Research On Functional Model Of Cortical Cortex Via Functional Magnetic Resonance Imaging

The brain,as the biological basis for abstract thinking,is the most advanced hub of the central nervous system of human beings.Notably,the cerebral cortex has very complicated folding patterns,consisting of various gyri and sulci,which play essential roles in the functional organization of brain.Therefore,research on cortical folding is the spotlight of the neuroscience field.However,it lacks systematical researches on the functional role differences between sulci and gyri.Studying the functional characteristics of the cerebral cortex will improve our understanding on the working mechanism of the human brain,and contribute to imaging marker identification of various neurological diseases.Functional magnetic resonance imaging(f MRI)technology can probe the neural activities with reasonable spatiotemporal resolution,making it possible to study the relationship between cortical folding patterns and functional architecture of cerebral cortex systematically.Based on f MRI data acquired under different paradigms,this dissertation explored the functional network connectivities of the cerebral cortex and their graph theory-based properties,as well as the frequency-specific characteristics of cortical f MRI signals using different algorithms,such as the sparse coding algorithm and deep neural networks.The research contents,innovations and related findings of this dissertation are summarized as follows:1.This dissertation proposed two novel sparse coding-based algorithms for f MRI data modelling and functional network identification:·Hybrid spatiotemporal sparse coding algorithm.The algorithm decomposes the individual f MRI matrix into a group-wise temporal dictionary,an individual-specific coefficient matrix,and a group-wise spatial dictionary.This algorithm can effectively reduce the dimensionality of f MRI data,discover discriminative patterns of every f MRI session,and robustly identify meaningful functional networks.Subsequent experiments of task state classification reached 100% accuracy.·Two-stage sparse coding algorithm.The algorithm establishes the spatial correspondence between the functional networks of different subjects,by imposing group-wise spatial constraints on the individual dictionary learning stage.Based on the temporal consistency of dictionary matrix and the spatial consistency of sparse coefficient matrix,this algorithm can automatically divide all brain networks into task-related networks,intrinsic connectivity networks and uncertain networks in an unsupervised manner.2.Based on above algorithms,this dissertation carried out research on functional organization of human brain by analyzing graph theory-based characteristics of functional network connectivities and cortical f MRI representation accuracy:·This experiment combined the sparse coding algorithm and complex network analysis to explore the graph theory-based characteristics of functional network connectivities(FNC)of cerebral gyrus and sulcus.The algorithms were applied to resting state and task-based f MRI data released by the human connectome project(HCP).The experiments revealed that gyral FNC has the highest proportion of strong connections,while sulcal FNC has the lowest proportion of strong connections,and the gyrus-sulcus FNC has the moderate proportion of strong connections.More importantly,gyral FNC has significantly higher global efficiency,higher local efficiency,and lower modularity than sulcal FNC.These graph theory-based differences indicate that gyri are the global functional hubs of the brain for information exchange,while sulci are local information processing units.·In order to characterize the degree of information sharing on cortical surface,this dissertation explored the f MRI representation accuracy in each cortical region.Large-scale analysis were conducted based on HCP emotion task-based f MRI data.One important finding is that in the four regions of interest in the default mode network,the signal representation accuracy of the gyri is significantly higher than that of the sulci.Similar findings were found in other typical resting-state networks.Considering that the dictionary matrix reflects the representative neural activity patterns of f MRI data,the signal has higher signal representation accuracy indicates that it contains more shared information with representative neural activities.Therefore,above results indicate that gyri may be more involved in different functional network interactions.3.This dissertation designed shallow and deep convolutional neural networks(CNN)to unearth signal composition patterns of f MRI data and characterize the discriminability and time-frequency characteristics of cortical f MRI signals:·This dissertation first designed a shallow CNN model to classify the functional signals of gyrus and sulcus to explore the differences in the spectral characteristics of neural activities on cortical surface.The advantage of this model lies in that it has easy interpretablity.The CNN distinguished the gyral and sulcal f MRI signals with moderate accuracy.More importantly,compared with gyral f MRI signals,the convolution kernels related to sucal f MRI signals show more diverse and higher-frequency patterns.In addition,the wavelet entropy metric further verifies the differences between gyri and sulci.The above findings prove that gyri and sulci play fundamentally different roles in the cerebral cortex.·This dissertation further designed deeper CNN models to improve the accuracy of f MRI signal classification and explore hierarchical features of the CNN models.This work is complementary to the above study.The model achieved higher classification accuracy in both human and macaque f MRI data.In addition,gyral f MRI signals have simple and low-frequency signal components,while sulcal f MRI signals have complex and high frequency signal components.4.In comparison with resting-state and task-based paradigms,the naturalistic paradigms can imitate various daily real-life scenarios,providing a new opportunity for the study of the functional division of the cerebral cortex.To this end,this dissertation initially explored the neural mechanisms of cerebral cortex in the naturalistic paradigm based on the inter-observer visual congruency(IOVC)metric:·This dissertation used the general linear model and inter-subject correlation analysis to examine the distribution of IOVC-related brain areas on cortical surfaces.The results show that cortical areas with positive correlation with IOVC are mainly located in the gyral regions,while cortical areas with negative correlation with IOVC are mainly located in the sulcal regions.In addition,when the subjects watch high-IOVC movie stimuli,their brain activities show significantly larger synchronization,which are also mainly located in the gyral regions.