Early diagnosis or detection of Alzheimer's disease (AD) and mild cognitive impairment (MCI) is crucial so as to intervene in advance and to better understand the neurodegenerative process. Gray matter volume (GMV) plays an important role in demonstrating unique anatomical characteristics of the brain regions and further differentiates AD, MCI and normal control (NC). In this study, 317 subjects (100 NC, 58 stable MCI (sMCI), 53 converted MCI (cMCI) and 106 AD) are selected from the Alzheimer's Disease Neuroimaging Initiative database. First, the differences of GMV patterns among the between-group comparisons at different time points and the development of longitudinal pattern within the same group are compared. Next, the longitudinal feature combination strategy is applied to construct the classification model by using a support vector machine (SVM) combined with the nested leave-one-out crossvalidation (LOOCV) method. The brain structure experiences a gradual change in the process of developing from NC to AD. In addition, the baseline GMV combined with the longitudinal measurements for 2 years of follow-up data yielded optimal classification results. Specifically, the AD-NC comparison achieves the best classification performance with 98.06% accuracy, 97.17% sensitivity, 99.00% specificity, 99.04% positive predictive value (PPV) and 97.06% negative predictive value (NPV). The comparison of the two subtypes of MCI (ie, sMCI and cMCI) also achieves high accuracy. Other between-group comparisons also receive high classification performance. According to statistics, caudate nucleus, hippocampus, temporal pole and lenticular putamen are the most important contribution areas to the between-group comparisons. Our research has the potential to improve the clinical diagnosis of subtypes of MCI and predict the risk of its conversion to AD.
K E Y W O R D SAlzheimer's disease, gray matter volume, longitudinal analysis, longitudinal classification, mild cognitive impairment