Study On The Brain Network Of Epilepsy Based On Non-Negative Matrix Factorization And Non-Negative Tensor Decomposition

Epilepsy is a neurological disorder that results from abnormalities in brain networks,and the treatment of which varies greatly at the cluster level due to the variety of onset and location of epileptogenic regions.Studies have shown that complex information interactions between brain regions accompany seizures throughout,and visualization of effectively connected brain networks can provide a global view to explore the dynamic properties of epileptic networks in multiple dimensions and better characterize seizure dynamics,with important implications for diagnosis,intervention and prediction of epilepsy.Firstly,intracranial EEG signals from patients with temporal lobe epilepsy are preprocessed to identify the study area.Representative channels were selected based on the patient’s electrode implantation information and the waveform of the EEG signal to ensure maximum spatial coverage of the entire brain region including the temporal lobe,frontal lobe,parietal lobe,insula,and hippocampal region.After completing data preparation,an effective connectivity matrix for epileptic patients is constructed in a sliding window based on a partially directed coherence analysis method.Secondly,thresholds were determined to build an effective brain network for the temporal lobe seizure period.The time-varying matrix was decomposed one by one using a non-negative matrix factorization technique to extract the common time-domain features and establish causal relationships between channels with the same time-domain features among the outgoing and incoming time-varying features to determine the information interaction paths of seizures.The results show that there are two patterns of temporal lobe seizures,each following a specific evolutionary pathway,and that the visualization of these two pathways characterizes the information interaction trends between brain regions during the seizure phase,providing new insights into the mechanisms of seizure onset,propagation and termination.Finally,the effective brain networks of temporal lobe epilepsy patients before,during and after seizures were analyzed from three perspectives: network components,modulation frequency and time-varying features using a non-negative tensor decomposition algorithm to extract dynamic features of sub-network connectivity patterns in temporal lobe epilepsy patients in multiple dimensions in order to investigate network abnormalities during seizures.The results show that there are multiple manifestations of abnormalities in the five network connectivity patterns and that the weight coefficients during the seizure phase show significant differences compared to the other phases.This method helps to characterize the kinetics of seizures and provides a reference for the diagnosis and treatment of epilepsy.