A plethora of neuroscience studies has found that Alzheimer?s disease (AD) can be understood as a dysfunction syndrome where the structural and functional connectivity of the large-scale network are progressively disrupted by molecular pathomechanism that is not fully understood. The disruptions to the network exhibit dynamic patterns at different stages of AD, which holds valuable clues to understand AD progression. Current network computational tools are designed for cross-sectional data only, which is insufficient to maintain temporal consistency in investigating longitudinal network changes. To address this problem, we will develop the first extensive computational tool for longitudinal network analysis. Specifically, we will propose a learning-based approach to precisely quantify the evolution of brain network from noisy imaging data (Aim 1). Sparse representation and tensor analysis technique will be integrated to seek for the consistent longitudinal brain networks. We will apply our longitudinal network analysis tool to the series diffusion-weighted imaging (DWI) data from ADNI database to investigate how Alzheimer?s disease attacks human brain network by inspecting the dynamic interactions between the hub and non-hub nodes in AD progression (Aim 2). The outcome of this project will be the first longitudinal brain network analysis tool in computational neuroscience and neuroimaging fields. We will release the software (both binary program and source code), to facilitate the network studies in other neuro diseases that show brain network dysfunction.
This proposal aims to develop a set of efficient computational tools specifically for longitudinal brain network analysis. In order to accurately measure the connectivity strength, we propose a learning-based approach to find the intrinsic brain network from the noisy imaging data. Sparse representation and tensor analysis technique have been integrated into brain network optimization for preserving the longitudinal consistency. It is worth noting that our data-driven approach is free of the ad-hoc thresholding technique which may undermine the repeatability of discovery. To the best of our knowledge, our learning-based approach is the first extensive computational tool for longitudinal network analysis in neuroscience and neuroimaging field. We will apply our longitudinal network analysis tools to the series diffusion-weighted imaging (DWI) data from ADNI database to investigate how Alzheimer?s disease attacks human brain network by inspecting the dynamic interactions between hub and non-hub nodes in AD progression. At the end of this project, we will release the software (both binary program and source code), to facilitate the network studies in other neuro diseases which are related to brain network dysfunction.
