PROJECT TITLE :

Multiview Feature Learning With Multiatlas-Based Functional Connectivity Networks for MCI Diagnosis

ABSTRACT:

For the diagnosis of Alzheimer's disease and its prodromal stage, also known as mild cognitive impairment, functional connectivity (FC) networks that were built from resting-state functional magnetic resonance imaging (rs-fMRI) have shown promising results (MCI). FC is typically calculated as a temporal correlation of regional mean rs-fMRI signals between any pair of brain regions. These brain regions are customarily parcellated according to a specific brain atlas. The majority of currently conducted research has utilized a predetermined brain atlas for all of the subjects. Nevertheless, it is unavoidable for the constructed FC networks to ignore potentially significant subject-specific information, in particular the subject-specific brain parcellation. FC networks constructed on the basis of a single atlas may not be sufficient to reveal the underlying complicated differences between normal controls and disease-affected patients due to the potential bias from that particular atlas. This is analogous to the drawback of the "single view" (as opposed to the "multiview" learning) in medical image-based classification. In this case, the "single view" learning is preferred. In the current investigation, we propose a multiview feature learning method with multiatlas-based FC networks as a means of enhancing MCI diagnosis. To be more specific, a three-step transformation is implemented in order to generate multiple individually specified atlases from the standard automated anatomical labeling template. From these atlases, a set of atlas exemplars is then selected. Multiple FC networks are built using these preselected atlas exemplars as the basis for their construction. These multiple FC networks provide multiple perspectives on the FC network-based feature representations for each subject. After that, we develop an algorithm for multitask learning in order to select joint features from multiple FC networks that have been constructed. In order to make a multiatlas-based MCI diagnosis, the selected features are combined before being fed into a support vector machine classifier. Extensive experimental comparisons are carried out between the method that has been proposed and other competing approaches, such as the conventional method that is based on a single atlas. The findings suggest that our approach results in a significant improvement to the MCI classification, demonstrating the method's potential application in the brain connectome-based individualized diagnosis of brain diseases.