| Objective:The purpose of this research was to understand and investigate the clinical characteristics of temporal lobe epilepsy(TLE)patients with cognitive impairment,and to use the machine learning(ML)techniques to compare the evolution patterns of algorithm features in electroencephalogram(EEG)signals between TLE patients with cognitive impairment and TLE patients without cognitive impairment.Finally,the classification performance of different feature extraction algorithms and classifiers for EEG signals was compared.Methods:During September 2020 to January 2023,46 adult patients diagnosed with TLE at the Epilepsy Center of the First Hospital of Jilin University were admitted to our study.The general demographic characteristics,clinical features,auxiliary test results and 30-minute EEG fragments in the tranquil period of the patients were entirely collected and recorded.EEG signal features were extracted using the log energy entropy(Log E),permutation entropy(PE),and central moment(CM)algorithms.Combining with different classifiers,such as support vector machine(SVM),k-nearest neighbors(KNN),decision tree(DT),and linear discriminant analysis(LDA),the classification model for EEG segments with and without cognitive impairment in TLE was constructed.Furthermore,we investigated the evolution of EEG signals in TLE patients with cognitive impairment based on various eigenvalue extraction algorithms,and compared the performance of different classification models in automatic EEG signals recognition.Results:(1)In TLE patients,the amount of IEDs was associated with altered cognitive function,and there was a statistically significant difference in the distribution of the moderate and large number of IEDs in patients with or without cognitive impairment(P=0.002).Age,education,duration of epilepsy,the age of onset,seizure type,seizure frequency,and other characteristics were not connected to cognitive change in patients with TLE.(2)Among the TLE patients with cognitive impairment,the scores of MMSE and Mo CA were higher in patients with unilateral IEDs than in patients with bilateral IEDs(P<0.05);other clinical characteristics had no significant effect on MMSE and Mo CA scores,such as duration of disease,the age of onset,and seizure frequency.(3)The cognitive impairment of TLE patients was significant in aspects of orientation,attention,recall,language structure,replication ability,visual space,naming,and delayed recall.(4)While PE features were increased in all leads compared to those in TLE without cognitive impairment EEG segments,Log E features and CM features were decreased in all leads in TLE with cognitive impairment EEG signals.(5)Among the classification models constructed by Log E features and four classifiers,the Log E-KNN and Log E-SVM combination performed better in the classification task for two distinct kinds of EEG segments,including TLE with cognitive impairment and TLE without cognitive impairment,with the accuracy,sensitivity and specificity of more than99.00%.Among the classification models constructed by PE features and four classifiers,in classification task for two EEG segment,including TLE with cognitive impairment and TLE without cognitive impairment,the PE-KNN combination had the better average classification accuracy and specificity with 87.29% and 99.94%,the classification sensitivity of the PE-SVM combination was 100.00%.Among the classification models constructed by CM features and four classifiers,in classification task for two EEG segment,including TLE with cognitive impairment and TLE without cognitive impairment,the CM-KNN and CM-SVM combination showed better classification accuracy,sensitivity and specificity,all of which were above 97.00%.Conclusion:(1)The amount of IEDs varied across TLE patients with and without cognitive impairment,although there were no statistical differences in the course,seizure type,seizure frequency,or other features.(2)TLE patients showed deficits in several cognitive areas,including orientation,attention,recall,language structure,replication ability,visual space,naming,and delayed recall.(3)Compared to EEG segments of TLE without cognitive impairment,the Log E features and CM features of TLE with cognitive impairment EEG segments were decreased in all leads,while PE features were significantly higher in all leads.(4)Log E-SVM,Log E-KNN,PE-KNN,CM-KNN,and CM-SVM models were proposed for automatic identification and detection of EEG segments in TLE patients with or without cognitive impairment. |
