Research On Brain Functional Connectome Based On Magnetic Resonance Imaging

Human brain is the most known complex network system in the world, and its functional architecture and operational principles adhere to two principles:functional segregation and functional integration. Functional segregation suggests that the whole brain network can be functionally divided into many sub-networks. The mutual coordination and collaboration between these sub-networks are implied by the functional integration.Functional connectome has been studied from different angles, for the connectomic-level analysis of the pathology and physiology of brain disorders. The main contents of the dissertation have been put forward:Firstly, we introduced the method for constructing large-scale whole-brain functional connectome based on the DICCCOL framework. In the DICCCOL framework, there are 358 structural and functional consistent brain regions of interest (ROIs), as network nodes. With functional connectivity, which is measured by the correlations between the blood oxygenation level dependent (BOLD) signals corresponding to those brain ROIs, as network edges, functional connectome is then derived. Meanwhile, we also elucidated the quantitative and qualitative methods for analyzing functional connectome.Then, given the complex and composite structure of functional connectome, we proposed a methodological framework, which was composed of the functional connectomic construction based on the DICCCOL framework and the atomic functional connectome (AFC) decomposition based on the divergence-based projective non-negative matrix factorization (DPNMF) algorithm, for the static AFC analysis. The applicability and effectiveness of the methodological framework were demonstrated by applying it to two mild cognitive impairment (MCI) datasets, which included both MCI subjects and their matched healthy control (HC) subjects, for their static AFC analysis.91% of the MCI subjects and 89.8% of the HC subjects were successfully classified.Additionally, given that the brain’s functional state is under dynamical changes but cannot be measured directly, we took functional connectome as the indirect measurement of the brain’s functional state, and then trained hidden Markov models (HMM) to analyze the hidden dynamic transition patterns of the brain’s functional states. The computational framework was then applied to the subjects with post-traumatic stress disorder (PTSD) and their matched HC subjects.91% of the PTSD subjects and 95% of the HC subjects were successfully classified via majority voting.Finally, considering the effectiveness of the AFC approach and the dynamics of functional connectomes, we proposed a framework for dynamic AFC analysis, which was composed of the DICCCOL framework, Bayesian connectivity change point model (BCCPM) and the DPNMF algorithm. This framework was then applied to the subjects with attention deficit/hyperactivity disorder (ADHD) and their matched HC subjects.92% of the ADHD subjects and 95% of the HC subjects were successfully classified. As compared with the static AFC analysis methodology, the dynamic AFC analysis framework can reveal more similarities and differences between the AFCs with disorder and the ones without disorder.