Research On Dynamic Characteristics Of Functional Brain Networks Via MRI Data

Brain science is one of the key research fileds in today’s world. On the road of investigating the complicated structure and function of our brain, magnetic resonance imaging(MRI) techniques have contributed a lot, e.g. Diffusion tensor imaging(DTI) data can provide structure information of white matter(WM) fibers in vivo and functional MRI(f MRI) technique can be used to investigate the function of brain. The brain network study and functional connectivity analysis based on MRI data is one important aspect of brain science research.During perception and cognition, the brain undergoes a set of information processing steps such as information perception, transfer, coordination, transform, storage and new information creation. The functional activity state of the brain will change accordingly. During the dynamic change procedure, whether the brain has kinds of temporally stable states called brain metastate? How can the dynamic brain metastates be accurately described and represented? During the procedure of brain completing cognitive tasks, how did the functional information interaction dynamically change between brain regions? The investigations to these questions have important meanings to understand thr brain’s functional congnition. Foucused on these hot issues, this thesis ultilized the brain’s DTI and f MRI data, based on the whole-brain functional connectivity network, did research on the dynamics of brain functional state from the following three aspects: 1) the expression of dynamic brain common state patterns; 2) the dynamic characteristic of brain resting state networks(RSN); 3) the tracing of dynamic information transfer path. The major research contents and innovation were summarized as follows:(1) This theis introduced one whole-brain network reference system with high accuracy and individual correspondence, i.e. DICCCOL(Dense Individualized and Common Connectivity-based Cortical Landmarks). It utilized the structural information of WM fibers provided by DTI data, which were expressed and compared through one kind of mathematical model, and then the globally optimal node group was searched out which has the least group structural variance between their corresponding fiber boundles. Finally, total 358 consistent spatially distributed cortical landmarks were acquired. These landmarks have high structural and functional correspondence, and are reproducible and predictable among individual brains. Besides, they can effectively represent the brain’s common structural connection patterns andmain functional brain areas. It is helpful for doing analysis and comparsion of the brain structure and function between different subjects or groups.(2) Based on DICCCOL whole-brain dynamic functional connectivity, this thesis discovered and proposed one specific expression of brain metastates. Through mapping f MRI data to the corresponding DTI space, the f MRI time series of each DICCCOL node can be extracted. A sliding time window based approach was applied to compute dynamic functional connectivity strength between each pair of DICCCOLs, and subsequently the corresponding 2D dynamic functional connectivity strength matrix was obtained by computing the connectivity strength of each DICCCOL. It is observed that this kind of matrix keeps relative stable during a short time period, i.e. brain metastates. Thus, the brain dynamic states could be described by the time-varying brain metastates based on DICCCOL whole-brain functional connectivity matrix. It makes preparation for studying the brain’s dynamic information processing and brain state variation in future.(3) Based on the sparse representation and classification of brain metastates samples, this theis presented one effective approach to represent dynamic common brain state pattern space. First,the metastates were represented by WQCP(whole-brain quasi-stable connectome pattern) samples, and then all the WQCP samples were collected as the input of FDDL(Fisher discriminative dictionary learning) for training and classification, and finally the dynamic brain states can be represented by a set of whole-brain brain functional connectome patterns, i.e. dynamic common brain state pattern space. Through the sparse representation of resting state and task brain WQCP samples, we found that the common state patterns during rest are completely different from that of task brains. The task brains with outliers, i.e. showing resting state ACP patterns, were further examined by activation detection experiments, and the result indicated that those subjects didn’t follow the designed paradigm well during the task scan time. It is demonstrated that our approach can provide effective assistance for f MRI data quality control.(4) Based on the resting state dynamic common brain state patterns, this thesis investigated the dynamic characteristic of resting state networks. Considering each dynamic common brain state pattern as one single view of whole-brain functional connectome, DICCCOL landmarks were clustered into several sub-networks containing dynamic information, i.e. dynamic RSNs, through multi-view spectral clustering. After doing comparison with static DICCCOL RSNs,we found that some RSNs including default mode network(DMN) and visual RSN show great stability during rest, while another RSNs including motor-related RSNs exhibit strong dynamics. It implied that those dynamic networks play a key role in dynamic functional brain interaction at rest. This study provides a new insight to RSN research and investigating the resting state brain’s functional information processing mechanism.(5) Based on the brain metastate time series and expression of dynamic common brain state patterns, this thesis tracked the information flow path under each metastate, i.e. the formation of brain metastate is considered to result from the dynamic information exchange among different brain regions. First, under each metastate, DICCCOL landmarks were clustered into several sub-networks and the average signal of each sub-nework was fitted and the corresponding activation time was detected. The sub-networks were re-oredered according to the sequence of their activation times. Then,one kind of probabilistic model of information flow was built and the dynamic programing was performed to achieve the most optimized information transfer path with the highest probability. The optimization path was represented by a sequence of key router nodes from each sub-network. Experimental results showed that the distributions of key router nodes with high frequency on normal adolescents and PTSD(post-traumatic stress disorder) adolescents during visual task had obvious variance and differences. The normal group tends to have high frequency router nodes in visual cortex, while the node frequency of the corresponding areas of the PTSD group was relatively lower. Besides, PTSD brains’ functional activity involved more cortex regions than normal brains. This study will give valuable reference to understand the functional information processing mechanism of PTSD brains.