Self-Weighting Grading Biomarker Based on Graph-Guided Information Propagation for the Prediction of Mild Cognitive Impairment Conversion

Abstract

Mild cognitive impairment (MCI) represents a transitional stage between normal aging and Alzheimer's disease (AD), with a higher risk to convert to AD. The information of AD and normal control (NC) subjects can aid the classification between progressive MCI and stable MCI. In this paper, we develop an effective biomarker by combining the auxiliary information of AD and NC subjects with the relationship of brain regions of MCI subject, which makes best of auxiliary information and improves the prediction accuracy of MCI-to-AD conversion. Specifically, a projection vector is first obtained for each MCI subject via graph-guided information propagation. Next, the information of projection vector is integrated using a self-weighting grading method to acquire the novel biomarker. Finally, the self-weighting grading biomarkers derived from multiple morphological features are combined to provide more accurate prediction of MCI-to-AD conversion. Experimental results on the Alzheimer's Disease Neuroimaging Initiative database demonstrate the effectiveness of the proposed biomarkers for the prediction of MCI conversion.