Functional Brain Network Learning With Spatio-temporal Information And Its Applications

Functional brain network(FBN)has become an effective tool to characterizing the functional interaction between different brain regions.It plays an increasingly important role in understanding the working mechanism of our brain and poring the pathogenesis of neurological/psychiatric diseases.Researchers have suggested that the analysis of FBNs are helpful for us to find the difference between patients and normal controls earlier than the clinical indicators.In the past decades,researchers had proposed various FBN estimation methods.However,the most existing methods.do not take the temporal and spatial information of the brain signals into account effectively,including the spatial relationship between signals that are from different brain regions and the temporal dependency within a signal.Therefore,in this paper,we mainly focus on incorporating the temporal and spatial information into the FBN estimation model,including the following two works:(1)Learning FBNs with latent temporal dependency.Recently,most of approaches estimate the FBNs under the assumption that signals are independently sampled,which ignores the temporal dependency and sequential order of different time points(or volumes).To address this problem,we introduce a latent variable into the traditional sparse representation model and a new regularization item to encode the temporal dependency within a signal.(2)Learning FBNs with spatial constraints.Most of the existing methods estimate FBNs only based on the dependency between the brain signals,which ignores the spatial relationship of the signals associated with different brain regions.Due to the space and material parsimony principle of our brain,we have reasons to believe that the spatial distance between brain regions has an important influence on FBN topology.In this approach,we not only utilize the statistical dependence of signals,but also encode the spatial relationship of signals though the low-high order form in the FBN estimation.In order to investigate the effectiveness of our proposed method,we have conducted experiments on two public databases,and the results show that our proposed methods are better than the baseline method in the sense of classification accuracy,sensitivity,specificity,F1,and AUC.