Exploring latent structures of Alzheimer's disease via structure learning

As a non-invasive brain imaging approach, functional magnetic resonance imaging (fMRI) acts an irreplaceable role when studying whole brain functional activities and dynamics. Large-scale and collaborative effort in collecting fMRI data, especially in neurology and psychiatry, offers a new source of statistical power to deepen the understanding of brain disease. However, due to the heterogeneous and dynamic nature of fMRI BOLD signals, we still lack a fundamental model to effectively identify and integrate the intrinsic functional characteristics at population level. Here we introduced a novel supervised structure learning method to explore latent structures of resting state fMRI (rs-fMRI) data that belongs to multiple groups. Using the dataset from Alzheimer's Disease Neuroimaging Initiative (ADNI) as a testbed, we successfully identified a 'TREE' structure in both whole brain and ROI based analysis, which reflects a virtual 'path' of Alzheimer's Disease (AD) progression as multiple stages.