Research On FMRI Decoding And Applications Based On MVPA

After millions of years of evolution,the human brain is able to handle complex infor-mation in conditions of constantly changing environment on account of a well-organized neural network which formed with abundant neurons and synapses.The appearance of modern brain imaging technologies,especially the functional magnetic resonance imaging(fMRI)has made it possible to observe the structure and function of the brain because of its non-invasive,non-radioactive and high spatial resolution features.As an advanta-geous tool for exploring the brain functions and revealing its working mechanism,fMRI has greatly promoted the progress of brain science research.Due to the high spatial resolution and specificity of imaging mode,fMRI data has the characteristics of high dimensionality and low signal-to-noise ratio.Traditionally,the fMRI data have been analyzed with a mass-univariate general linear model approach to reveal task-related brain areas by treating each voxel separately.One of the limitations about this approach is that the interrelationship among voxels of spatially distributed brain areas is not considered because it works on isolated voxels,hence,the joint infor-mation among them is ignored.As an alternative method,multivariate pattern analysis(MVPA)approach has been widely applied in neuroimaging data analysis and presented promising prospect in the analysis of fMRI data for their superior performance in localiz-ing spatial patterns of activity that differentiated across experimental conditions/tasks.In this dissertation,task state fMRI data is taken as our research object and MVPA approach is chosen as research method to decode the cognitive state of the human brain and localize brain activation areas corresponding to experimental conditions/tasks.The main contents and contributions of the dissertation can be summarized as follows:1.On the basis of previous method,an orthogonal matching pursuit(OMP)based feature selection method was proposed to solve the high dimensional problem of fMRI data.The method was then applied in the competition of fMRI data analysis.The effectiveness of this new method was verified by comparing with other two methods subsequently.2.On the basis of the former chapter and the functional specialization principle of fMRI data analysis,a method that combine a forward feature selection scheme with a sparse regularization and permutation testing for feature selection in multivariate pattern classification settings was introduced.The effectiveness of this novel approach in potential applications was demonstrated via the data analysis results.3.MVPA method based on the searchlight.In this fMRI study,we used familiar or unfamiliar face images,auditory names,and audiovisual face-name pairs as stimuli to determine the influence of semantic familiarity on audiovisual integration.Specifically,using the fMRI data,we decoded familiarity categories of the stimuli(familiar vs.un-familiar)and calculated the reproducibility indices of brain patterns corresponding to familiar and unfamiliar stimuli.The results indicate that audiovisual integration at the semantic level enhanced and enriched the neural representations of semantically famil-iar congruent face-name pairs but not semantically familiar incongruent or unfamiliar pairs.Furthermore,this modulatory effect of semantic familiarity for congruent stimuli on audiovisual integration might have arisen from enhanced effective connectivity that influences information flow from the heteromodal bilateral perirhinal cortex to the brain areas encoding familiarity.4.Inspired by the previous chapter that the influence of semantic familiarity on the effective network connection,we conducted an fMRI experiment in which subjects were presented with face-name pairs that were semantically familiar or unfamiliar and instructed to judge the familiarity of the stimulus(familiar vs.unfamiliar),and esti-mated the whole-brain scale effective connectivity networks on acquired fMRI data.We used an MVPA approach based on effective connectivities to reliable decode the semantic familiarity face-name pairs with high classification accuracy,and identified the features with high discriminative power at the same time,which help us provide insights in brain network of face-name pairs recognition.Furthermore,the brain regions with high discrim-ination ability were also confirmed by the previous studies and the comparison between the two task-specific networks is also very persuasive.In particular,there is a high de-gree of consistency with the viewpoint of information flow and the nature of effective connectivity.