The Spatio-temporal Manifolds Of Dynamic Functional Connectivity In Human Brains

Dynamic functional connectivity(dFC)focuses on the time-varying couplings between the two brain regions isolated from each other in brains,and the temporal correlation of neural activity sequences exhibits instantaneous dynamic changes.The potential correlation between the functional connectivity and the behavioral information suggests that the dynamics of functional connectivity is partly due to neural activity and may be related to the transformation of the cognitive or alert states,as well as some diseases.The change of the connectivity characteristics also indicates the physiological significance of dFC,and can be used as a biomarker for disease diagnosis to some extent.In order to explore the objective laws and useful information in high-dimensional magnetic resonance data,dimensionality reduction is important to mine the implicit information of dFC datasets.For this purpose,we systematically studied the impact of spatiotemporal resolution on dFC analysis based on HCP dataset and ASD dataset.Then,we proposed the combination model of dynamic functional connecitivity and the local linear embedding methods and explored the spatiotemporal manifolds of sleep deprivation,the resting state,as well as task states with K-means clustering and nonlinear spectral algorithm.The main contents of this paper are as follows:Manifold learning for the temporal dynamics of dFC.Based on the sliding time window correlation,local linear embedding and K-means clustering method,we studied the temporal dynamic features of functional connectivity in human brains.The results showd that we successfully extracted the potential relation between the static functional connectivity and the dFC of the low-dimensional manifolds.Whhat’s more,the brain functional connectivity in the low-dimensional space was different from that in Euclidean space.By observing the manifold distribution of the functional connectivity’s evolvement in low-dimensional space,we thought about the meaning of the dynamic functional states’ evolvement and the exploration of the evolutionary mechanism for the brain activity under different tasks.Functionally consistent coupling patterns analysis of dynamic functional connectiviy.Based on the sliding time window correlation and local linear embedding methods,we used a K-means clustering algorithm to separate the coherent coupling patterns of the brain dynamcic function connectiviy.The results showd that,in the resting-state,the functionally consistent coupling patterns had a significantly consistent functional connection mode,comparing with the functional network in the brain template.Some patterns showd strongly functional homogeneity.It is worth noting that the network interactions between the association cortex and whole brain were separated in one pattern,which had not been isolated separately in previous high-dimensional clustering studies.It indicated that the main cortex brain region and the association cortex are discrimative in the nonlinear functionally consisit couplings patterns.The manifold structure topology for dynamic functional connenctivity.Based on sliding time window correlation and local linear embedding algorithm,we used the nonlinear spectral method to detect the topological structure in the embbedded dataset,rank each functional connectivity pair in the whole brain topology based on ”hub-like”degree.It showd that the ranking score for the dFC pairs had significant behavioral relevance and physiological significance.Using the topological properties under different states,we further effectively distinguished the 7 task states and the resting state,as well as the sleep state(acute sleep deprivation and full rest state).This results indicated the practical applicability of our model.