Functional Connectome-based Location And Classification Of Epilepsy-related Networks In BECTS

Objective:The aim of this study is to investigate the reliability of interictal epileptiform-related brain activation reflected by granger causality density in BECTS,and the classification performance of granger causality density(GCD)based support vector machine(SVM)in distinguishing the children with benign epilepsy with centrotemporal spikes(BECTS)with interictal epileptiform discharges(IEDs)from those without IEDs(non-IEDs),as well as the classification performance of resting-state dynamic causal modeling(rs-DCM)based SVM in distinguishing the BECTS from the healthy children and the potential characteristic network models.Methods:Generalized linear model was used to observe the IEDs-related brain activations.The differences in brain areas were used for structural equation parametric modeling(SEM)and dynamic causal modeling(DCM)to determine optimal directed network models.GCD was calculated on four metrics,including inflow,outflow,total-flow(inflow+outflow)and int-flow(inflow-outflow)connectivity.Multiple correlation-modality analyses(pearson,multiple regression,across-voxel and across-subject correlations),IEDs-related brain activation map and lateralization of localize IEDs were used to explore the reliability of the GCD analysis in discribing the IEDs-related brain activations.GCD,SEM and DCM were used to determine optimal directed epilepsy-related network models.SVM was used to evaluate the classification performance of the GCD-SVM and rs-DCM-SVM in distinguishing BECTS with and without IEDs,and BECTS from healthy children.Results:Multiple correlation-modality analyses,IEDs-related brain activation map and clinical EEG recording exhibited good couplings between GCD maps and EEG-fMRI map.which can be used to discribe the IEDs-related brain activations.GCD has high discriminating brain regions between IEDs and non-IEDs subgroups mainly in the Rolandic area,caudate,dorsal attention network,visual cortex,language networks,and cerebellum.Combinations with more than three GCD metrics could receive good classification performance(best value;AUC,0.928;accuracy rate,90.83%;sensitivity,90%;specificity,95%).Specially,the combination of inflow,outflow and int-flow connectivity received the best classification performance among all combinations with the best AUC,classification accuracy and specificity,as well as stable classification performance.GCD analysis exhibited that the upper portions of either pre-or post-central gyrus mainly received information flow while the Rolandic areas mainly sent out.The IEDs promoted the targeted and driving effect.From non-IEDs to IEDs status,the thalamus load may play important roles in the modulation and regulation of the directed information flow of epileptiform-related brain networks.The reliability of the SEM and DCM methods of is proved by multicentric normal children showing the same optimal network architecture,which does not change with the scanning environment and number of sample sizes.The optimal network architecture of the two BECTS subgroups showed that the information flow arises from the right Rolandic areas to the left Rolandic areas,and subsequently propagates to the thalamus and the upper portions of either pre-or post-central gyrus.The causal effective connectivity from the upper portions of either pre-or post-central gyrus to the thalamus was only shown in the IEDs subgroup.The optimal network architecture of the two BECTS subgroups both differed from healthy children.The rs-DCM-SVM classifier received high classification performance for distinguishing the BECTS from the healthy children with classification accuracy of 88.16%%and AUC value of 0.92.External data also received the similar classification performance(classification accuracy of 81.51%and AUC value of 0.83).Total of 108 positive and 125 negative effective connectivity edges were used as useful features for classification.The left thalamus,left language-related brain region,left precentral gyrus,left inferior parietal cortex,right Rolandic areas,bilateral intraparietal sulcus,right striatum and right inferior parietal cortex received the highest weight of contribution to the classification.These areas located in the sensorimotor,subcortical,dorsal attention,and language networks.Conclusions:The GCD might be served as a potential biological indicator to discribe the changes of the IEDs-related brain networks.The GCD-SVM and rs-DCM-SVM might be promising models for BECTS early diagnosis and prediction of IEDs degree.The high excitatory and high inhibitory connectivity among the epileptiform-related brain networks may be an important neural network basis for mediating epileptogenesis.These findings will shed new light on the pathophysiological mechanisms of BECTS development and progression.