【摘要】 目的利用支持向量机(SVM)和随机森林(RF)构建模型,对正常认知(NC)组、早期轻度认知障碍(EMCI)组、晚期轻度认知障碍(LMCI)组和阿尔茨海默病(AD)组进行分类研究,比较模型分类的准确率。资料与方法选取ANDI数据库中543例研究对象,按照数据库对发病进程的分类,分为NC组139例、EMCI组220例、LMCI组108例和AD组76例。272项结构性MRI(sMRI)经特征提取出28项sMRI数据后进入SVM、RF分类训练器,用10-折交叉验证训练模型获得分类准确率,用受检者工作特征(ROC)曲线评价两种分类模型的分类效能。结果 RF和SVM分类器在NC组与AD组、EMCI组与AD组和NC组与LMCI组均有较高的预测准确率,其中对NC组与AD组准确率最高(RF为96.45%,SVM为90.90%)。基于28项sMRI特征的RF分类器在NC与EMCI、NC与LMCI、NC与AD两两分类的敏感度、特异度和ROC曲线下面积均高于SVM。结论基于28项s MRI的RF模型可为AD的分类诊断提供指导。更多还原

【Abstract】 Purpose To conduct classification study of the normal cognitive(NC) group, early mild cognitive impairment(EMCI) group, late mild cognitive impairment(LMCI) group and Alzheimer’s disease(AD) group by using support vector machine(SVM) and random forest(RF) models, and compare the accuracy of model classification. Materials and Methods A total of 543 subjects in the ANDI database were selected, and according to the classification of the pathogenesis process by the database, all subjects were divided into the NC group(139 cases), EMCI group(220 cases), LMCI group(108 cases) and AD group(76 cases). Twenty-eight structural magnetic resonance imaging(sMRI) data of 272 sMRIs items were extracted by features and then entered into SVM and RF classification trainers. The classification accuracy was obtained by 10-fold cross-validation training model, and the classification efficiency of the two classification models was evaluated by the receiver operating characteristic(ROC) curve. Results RF and SVM classifiers had higher prediction accuracy in NC and AD group, EMCI and AD group, NC and LMCI group, with the highest accuracy for NC and AD group(RF was 96.45% and SVM was 90.90%). RF classifiers based on 28 sMRI features had higher sensitivity, specificity and ROC curve area than SVM in pairwise classification of NC and EMCI, NC and LMCI, NC and AD. Conclusion The RF model based on 28 sMRIs can provide guidance for the classification and diagnosis of AD.更多还原