Accessing Latent Connectome of Mild Cognitive Impairment via Discriminant Structure Learning

For decades, many potential measures have been proposed and examined in terms of their predicting capability for mild cognitive impairment (MCI). The development of non-invasive markers from multiple imaging modalities including MRI (Tl-wighted), diffusion tensor imaging and functional MRI are of great interest. However, most of previous studies focused on classification, prediction or identification of statistical differences among different groups (Normal controls, MCI/AD) and clinical stages (longitudinal studies). It is still largely unknown whether there exists a way to quantitatively model the entire progression process of the disorder. Here we introduced a novel supervised discriminant structure learning method to explore latent structures of both structural and functional connectome of MCI patients. Our result shows that a latent structure reflecting the entire progression process of MCI can be learned from both structural and functional connectome. When considering more connectome features, the learned structure using functional data tend to display more heterogeneous patterns.