Classification and diagnosis of Alzheimer's disease based on multimodal data

Alzheimer's disease is an irreversible neurodegenerative disease, and exploring early diagnostic methods can benefit patients in obtaining accurate and effective treatment. This study adopted multimodal data of clinical neuropsychological examinations and functional Magnetic Resonance Imaging brain network properties constructed by graph theory. Scales, global brain network properties and local properties with significant differences were used as features in patients with Alzheimer's disease, patients with mild cognitive impairment, and normal elderly people. The feature significances were analyzed, and three features and feature combinations calculated using Support Vector Machine and Naive Bayes Classifiers were compared. The results indicated that the scales and local brain network properties had better classification effects in the diagnosis, and the trichotomous classification accuracy of the two classifiers for all feature combinations was 85.07% and 88.06%, respectively. The feature selection method proposed in this paper has an auxiliary effect on the classification diagnosis.