Neuroimaging biomarker based prediction of Alzheimer'S disease severity with optimized graph construction

Abstract

The prediction of Alzheimer's disease (AD) severity is very important in AD diagnosis and patient care, especially for patients at early stage when clinical intervention is most effective and no irreversible damages have been formed to brains. To achieve accurate diagnosis of AD and identify the subjects who have higher risk to convert to AD, we proposed an AD severity prediction method based on the neuroimaging predictors evaluated by the region-wise atrophy patterns. The proposed method introduced a global cost function that encodes the empirical conversion rates for subjects at different progression stages from normal aging through mild cognitive impairment (MCI) to AD, based on the classic graph cut algorithm. Experimental results on ADNI baseline dataset of 758 subjects validated the efficacy of the proposed method.