EEG Processing Based On Multi-channel Signals And Its Application In Analysis Of Stroke Patients' EEG

Electroencephalogram (EEG) reflects the electrical activities of brain neurons from cerebral cortex, which contains losts of physiological and disease information. The finding of EEG contributes to the further study of brain structures and thinking patterns. So far, EEG is widely used in the study of information input, external-device control, psychology and rehabilitation of some diseases, and as the development of signal processing techniques, more EEG processing methods are used in the analysis of EEG Traditional EEG processing methods contain three procedures, which are preprocessing, feature extraction and classification. The EEG processing methods are the key to dig the meaningful hidden information in EEGCommonly, P300 feature extraction methods extract P300 features from too many channels of signals, which leads to a low speed of system. That's an obstacle for online experiment. To solve this problem, three P300 feature extraction methods based on multi-channel EEG signals are proposed, and the methods based on the multiplication of multi-channel EEG signals are applied to analyze the features of EEG from stroke patients.First of all, three P300 feature extraction methods based on multi-channel EEG signals are proposed. In these three methods, P300 features are extracted from the multiplication, convolution and matrix multiplication of the EEG signals from two channels respectively, and a SVM is used for classification. Datasets from P300 speller, deception detection based on facial images and deception detection in simulated network fraud condition are used to verify the performance of these methods. The results show that, compared with the related literatures, the data amount is reduced by 93.8% at most, and all the three methods achieve excellent accuracies on the datasets above. The feature extraction method based on the multiplication of multi-channel EEG signals achieves the best performance with an accuracy of 93% on the dataset from P300 speller, and on datasets from deception detection based on facial images and deception detection in simulated network fraud condition, the methods based on the convolution and matrix multiplication of multi-channel EEG signals achieve the highest classification accuracies of 88.6% and 90.6% respectively. Compared with the related literatures, these three methods increase the accuracies by 15.6,5.8 and 11.9 percentage points at most on datasets from P300 speller, deception detection based on facial images and deception detection in simulated network fraud condition respectively. It is believable that these research results can provide a help for the selection of P300 feature extraction methods on different datasets.Secondly, we record the free imagery and motor imagery EEG data from 12 stroke patients using a motor imagery EEG stimulator which is designed by ourselves. The proposed method based on the multiplication of multi-channel EEG signals is suitable for processing the EEG signals whose features are centralized in some areas over the brain surface, such as P300, imagery EEG Therefore, the EEG data from all channels are used to analyze free imagery EEG signals, and the EEG data from C3, C4 and the channels around them are used to analyze event related synchronization/d-esynchronization (ERS/ERD) of motor imagery EEG signals in time and frequency domain by the processing method based on the multiplication of multi-channel EEG signals. The results shows that in time domain, among 83.3% of the patients, the energies of the free imagery EEG from ill side of the brain are lower than the ones from healthy side, and among 66.7% of them, the phenomenon of ERS/ERD is not so notable when they complete the task of upper limb's motor imagery which is the opposite to the ill side of the brain. However, during the analysis in frequency domain, the two numbers above changes to 83.3% and 58.3%. Furthermore, we observe the phenomenon of ERS/ERD by studying the EEG data from single side of brain in frequency domain. The results show that among 66.7% of all the 12 patients, the EEG data from their ill sides of brains donot have the right phenomenon of ERS/ERD. All the research results indicate that stroke leads the restraint of the free imagery and motor imagery EEG signals, which can provide a reference for the judgement of stroke patients' recovery condition and improvement of stroke rehabilitation methods.Finally, we introduce the deception detection platform based on Guilty Knowledge Test (GKT) paradigm, EEG data acquisition platform based on the motor imagery EEG stimulator and the EEG analysis tool--MATLAB.