A Study On The Phase Amplitude Coupling Network Of Resting-State EEG For Children With Attention-Deficit/Hyperactivity Disorder

Objective:To analyze the cross frequency phase amplitude coupling in brain network of electroencephalogram(EEG)from children with attention-deficit/hyperactivity disorder(ADHD)before and after neurofeedback training.To study the possible mechanism of neurofeedback training,and explore the possibility of using network coupling parameters as a diagnostic tool for ADHD.Methods:A total of 60 subjects were recruited in this study,30 in ADHD group(24males,aged 8.93±2.22 years,provided with neurofeedback training),and 30 in normal control group(22 males,aged 9.93±2.46 years).The 19-channel eye-closed resting EEG were recorded from both the control group and the ADHD group before,during and after the neurofeedback training.After segmenting the data and removing the marginal effect,we used the modulation index method to calculate the coupling intensity between the low-frequency phase signal and the high-frequency amplitude signal to compare the differences of the cross frequency phase amplitude coupling of the four groups of data as well as the differences of the coupling mode.Graph theory method was put into use to analyze the coupling of network in different frequency pairs and then we studied the correlation between the coupling index and the graph theory index.Finally,four machine learning methods,binary logistic analysis,support vector machine,ensemble learning and artificial neural network,were taken to classify the coupling network indexes of ADHD group before neurofeedback training and control group.The statistical method of difference test is nonparametric test and the significance level is set to 0.05.Results:(1)There was a strong and significant coupling between delta rhythm and higher rhythm in EEG signal.(2)In ADHD group,the coupling intensity of the three stages before,during and after neurofeedback were higher than that of the normal children.(3)Significant differences were revealed in the average coupling intensity between the children with ADHD and normal children in the frequency pairs ofδ-γ_L,θ-γ_L,α-γ_L,β_H-γ_L,θ-γ_H,α-γ_H,β_L-γ_H,β_H-γ_H,especially in the coupling of low-frequency and high gamma rhythm.(4)Neurofeedback training affected significantly the coupling network ofα-γ_L,and there was no difference between the coupling in network of after neurofeedback training of ADHD group and that of normal children.It had a certain impact on the coupling network ofθ-γ_H,α-γ_H,β_L-γ_H,β_H-γ_H,and it is because of neurofeedback training that the coupling network characteristics of ADHD children tend to be similar as the normal children’s,but significant differences still cannot be ignored.It had no obvious impact on the coupling network ofδ-γ_L,θ-γ_L,β_H-γ_L.(5)Under the large-scale network cost,the ability of brain network separation,integration and average local efficiency of ADHD children were increasing.Neurofeedback training greatly corrects the abnormality ofα-γ_L coupling network of ADHD children.(6)Neurofeedback training has little influence on the network characteristics such as radius,diameter,assortativity,degree distribution,etc.The coupling network of ADHD children and normal children with neurofeedback training is basically in the form of random network,without obvious small worldness.(7)Neurofeedback training reduced the significantly increased local efficiency,local clustering coefficient and intermediary centrality in the central area,right temporal area and occipital area of ADHD children.(8)In the coupling network with multiple frequency pairs,the coupling intensity had a positive correlation with the network separation ability,integration ability,average local efficiency,node importance,and a negative correlation with the characteristic path length,radius,and diameter.(9)The classification effect of artificial neural network,support vector machine,binary logistic analysis and ensemble learning was gradually increasing.Ensemble learning achieves accurate classification(accuracy,sensitivity and accuracy are all 100%)when using all feature sets ofα-γ_Hcoupling network.The average accuracy rate of the ensemble learning in the classification ofθ-γ_H,α-γ_H,β_L-γ_H,β_H-γ_H coupling network reach 96.1%.Conclusions:There is a close relationship between the intensity of phase amplitude coupling and the coupling network index.There are obvious differences between ADHD children and healthy children,especially the coupling between low frequency and high gamma rhythm.The positive change ofα-γ_Lcoupling network may be the mechanism of neural feedback training to improve ADHD symptoms.The coupling network indexes ofθ-γ_H,α-γ_H,β_L-γ_H,β_H-γ_H(especiallyα-γ_H)in resting EEG,combined with ensemble learning,can provide a quantitative method for assisting diagnosis and evaluating the therapeutic of ADHD.