Prediction Of Individuals’ ASR Based On Resting State FMRI

The Adult self-report(ASR)is a temperamental and psychological assessment scale for adults,which mainly includes evaluations of adult anxiety,attention,introversion,aggressive behavior,invasive behavior,etc.It has received widespread acceptance in the field of psychology and mental illness research.On account of the fact that the measurement of ASR score is still mainly dependent on the subjective scores of questionnaire,the objective measurement of ASR scores will be conducive to the accurate evaluation of adult mental and mental health conditions.Resting state functional magnetic resonance imaging(rs-fMRI)can effectively capture the subtle differences of individual brain function,and provide an effective way for the objective evaluation of individuals’ cognitive parameters.Based on the rs-fMRI data from 319 adult samples,the predictions of internalizing score,externalizing score and total score in ASR were carried out in this study by extracting brain functional network connectivity from different perspectives.The details of three tasks in this paper are as follows:(1)Prediction of ASR score based on classic functional connectivity.In this study,classic functional connectivity was used as the feature,least absolute shrinkage and selection operator(LASSO)algorithm was implemented as the feature selection approach,and random forest and support vector regression were introduced for predictive model construction of three kinds of ASR scores.The results showed that based on classic functional connectivity,the prediction accuracies of support vector regression and random forest were comparable,and the prediction of externalizing score(correlation coefficient between the predicted and the actual score R=0.496-0.504)was better than that of internalizing score and total score(R=0.335-0.395).The functional connectivity related to the frontoparietal network and the default-mode network was crucial for predicting individual ASR score.(2)Prediction of ASR based on the Granger causality.As the classic functional connectivity lack the consideration of driving relationship between brain regions(i.e.,the directional connectivity),Granger causality was introduced as the index of driving relations in this study.Granger causality was implemented as the feature and ASR score prediction was performed using the same algorithms of feature selection and predictive model construction in the previous study.The results showed that the predictions based on Granger causality were better than those based on classic functional connectivity,and the correlation coefficient between the predicted and the actual ASR externalizing score increased to R=0.506-0.526.The connections related to components of the default-mode network and the sensorimotor network occupied a significant proportion in prediction of three kinds of score.(3)Prediction of ASR based on pairwise likelihood ratio of the linear non-Gaussian acyclic model(LiNGAM).LiNGAM is a special kind of Bayesian network,and the pairwise likelihood ratio of it represents the causality or effect relationship between network nodes.The pairwise likelihood ratio of LiNGAM was introduced to extract the effective connectivity features of brain functional network,and the same algorithms of feature selection and predictive model construction as the first study were used for ASR score prediction.The results showed that ASR score prediction based on pairwise likelihood ratio was effective,and the correlation coefficient between the predicted and the actual score in ASR externalizing score prediction attained R=0.506-0.520.The connections associated with components of the default-mode network and the sensorimotor network were the predominant in the predictions of three kinds of score.The innovations of this study are as follow:(1)It was an early attempt for objective evaluation of ASR score based on rs-fMRI data in the field,and the predictions were observed to be of relatively high accuracy.(2)Granger causality was introduced as an index of directed connectivity between brain regions,and ASR score prediction was performed based on this kind of feature.(3)Regarding the brain functional network as a directed acyclic graph,the pairwise likelihood ratio of LiNGAM was introduced,and ASR score prediction based on the causality connectome of network nodes was carried out.This study not only performed effective object evaluation of ASR score based on rs-fMRI,but introduced new features based on rs-fMRI for individualized predictions of cognitive variables based on rs-fMRI.