Consistency Of The Representations Of The Static Characteristics Of Brain Function Based On Naturalistic FMRI Across Different Stimuli

Naturalistic functional magnetic resonance imaging(fMRI)is attractive not only for the intrinsic advantages of fMRI technology,but for the approximation of naturalistic paradigm to daily life.Many studies on brain function utilized such representations of the static characteristics of brain function as functional connectivity(FC),inter-individual correlation(ISC)and inter-individual functional correlation(ISFC)extracted from naturalistic fMRI based on full time courses(as opposed to representations of the dynamic characteristics based on partial time courses).To date,however,little is known about the consistency of these representations,especially their consistency across different naturalistic stimuli.Cross-stimuli consistency of the representations of brain function deserves in-depth studies,as this consistency directly affects the generalizability of predictive models based on them.Besides,most of the existing representations of the static characteristics of brain function focus on the synchrony of fMRI signals,while the brain activation induced by naturalistic stimuli also contains much information.However,given the complexity and variability of naturalistic stimuli,there are no studies on the detection of brain activation based on naturalistic fMRI.To address these issues,this study carried out the following three research works:(1)Cross-stimuli consistency of natural-state FC.In this work,FC was extracted from four different sets of naturalistic fMRI(with movie as stimuli).The cross-stimuli consistency of natural-state FC in exact values was evaluated by ICC,and that in spatial patterns was evaluated by individual identification.Moreover,with sex classification and grip strength prediction as test cases,the generalizability of predictive models based on natural-state FC was evaluated.The results showed that the cross-stimuli consistency of natural-state FC was good in both exact values and spatial patterns.And predictive models trained based on FC extracted from one naturalistic fMRI run can make good predictions on FC extracted from runs with different stimuli,which indicates the predictive models were generalizable.(2)Cross-stimuli consistency of ISFCs.In this work,ISFCs were extracted from four different sets of naturalistic fMRI(with movie as stimuli),and the intensity of the whole-brain ISFCs was evaluated.And the cross-stimuli consistency of ISFCs were analyzed in both exact values and spatial patterns.In addition,the generalizability of predictive models based on ISFCs was assessed.The results showed that the overall intensity of the whole-brain ISFCs was weak,and the consistency of the ISFCs across different naturalistic stimuli was low in both exact values and spatial patterns.Moreover,the prediction accuracies with the same naturalistic stimulus for model training and testing were significantly higher than those with different naturalistic stimuli,indicating that the predictive models based on ISFCs were of poor generalizability.(3)Cross-stimuli consistency of NAs.This study utilized common signals among subjects to construct response models,and NAs were obtained with general linear model(GLM).The cross-stimuli consistency of NAs was analyzed in both exact values and spatial patterns,and the generalizability of predictive models based on NAs was evaluated.The results showed that the intensity of NAs was strong in general,and the cross-stimuli consistency of NAs was high in both exact values and spatial patterns.Moreover,the predictive models based on NAs were generalizable.The innovations of this study are as follows:(1)The cross-stimuli consistency of natural-state FC was verified and the generalizability of predictive models based on natural-state FC was evaluated to make a valuable exploration for the practical application of natural-state FC.(2)The cross-stimuli consistency of ISFCs was found to be low,which warns that the ISFC-based conclusions may only be applicable to the single given stimulus.(3)A method was proposed to detect brain activation under naturalistic stimuli(NA).The cross-stimuli consistency of NAs was analyzed and it was found that the consistency of NAs across different stimuli was great,suggesting that NA could be widely used in the data analysis based on naturalistic fMRI.