The new graph kernels on connectivity networks for identification of MCI

4th International Workshop on Machine Learning and Interpretation in Neuroimaging, MLINI 2014 and workshop on Neural Information Processing Systems, NIPS 2014

Montreal QC

2016

Brain connectivity networks have been applied recently to brain disease diagnosis and classification. Especially for both functional and structural connectivity interaction, graph theoretical analysis provided a new measure for human brain organization in vivo, with one fundamental challenge that is how to define the similarity between a pair of graphs. As one kind of similarity measure for graphs, graph kernels have been widely studied and applied in the literature. However, few works exploit to construct graph kernels for brain connectivity networks, where each node corresponds a unique EEG electrode or regions of interest(ROI). Accordingly, in this paper, we construct a new graph kernel for brain connectivity networks, which takes into account the inherent characteristic of nodes and captures the local topological properties of brain connectivity networks. To validate our method, we have performed extensive evaluation on a real mild cognitive impairment (MCI) dataset with the baseline functional magnetic resonance imaging (fMRI) data from Alzheimers disease Neuroimaging Initiative (ADNI) database. Our experimental results demonstrate the efficacy of the proposed method.