Research On Individual Differences In IQ Based On Dynamic Function Network Analyses Of FMRI Data

As an important physiological parameter for individual cognition and the judgment of some neurological disease,intelligence quotient(IQ)has been the research focus in the field of cognitive neuroscience.Resting state functional magnetic resonance imaging(fMRI)is one important technique for studying human brain function.The analysis of the dynamic properties of brain functional network is a hot topic in functional brain research.However,most studies on individual differences in IQ still based on the assumption of temporal stationarity,ignoring the dynamic changes of functional connections among different brain regions.This study investigated the relationship between dynamic functional networks and IQ based on resting state fMRI.The study was performed on the resting state fMRI data of 97 healthy children.The main contents of this paper are as follows:(1)Research on the relationship between dynamic properties of brain functional network and IQ.This study analyzed the relationship between dynamic properties of brain function network and IQ from two perspectives,namely,dynamic switches between functional network states,and temporal fluctuations in functional connection time-series.The results showed that the dwell time of some functional network states were correlated with IQ.In addition,the significant correlations between the dynamic fluctuations of dynamic functional connection among different brain regions and IQ were observed,which provided an important basis for the follow-up research on IQ prediction based on the dynamic functional connection features.(2)IQ Prediction based on dynamic functional connection(DFC)features.The prediction model was extended from the perspective of features and algorithms,four DFC features in time domain and frequency domain were extracted,and multiple regression algorithms were used to predict children's IQ.The experimental results show that partial dynamic characteristics(DFC_Mean feature and FFT_Feature of specific frequency band(0.075-0.1 Hz))can predict IQ well.Moreover,the best result(correlation between the estimated and the actual value R=0.54)was obtained by the combination of minimum angle regression algorithm and frequency domain features,which exceeded the previous IQ prediction results based on the dynamic functional connection feature(R=0.42).(3)IQ prediction based on multi-view learning.Since both dynamic and static features contribute to IQ prediction,multi-view learning can be used to integrate complementary information among different features.We take the lead in introducing multi-view learning methods to carry out individual IQ prediction research.The results showed that the multi-view learning method based on static and dynamic features of functional connections fail to exceed the single-view prediction results.The reason may be that static and dynamic features are based on fMRI data extraction,perhaps because both static and dynamic features are extracted based on fMRI data,the complementarity between features is limited.The innovations of this study are as follow:(1)studied the individual differences of IQ from the perspective of dynamic functional connection;(2)in addition to fully analyzing the prediction effect of multiple time domain features based on dynamic functional connections,the frequency domain features were introduced to carry out IQ prediction research;(3)introduced the minimum angle regression algorithm which suitable for high-dimensional and small sample data,and obtained better prediction results;(4)introduced multi-view learning method to provide new ideas for individual IQ evaluation research from the perspective of feature fusion.