共数据库，选择50例正常对照组（NormalControl，NC）与50例MCI患者。十折交叉验证结果显示文中所提出方法的敏感度为84％、特异度为78％、准确度为81％，优于前向神经网络、决策树、支持向量机、齐次与非齐次核支持向量机方法。文中构建的RBF核支持向量机有效，可用于MCI检测。

英文摘要：

In order to detect mild cognitive impairment（MCI） and reduce the morbidity rate of Alzhei- mer disease, a novel MCI detection system based on radial basis function-kernel support vector ma- chine（RBF-kSVM） was developed. The system read the 3D magnetic resonance（MR） images with preprocessing, and then employed the principal component analysis（PCA） to reduce the feature dimen- sions, followed by using RBF kernel SVM as the classification model. The parameter of RBF was cho- sen by optimization method. OASIS public data were obtained from Internet, and 50 normal controls （NCs） and 50 MCIs were picked up. The results of 10-fold cross validation showed that the proposed method achieved desired results as 840~ sensitivity, 78~ specificity and 81~ precision, which were superior to the results of forward neural network, decision tree, SVM, homogeneous and inhomoge- neous polynomial kSVM. So the proposed RBF-kSVM is remarkably effective in detecting MCI.