Analysis Of EEG Characteristics Of Children With Attention Deficit Hyperactivity Disorder Based On Non-negative Tensor Decomposition

Attention deficit/hyperactivity disorder(ADHD)is a neurodevelopmental disorder that is characterized by attention deficit,hyperactivity or impulsivity.At present,the diagnosis of children with ADHD mainly relies on behavior and scale assessment,but this assessment method is easily affected by subjective factors,so the measurement results may not be accurate.Therefore,finding a method that can objectively evaluate ADHD has important research significance.Electroencephalogram(EEG)has become an electrophysiological technique widely used in ADHD research.In the process of recording EEG,it is unavoidable to be interfered by ocular artifacts,so that the result of subsequent analysis and processing of EEG signals may not be accurate.Therefore,it is a very important step to remove ocular artifacts in EEG signals.EEG measurements such as spectral power and functional connectivity have been used to study the neural basis of ADHD.However,the current research on the power spectrum has produced different results,and the traditional brain network research is mainly a fixed single frequency research method,and the current extraction of the brain network is based on a single variable.The research on power spectrum and brain network continues.This article focuses on the assessment of children with ADHD.The main research is as follows:(1)Remove the ocular artifacts in the multi-channel EEG signal.We propose a method of automatic ocular artifacts removal(AOAR)based on non-negative matrix factorization and empirical mode decomposition.This method is compared with other 7 methods based on simulated data and real EEG data.The results show that the AOAR method is effective and reliable in processing EEG data.(2)In view of the problem of different results obtained from power spectrum studies,we suspect that traditional methods cannot clearly distinguish between rhythmic and non-rhythmic activities.We propose a modified Better OSCillation detection(MBOSC)to identify rhythmic brain activity.The performance of the MBOSC method is evaluated in simulation data.Then use relative power and MBOSC to analyze resting state EEG signals.The results show that the frequency range detected by MBOSC is more concentrated,the relative power change in the alpha band is caused by rhythmic brain activity,and the relative power change in the delta band is caused by non-rhythmic brain activity.(3)Aiming at the shortcomings of the current brain network extraction and analysis methods,we use non-negative tensor decomposition to extract the characteristics of the brain network.Analyze the brain network from two aspects:static brain network and dynamic brain network.The results show that the two groups have significant differences in the delta,theta,alpha and beta frequency bands at the same time under the two networks,and the energy of the ADHD group is lower than that of the typically developing group.In ADHD group,the connection between occipital lobe and frontal lobe and the connection dependence between frontal lobe increased,and the connection between frontal lobe and parietal lobe and the connection strength between occipital lobe and occipital lobe decreased.