Research On EEG Signal Analysis And Cognitive Mechanisms Based On Deep Learning

The brain contains hundreds of millions of nerve cells,which process all kinds of information all the time.Studying the operating mechanism of the brain will help to deepen our understanding of it.By studying the electroencephalogram(EEG)signals of different task states and extracting the key components of EEG(such as time,frequency,activation distribution of brain area,etc.)by deep learning,the mechanism of brain nerve operation is explored.In this paper,deep learning is used to analyze and study the EEG key components extraction effect of healthy people under different cognitive experiments(auditory attention experiment and P300 intention recognition experiment)from different perspectives,and the variation EEG information extraction of patients(major depressive disorder)compared with healthy people under similar cognitive experiments(probability learning experiment).The main contents are as follows:1.Under the experimental paradigm of auditory attention in our laboratory,two methods of EEG time-frequency analysis and deep learning decoding are used to study.The results of the time-frequency analysis showed that around 250 ms after the appearance of the speech keyword,the rhythmic energy changes in the EEG signal segment of focused attention were more significant and concentrated than in the non-attentive signal segment,and the brain regions with significant changes were mainly distributed in the temporal and central frontoparietal lobes.The feature map decoded by deep learning is a2 D feature map obtained by mapping the EEG to the cortex using the source localization method,using the EEG cortical signal to calculate the rhythm entropy feature,and then using the Mollweide Projection.The test accuracy of each subject was calculated using the leave-one-subject method,and the results showed that the average accuracy was84.4%,which was higher than the chance level of 58.8%,and only one subject was below the chance level,indicating that the deep learning model achieved better classification results.Deep learning visualisation techniques were used to extract the location distribution of key brain regions and combined with time-frequency analysis to explain the auditory attention mechanism.2.A poker picture P300 experiment was designed and EEG signals were collected to investigate intention recognition experiments based on the P300 signal.The event-related potential(ERP)method was used to compare and analyze the ERP components of the biased stimulus(1 stimulus picture as the target)and the standard stimulus(7 stimulus pictures),and the results showed that only the target stimulus had a significant P300 ERP component.Using deep learning methods to classify single-trial EEG,the leave-one-subject method showed that the average classification accuracy of the subjects was 67.4%,which exceeded the chance level by 19.8%.The key information was obtained through the gradient visualization method and the time and electrode channels of the P300 components were compared.The results showed that the deep learning extracted key temporal and spatial information(key electrode positions).Through source-localized cortical signal analysis,more precise brain area activity for categorizing signals is found in cortical source signals.3.Based on the good results obtained in the study of healthy people,apply deep learning to the diseased population to extract the variant EEG information of the diseased population.In the probabilistic learning experimental dataset,ERP analysis showed that patients with major depression had greater negative feedback waves in negative feedback stimuli compared to healthy subjects.Using the EEGNet deep learning model to decode the dataset,the four-category results obtained by leave-one-subject cross-validation showed that patients with major depressive disorder received the highest classification accuracy rate of 71%(chance level 28.8%)under negative stimuli.Using gradient visualisation and source localisation analysis it was obtained that patients with major depression had abnormally enhanced activity in the central frontal brain region under negative stimuli,and the results from ERP analysis indicated that patients with major depression were overly attentive to negative feedback stimuli.