The Research Of Resting State EEG In Mild Cognitive Impairment And Alzheimer’s Disease

Objective:(1)To study the incidence of seizures in mild cognitive impairment(MCI)and Alzheimer’s disease(AD)and the clinical characteristics of patients with seizures.(2)To analyze the EEG abnormalities diagnosed by routine EEG in patients with MCI and AD,and explore their relationship with cognitive function and disease progression.(3)To analyze the diagnostic,assessment of disease progression and predictive effects of EEG features in MCI and AD based on resting-state EEG signals and machine learning.Methods:(1)Retrospectively analyzing the incidence of seizures in 205 MCI and 297 AD patients.The general information of the patients,dementia-related clinical features,age of seizures,and seizure patterns were collected,and the characteristics of seizures among MCI and AD patients were summarized.(2)Retrospectively analyzing the clinical characteristics and routine EEG diagnosis reports of 205 MCI and 297 AD patients,summarized the EEG abnormalities in MCI and AD patients,and analyzing the correlation between EEG abnormalities and cognitive function and their predictive effect on disease progression.(3)Clinical information and EEG data were collected from 189 MCI patients,330 AD patients,57 vascular cognitive impairment(VCI)patients,47 frontotemporal dementia(FTD)patients,21 DLB patients and 246 healthy controls.Among them,57 MCI patients and 150 AD patients completed the detection of APOE genotype,and 28 MCI patients and 87 AD patients completed the detection of CSF biomarkers.Six EEG features were extracted which included absolute power spectral density(abs＿psd),relative power spectral density(rel＿psd),Hjorth parameters(activity,mobility,and complexity),sample entropy,time-frequency characteristics,and microstate.Three classifications were performed for the HC,MCI and AD cohorts using linear discriminant analysis and support vector machines.EEG features that could identify other subtypes of dementia were explored using the ANOVA test.The correlations between EEG features and MMSE,Mo CA scores,and CSF biomarkers(Aβ,tau,etc.)in the MCI and AD groups were evaluated using Pearson correlation analysis.Multimodal features were independently selected using machine learning to train a series of random forest regression models for prediction of MMSE,Mo CA,age of onset,and disease duration,respectively.Results:(1)The incidences of seizures in MCI and AD patients were 1.46%and 3.7%,respectively;the seizures in MCI and AD patients were diverse,50% were generalized tonic-clonic seizures,and 28.6% were focal seizures with impaired consciousness.seizures,14.3% were absence seizures,and 7.1% were myoclonic seizures;the course of AD patients with seizures was significantly shorter than that of AD patients without seizures(2.78 ± 2.46 vs 4.91 ± 3.67,p = 0.006),AD patients combined with seizures also had significantly lower MMSE scores than AD patients without seizures(8.18 ± 3.97 vs 11.66 ± 6.63,p = 0.017).(2)In MCI patients,normal EEG accounted for 46.3%,borderline state accounted for 20.0%,mildly abnormal accounted for 28.8%,moderately abnormal accounted for 2%,and severely abnormal EEG accounted for 2.9%.In AD patients,normal EEG accounted for 25.3%,borderline state accounted for 15.2%,mildly abnormal accounted for52.2%,moderately abnormal accounted for 5.4%,and severely abnormal EEG accounted for 2.0%.The MMSE score of the abnormal EEG group in AD patients was lower than that of the normal EEG group,and the difference was statistically significant(10.64 ± 6.00 vs 14.16 ± 6.42,p <0.001).Regarding the effect of EEG on disease progression,a total of 18 MCI and 18 AD patients were included in the analysis.It was found that in MCI patients,individuals with EEG abnormalities had faster MMSE decline(p = 0.016),while no correlation between EEG abnormalities and cognitive decline was found in AD patients(p = 0.384).(3)In order to distinguish MCI,AD and HC,we obtained a total of178 dimensions of key EEG features.The optimal feature set almost covers some features of each EEG feature category,especially the absolute power spectral density and the complexity of EEG signals.The binary classification of HC vs.MCI achieved about 80% accuracy and80.9% F1-score,while the binary classification of HC vs.AD achieved about 85% accuracy and 85% F1-score.For the three classifications of HC,MCI and AD,the accuracy rate and F1 score both reached about70%.Key EEG features mainly include abs＿psd in occipital theta band,rel＿psd in occipital theta band,parieto-occipital Hjorth mobility can also be used to differentiate HC,MCI,AD,DLB,VCI and FTD(Fs > 40,p <0.0001).The abs＿psd of O2 lead theta band was negatively correlated with Aβ42(r =-0.358,p < 0.001)and positively correlated with p-tau(r =0.442,p < 0.001).The rel＿psd of the theta band in lead O1 was positively correlated with Aβ42(r = 0.373,p < 0.001)and negatively correlated with p-tau(r =-0.447,p < 0.001).In terms of cognitive assessment,Hjorth mobility of channels O1,O2,and P4 was found to be positively correlated with MMSE and Mo CA scores(r = 0.416-0.464,p < 0.001).In predicting clinical features using multimodal features,it was found that mixed features(EEG+CSF/APOE+gender/age)outperformed CSF/APOE features over EEG-only features for prediction of patient MMSE scores.Similar results were observed when predicting patients’ Mo CA scores.With regard to prediction of age at onset,hybrid features were superior to EEG-only features over CSF-APOE features.In contrast,with regard to the prediction of disease course,only EEG features achieved the best performance.Conclusions:(1)Patients with MCI and AD have an increased risk of seizures during disease progression,and patients with faster cognitive decline and higher severity of cognitive impairment are more prone to seizures.(2)Routine EEG examination may be useful in the differential diagnosis of MCI and AD,abnormal EEG in AD patients is associated with the severity of cognitive impairment,and may be used for prognosis prediction in MCI patients.(3)EEG features can be used in the diagnosis and differential diagnosis of MCI and AD patients and are correlated with cognitive severity and pathological markers.EEG features are reliable biomarkers for MCI and AD diagnosis and disease progression monitoring.