EEG-fMRI Information Fusion And Its Application

With development of neuroimaging techniques,many noninvasive recordings have shown respective strengths and weaknesses in understanding the brain functions and dysfunctions.And,integrating information of various neuroimaging multimodalities has potential to achieve high spatiotemporal mapping of brain noninvasively.Typically,in view of noninvasiveness and complementarity of spatiotemporal resolution,electroencephalogram(EEG)and functional magnetic resonance imaging(fMRI)multimodal fusion has been widely applied in cognitive and clinical neurosciences.The main work in this dissertation concentrates in EEG-fMRI fusion based on spatial and/or temporal information and its application in epilepsy.We start with an introduction of a novel spatiotemporal consistency measure and its application in frontal lobe epilepsy.Then,two kinds of multimodal information fusion methods are developed;and a proposed method,named eigenspace maximal information canonical correlation analysis,is applied in the study of juvenile myoclonic epilepsy(JME).Finally,based on spatiotemporal matching between EEG and fMRI modalities,a hierarchical framework is proposed to ensure the relative reliability of the integrated results.The main works and findings in this dissertation are introduced as follows:1.A novel measure,named four-dimensional(spatiotemporal)consistency of local neural activities(FOCA),is proposed to characterize the spatiotemporal information of local spontaneous activity.The FOCA measure integrates both temporal homogeneity of local adjacent voxels,and regional stability(i.e.spatial consistency)of brain activity states between neighboring time points.Using simulation,resting-state and task fMRI data,the performance of FOCA is assessed first.Then,the FOCA measure is applied in the study of frontal lobe epilepsy(FLE).These results suggest that FOCA may provide important spatiotemporal information that will help in the understanding of the brain function and dysfunction.2.Based on canonical correlation analysis(CCA),we propose a new multimodal information fusion method,named local multimodal serial analysis(LMSA).This method emphasizes the common temporal information of EEG and fMRI modalities to decrease the uncertainty in multimodal fusion,and it can tolerate variably shaped hemodynamic response functions(HRFs)in the local region.A simulation shows its superiority for detecting weak changes of fMRI blood oxygenation level-dependent(BOLD)signals related with EEG features.Furthermore,using simultaneous EEG-fMRI data of familial cortical myoclonic tremor and epilepsy(FCMTE)patients,LMSA also performs well in revealing the potential BOLD changes related with epileptic discharges.These results show the superior performance of LMSA in comparison with the traditional EEG-informed fMRI analysis(general linear model),and suggest that LMSA has the potential for applications in simultaneous EEG-fMRI information fusion.3.Many important problems in the analysis of neuroimages can be formulated as discovering linear and nonlinear relationships between two sets of variables.Here,we propose a new unsupervised and data-driven method,termed the eigenspace maximal information canonical correlation analysis(emiCCA).The emiCCA symmetrically analyzes both modalities simultaneously(avoiding any possible bias),and has the potential to uncover the underlying linear and/or nonlinear relationships between various data sets.Results of the simulation and task fMRI data show the superior performance of emiCCA in comparison with linear CCA and kernel CCA.These findings suggest that emiCCA is a promising technique for exploring various data.4.In this work,emiCCA and functional network connectivity(FNC)analysis are applied to investigate complex discharge-affecting networks in JME patients using simultaneous EEG-fMRI data.The present study provides evidence of complex discharge-affecting networks comprising the default model(DMN),self-reference(SRN),basal ganglia(BGN)and frontal networks in JME patients.Furthermore,our findings suggest that the BGN,DMN and SRN play intermediary roles in the propagation of epileptic discharges,and these roles further tend to disturb the switching function of the salience network.We postulate that emiCCA and FNC analysis are potential tools to provide important insights into understanding the pathophysiological mechanism of epilepsy such as JME.5.Inspired by spatial and temrpoal matching in EEG and fMRI information fusion,we propose a new hierarchical framework to ensure the relative reliability of the integrated results.As an example,spatial and temporal matching are implemented by network-based source imaging(NESOI)and maximal information coefficient(MIC),respectively.Results of synthetic and real data demonstrate that this framework will provide further insights into multimodal information integration and will likely provide important information furthering our understanding of various cognitive processes.In addition,the framework is certainly open for any other temporal or spatial matching approach.