Research On Machine Learning Classification Of Schizophrenia Based On EEG Features

Schizophrenia(SCZ)is a severely harmful chronic mental illness,which may bring a heavy burden to individuals and families.At present,the etiology and pathogenesis of SCZ are not fully understood.Clinically,it mainly depends on the clinical experience of clinicians and oral statement of patients and family members for diagnosis and curative evaluation,lacking of objective indicators.Therefore,looking for objective electrophysiological markers has important clinical significance.Anti-schizophrenia drugs are generally used in the clinical treatment of SCZ.This paper was aimed at SCZ patients with taking Clozapine.Starting with resting state electroencephalography(EEG)data,through power spectrum and brain network analysis,EEG data before and after medication were compared to explore SCZ specific electrophysiological indicators,and on this basis,machine learning(ML)methods are used to further verify the role of these electrophysiological indicators in the clinical diagnosis and efficacy evaluation of SCZ.Details are as follows:1.Based on the resting state EEG data of SCZ patients,the differences in the brain network topology and network properties of power spectrum were compared with those of Healthy Control(HC).The results showed that SCZ patients had significantly higher power spectra in the high frequency bands(beta1,beta2,and gamma)at the frontal,temporal,and parietal lobe.The connection of gamma-band brain network was significantly enhanced in the frontal lobe,frontal lobe-parietal lobe,and frontal lobetemporal lobe,and the frontal lobe-central area.The brain network properties further showed that the degree of information interaction in the brain area of SCZ patients was increased.After two months of treatment with clozapine,the increase in the power spectrum of high frequency has been relieved at some brain regions,mainly in the frontal lobe.Meanwhile,the connection between the frontal lobe,frontal lobe-apical lobe,and the hemispheres was significantly reduced.These results indicated that the power spectrum in high frequency bands(beta1,beta2,and gamma)and the brain network of gamma band were highly correlated with SCZ.2.Based on the above resting EEG data analysis,the classification selection algorithm was applied in the power spectrum of the high frequency bands(beta1,beta2 and gamma)and the brain network of gamma band,to select the most identifiable feature,and then to sort the data.The results indicated that the SVM classifier had a better effect after using the RF algorithm to get a subset from the gamma network feature.Among them,the accuracy was 96.15% for SCZ classification,and 96.15% for the efficacy evaluation of Clozapine.The selected features also had a good interpretability.These results further suggested that the topological characteristics of the brain network in the gamma band might be used as an objective electrophysiological indicator for the diagnosis of SCZ and the efficacy evaluation of Clozapine.This study mainly introduced the feature selection algorithm into the machine learning method and found that the topological characteristics of the gamma brain network could be used as an objective electrophysiological indicator of SCZ to assist clinical diagnosis,meanwhile,it also has a certain reference value in assisting the evaluation of clinical drug efficacy.