Integrating Multiple Network Properties for MCI Identification

Abstract

Recently, machine learning techniques have been actively applied to the identification of Alzheimer's disease (AD) and mild cognitive impairment (MCI). However, most of the existing methods focus on using only single network property, although combination of multiple network properties such as local connectivity and topological properties may be more powerful. Employing the kernel-based method, we propose a novel classification framework that attempts to integrate multiple network properties for improving the MCI classification. Specifically, two different types of kernel (i.e., vector-kernel and graph-kernel) extracted from multiple sub-networks are used to quantify two different yet complementary network properties. A multi-kernel learning technique is further adopted to fuse these heterogeneous kernels for MCI classification. Experimental results show that the proposed multiple-network-properties based method outperforms conventional single-network-property based methods.