Combining Multiple Network Features for Mild Cognitive Impairment Classification

Abstract

Connectivity-network-based techniques have been recently developed for the diagnosis of Alzheimer's disease (AD) as well as its prodromal stage, i.e., mild cognitive impairment (MCI). However, most existing methods focus on using only a single property of connectivity networks (e.g., The correlation between paired brain regions), which can not fully reflect the topological information among multiple brain regions. To address that problem, in this paper we propose a novel connectivity-network-based framework to fuse multiple properties of network features for MCI classification. Specifically, two different types of network features (i.e., Brain region and sub graph) are respectively used to quantify two different properties of networks, where two kinds of feature selection methods are further performed to remove the irrelevant and redundant features. Then, multi-kernel learning technique is adopted on those corresponding selected features to obtain the final classification results. We evaluate our proposed method on a real MCI dataset containing 12 MCI patients and 25 healthy controls. The experimental results show that by using multiple properties of network features our method achieves better performance than traditional methods using only single property of network features.