EEG Analysis Of Parkinson’s Disease Based On Communication Model

In previous literature,brain network(BN)based on constructing electroencephalogram(EEG)signals can reveal the functional mechanisms of the brain.However,this method still has many shortcomings,such as:(1)ignoring the influence of each brain region on itself when transmitting information;(2)a large number of literature have proved that multivariate signal(MS)can be decomposed into some information in the time-space-frequency domain,but EEG as a MS,BN methods do not comprehensively analyze its multivariate information transmission(MIT)from these domains at the same time;(3)although the brain is considered to be the most complex communication system known to humans,there is currently no research to reveal how information is transmitted within the brain from the perspective of communication system.Therefore,this thesis combines dynamic mode decomposition(DMD)with parallel communication model(PCM)to explore the information transmission mechanism of the brain and its influence by Parkinson’s disease(PD),by using EEG signals of PD patients and healthy control(HC)group.Our specific work is as follows:(1)Firstly,the preprocessed EEG signal is divided into frequency bands(alpha,beta,delta and theta).The EEG signal at each frequency band is treated as a large system,and then DMD is used to decompose it into subsystems with their own intrinsic oscillation frequencies.(2)At the same time,in each subsystem(corresponding to each point at frequency domain),a transition probability matrix(TPM)is calculated to describe how all leads transmit information at two consecutive moments,which describes the information transmission properties of the MS in the time-space-frequency domain,including the information transmission within and between the leads,representing a basic communication model.(3)Then,a PCM with parallel relationships in frequency domain is constructed by TPMs of subsystems.Then eight communication parameters(CPs),such as channel capacity and loss entropy etc.,and correspondingly their distribution indexes(DIs),such as skewness and kurtosis etc.,both are calculated at each EEG frequency band,to reveal information transmission phenomena of MS in time-space-frequency domain.(4)In order to verify feasibility of the PCM,significance test on extracted CPs and their DIs are performed.After Bonferroni correction for Wilcoxon rank sum test,most features have significant difference in four EEG frequency bands.(5)Finally,nine common machine learning classification algorithms are employed for PD’s classification based on extracted features.Compared with the unparallel communication model(UCM),traditional BN methods(partial correlation,phase locking value,etc.)and MS analysis methods(wavelet transform,empirical mode decomposition,etc.),the PCM can distinguish PD patients with higher accuracy.The results show that CPs extracted from EEG signals reveal the interaction mechanism between brain regions from the perspective of the communication field.This study quantitatively analyzes the specific amount of information transmitted during the communication process.After classification prediction bewteen PD and HC,the PCM’s classification accuracy mean values in alpha,beta,delta,and theta frequency bands reach 0.85,0.91,0.94,and 0.89 respectively.The PCM is significantly higher than most traditional BN and MS analysis methods and also has certain superiority in computation time,which indicates the PCM has significant advantages in disease detection.This thesis proposes the PCM based on DMD for EEG signal research of PD patients,and first demonstrates the feasibility of analyzing MIT in time-space-frequency domain simultaneously,which is impossible for traditional methods.Moreover,our study also shows that the MIT of MS in time-space-frequency domain is completely different from signal decomposition,because the meaning revealed by information transmission is not the same as signal decomposition.On this basis,the proposed method does not treat the brain as a closed system but an open system that is more consistent with previous physiological studies,since the brain can communicate not only internally but also with the external world such as peripheral nervous system and so on.Some CPs also can characterize information originated from external environment or are transmitted to it.Finally,compared with traditional methods,the proposed method not only provides a new perspective for disease detection,but also uncover many new phenomena behind MS that have not been discovered yet.In summary,PCM can quantify MIT of MS comprehensively in time-space-frequency domains,thus provide new possibilities for many future applications.