EEG Research On Biomarker For Episodic Memory Deficit And Automatic Seizure Detection And Prediction

Epilepsy is a brain nervous system disease and one of the major health and safety issues in the world.Temporal lobe epilepsy(TLE)is common refractory epilepsy,often accompanied by episodic memory(EM)impairment,which greatly reduces patients’ life quality and increases the burden on social healthcare.The diagnosis time of episodic memory impairment,the frequency and intensity of seizures are important factors that affect the progression of TLE patients’ EM deterioration.Early diagnosis will help obtain the optimal time window for treatment.Accurate seizure detection can remind patients to seek timely treatment and help doctors judge the condition,thereby reducing the seizure frequency and intensity.Early seizures warning will provide patients and medical staff with sufficient preparation time,thereby reducing the intensity of the attack and avoiding further memory deterioration caused by secondary brain damage during the attack.However,the traditional scales used for memory function assessment in clinical practice have strong subjectivity,low repeatability,and low sensitivity,which makes it easy to miss the early diagnosis window for EM impairment.Moreover,seizure detection is usually completed through visual examination by doctors,which is timeconsuming and experience-dependent,resulting in patients in economically underdeveloped areas unable to receive timely diagnosis and treatment.Seizure prediction is usually alerted through common sensory symptoms,but they vary among different individuals and generally appear only within several seconds or minutes before a seizure,which provides insufficient preparation time.Therefore,this study explores the possibility of alleviating or even curing EM impairment in TLE patients from the three important aspects: early diagnosis,seizure detection,and seizure prediction.The main work is summarized as follows:(1)In response to the limitations of traditional scales used for memory function assessment,such as strong subjectivity,low repeatability,and low sensitivity,eventrelated potentials(ERP)were used to investigate the biomarkers of early EM impairment in TLE patients.A visual change detection task with image stimuli was designed to assess episodic memory.During the memory encoding and decoding phases,the ERP signals were recorded from thirty TLE patients and thirty healthy controls.Given that EM is a complex process involving attention and working memory(WM),the amplitudes and latencies of EM-related ERP(FN400,late positive potential(LPC),and late posterior negativity(LPN)),attentional ERP(P100,and N100),and WMrelated ERP(P200,and P300)were calculated.Then we conducted the inter-group and inter-condition statistical analysis for Wechsler Memory Scales-Chinese Revision(WMS-RC)assessment,experimental behaviors,and the ERP metrics.In addition,the correlation analysis among scale results,behavioral data,and ERP was carried out.The results showed that the TLE patients performed worse in WMS-RC and the memory task.The ERP results showed that the TLE patients did not suffer from attention but WM impairment.For EM-related components,differences were observed between the TLE patients and the controls: the lack of the FN400 effect,the inverse of the LPC effect,the reduced FN400 and LPC,and the increased LPN.No significant inter-group difference was detected for the latencies of all the ERPs.Additionally,there were significant correlations among WMS-RC scores,behaviors,and some ERP amplitudes.The following conclusions can be drawn: the EM deficit was more associated with WM damage than attention damage.More importantly,the impaired old/new effect of the FN400 and LPC,and the abnormal amplitudes of FN400,LPC,and LPN might be the potential physiological biomarkers for the visual EM deficit in TLE patients.These findings were expected to guide clinical evaluation and intervention of TLE patients,and help obtain the optimal time window for EM impairment treatment.(2)In response to the strong dependence of seizure detection on experienced doctors in clinical practice,as well as the low universality and short warning time window of sensory symptoms for seizure prediction,this study proposes an automatic seizure detection and prediction method based on brain connectivity features.Two window lengths(1 s and 8 s)were employed for EEG data segmentation.Five physiological wave bands(i.e.,δ,θ,α,β,and γ)and five connectivity measures(i.e.,Pearson correlation coefficient(PCC),phase locking value(PLV),mutual information(MI),Granger causality(GC),and transfer entropy(TE))were used to extract imagelike features,which were fed into a support vector machine(SVM)for the subjectspecific model(SSM)and into a CNNs-Meet-Transformers(CMT)network for the subject-independent model(SIM)and cross-subject model(CSM).Finally,t-SNE feature reduction and silhouette coefficients were used to measure feature separability for feature selection.After feature selection,the time and storage efficiency under the optimal feature combination were also analyzed.The classification results on the CHBMIT dataset showed that the features extracted in the 8s-window were more effective than those in the 1s-window.For seizure detection,the best accuracies of SSM,SIM,and CSM were 99.29,100,and 94.59%,respectively.The highest accuracies for seizure prediction were 98.99,98.98,and 84.58%,respectively.In addition,PCC and PLV features in the β and γ bands showed good performance and high speed.Based on the above results,the proposed brain connectivity features demonstrate good reliability and practical value in automatic seizure detection and prediction,which showed that brain connectivity matrix can be a potential two-dimensional feature candidate.This method could help remind epileptic patients to prepare intervention measures in advance,and reduce seizure frequency and intensity,thus slowing down the EM deterioration process.In summary,the work revolves around the three important aspects: early diagnosis,seizure detection,and seizure prediction.This study explores the potential ERP biomarkers for EM impairment,and proposed methods for automatic seizure detection and prediction,which are expected to improve the possibility of alleviating or even curing EM deficit in patients with temporal lobe epilepsy,leaving certain guiding significance and application value in clinical diagnosis and treatment.