| As we all know,the human brain is the most complicated and sophisticated system to human cognition in the world currently,therefore,brain science has become one of the important research areas.The study of brain function and structure,and how to decode brain function based on brain structure is a popular topic and a major challenge for human science as well.The rapid development ofMRI technology makes it possible to acquire image inside the human brain in a non-invasive way.MRI(Magnetic Resonance Imaging)and DTI(Diffusion Tensor Imaging)techniques have played an important role in the study of brain macrostructure and microstructure.Subsequently,in the 1990 s,the emergence of fMRI(functional Magnetic Resonance Imaging)provides a new window for researchers to observe the dynamic activities and activation state inside the brain.The mechanism of human brain function is closely related to human brain structure,therefore,researches on brain structure,function and their potential relationship based on MRI and fMRI have been receiving much attention.To explore and solve the related questions to some extent,this thesis conducted several corresponding researches in terms of brain function based on sparse coding technique as well as deep convolutional neural network(CNN)model.For the studies on brain structure,three related work was carried out in the paper to analyze cortical folding patterns.Finally,the thesis did a preliminary study on the potential relationship between brain function and brain structure in terms of functional role of gyri,sulci,2-hinge and 3-hinge gyral area from the perspective of neural activity.Specifically,the main points of the thesis can be summarized as follows:(1)This thesis is the first to propose the whole brain functional network representation based on temporal sparse coding.Specially,the brain volume at each time point of fMRI data was treated as a learning sample in temporal sparse coding technique to perform the decomposition of overlayed brain networks.The experimental result indicated that temporal sparse coding could effectively detect meaningful brain functional networks.Meanwhile,based on these detected networks,the thesis also made progress in the study on abnormal brain function of patients with brain disorders(ADHD: Attention Deficit Hyperactivity Disorder,m TBI: Mild Traumatic Brain Injury).For instance,more effective functional connectivity biomarkers for differentiating healthy controls and ADHD patients could be identified based on the brain functional networks detected by temporal sparse coding technique.In the investigation of m TBI,besides temporal sparse coding approach,the thesis also tried temporal concatenated spatial sparse coding to do a novel representation of brain states,the corresponding results not only indicated that group-wise statistical difference on the network composition of brain states could be found between healthy controls and m TBI patients at two different temporal stages(acute stage,sub-acute stage),but also revealed meaningful network interaction changes in m TBI.(2)This thesis proposed an effective 3D deep convolutional neural network(CNN)model that can characterize and differentiate Autism Spectrum Disorder(ASD)from healthy controls.In this model,it took meaningful spatial brain network overlap patterns as input.The corresponding experiment results demonstrated that the spatial distribution patterns of connectome-scale functional network maps per se have little discrimination power in differentiating ASD from controls via the CNN framework.In contrast,the spatial overlap patterns instead of spatial patterns per se among these connectome-scale networks learned via the same CNN framework have remarkable differentiation power in separating ASD from controls.In addition,in terms of algorithm,the thesis subsequently proposed a new framework—iteratively optimized convolutional neural networks(IO-CNN),this model successfully enables the functional brain networks recognition task to a fully automatic largescale classification procedure and replaces the traditional overlap rate calculation method by greatly improving the accuracy of functional network identification.(3)Starting from the morphometric features of the cerebral cortex folding pattern,in the study of brain structure,this thesis firstly proposed a novel computational framework which was used to study the morphometric patterns of major gyral crestlines and sulcal baselines(eight gyral crestlines,two sulcal baselines)in autism spectrum disorder(ASD).The experiment results indicated ASD patients show more curved patterns dominantly.Meanwhile,in combination with cortical thickness calculation results of corresponding gyral crestlines or sulcal baselines,the findings in the thesis coincide well with the findings in brain developmental studies that thinner cortex will be more curved.Moreover,to enable largescale cortical folding modeling and analyses,inspired by the observation that the lines of gyral crests can form a connected graph on each brain hemisphere,the thesis subsequently proposed a new representation of cortical folding organization — “gyral net”,and developed an automatic data processing and analysis pipeline which is capable of handling big datasets to extract 3-hinges gyral area.Finally,inspired by “axonal pushing” theory,the thesis proposed a hypothesis related to the relationship between denser growing fiber connections and 3-hinge gyral folding and for the first time investigated 3-hinge gyral area to study cortical folding mechanism which then verified its fundamental biomechanical mechanisms via a series of 3-dimensional finite element soft tissue models.By extracting 3-hinge gyral regions in macaque/chimpanzee/human brains,quantifying and comparing the relevant DTIderived fiber densities in 3-hinge and 2-hinge gyral regions,the phenomenon that DTIderived fiber densities in 3-hinge regions are much higher than those in 2-hinge regions could be consistently observed.Therefore,the thesis hypothesized that besides the cortical expansion,denser fiber connections can also induce the formation of 3-hinge gyri.Such integrative approach combining neuroimaging data analysis and computational modeling appeared effective in probing a plausible theory of 3-hinge gyri formation and providing new insights into structural and functional cortical architectures and their relationship.(4)From the perspective of neural activity,this thesis explored the intrinsic functional differences among gyri,sulci,2-hinge and 3-hinge joints and confirmed the different functional roles of neural activities on gyri and sulci,as well as 2-hinge and 3-hinge gyral folding areas.In particular,this study learned 1D convolution neural network(CNN)models with relatively high classification accuracies based on a mixed dataset including both healthy controls and ASD patients.The experiment results not only demonstrated the different functional roles of rsfMRI signals in different brain anatomical areas(gyri,sulci,2-hinge and 3-hinge gyral folding areas)but also observed that the functional difference of neural activities between gyri and sulci was more obvious than that between 2-hinge and 3-hinge gyral folding areas and such differences could be consistently found in the brains of both healthy controls and ASD patients.Besides,further analyses on learned characteristic features to differentiate gyral/sulcal,2-hinge/3-hinge joint rsfMRI(resting state functional Magnetic Resonance Imaging)signals were also designed and performed to interpret the corresponding findings. |
PDF Full Text Request