Classification of Alzheimer’s Disease Using Conventional Machine Learning Methods with Cortical and Genetic Characteristics

Abstract

The conventional machine learning (ML) methods have been used in the classification of Alzheimer’s disease (AD), while the performance was still needed to be addressed for being short of the standard datasets and implemented approaches. In our study, one scheme was proposed to verify the robustness of proposed methods. A total of one hundred T1-MRI data were downloaded and divided into three groups: AD (34 cases), mind cognition impairment (MCI) (45 cases) and normal control (NC) (21 cases). In our method, four cortical characteristics of cortical thickness (CT), gray matter volume (GMV), local gyrification index (LGI) and cortical surface area (CSA) for brain regions and one genetic characteristic of ApoE alleles were used. With the optimized characteristics by one-way analysis of variance (ANOVA), Krystal-Wallis H test and ReliefF algorithms, five MLs including random forest (RF), support vector machine (SVM), linear discriminant analysis (LDA), logistic regression (LR) and decision trees (DTs) are compared. With the improved cortical features, the classifiers of RF and LDA showed the highest accuracy of 89% for the binary classification (AD vs NC). The DTs and SVM classifier achieved 92% for the three-classification (AD vs MCI vs NC). The chosen brain cortical and genetic characteristics showed valuable capability in AD classification and the proposed scheme was robust for performance evaluation.