Machine Learning Classification of Alzheimer’s Disease Using Joint Features of Diffusion Tensor Imaging and Clinical Scales

Abstract

The computer-aided diagnosis techniques have been widely applied in the Alzheimer’s disease (AD) classification, especially machine learning algorithms. And the selection of features acts on the AD classification. In this paper, we collected the diffusion tensor imaging (DTI) images of 375 subjects from the ADNI, including 43 AD, 187 mild cognition impairment (MCI) and 145 normal controls (NC). The DTI indexes of fractional anisotropy, mean, axial and radial diffusivity (FA, MD, AxD and RD) measured by PANDA software and clinical scales were screened by one-way analysis of variance (ANOVA) and ReliefF algorithms, respectively. Then the optimized DTI features and the joint features were used for the classification of AD. The performance of six classifiers of random forest (RF), support vector machine (SVM), logistic regression (LR), linear discriminant analysis (LDA), K-Nearest Neighbor (KNN) and decision trees (DTs) were evaluated by the accuracy (ACC), precision (PRE), recall (REC), F1-Score and area under curve (AUC). Our results showed that the LR and RF had the highest classification accuracy of 97.87% with the joint features. It was concluded that the joint features of diffusion tensor imaging and clinical scales cloud improve the accuracy of AD classification.