[P4–050]: GRAPH‐THEORY ANALYSIS SHOWS A HIGHLY EFFICIENT BUT REDUNDANT NETWORK IN MCI TAU PROPAGATION

Weighted Random Support Vector Machine Clusters Analysis of Resting-State fMRI in Mild Cognitive Impairment

The identification of abnormal cognitive decline at an early stage becomes an increasingly significant conundrum to physicians and is of major interest in the studies of mild cognitive impairment (MCI). Support vector machine (SVM) as a high-dimensional pattern classification technique is widely employed in neuroimaging research. However, the application of a single SVM classifier may be difficult to achieve the excellent classification performance because of the small-sample size and noise of imaging data. To address this issue, we propose a novel method of the weighted random support vector machine cluster (WRSVMC) in which multiple SVMs were built and different weights were given to corresponding SVMs with different classification performances. We evaluated our algorithm on resting state functional magnetic resonance imaging (RS-fMRI) data of 93 MCI patients and 105 healthy controls (HC) from the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort. The maximum accuracy given by the WRSVMC is 87.67%, demonstrating excellent diagnostic power. Furthermore, the most discriminative brain areas have been found out as follows: gyrus rectus (REC.L), precentral gyrus (PreCG.R), olfactory cortex (OLF.L), and middle occipital gyrus (MOG.R). These findings of the paper provide a new perspective for the clinical diagnosis of MCI.