Feature Extraction Of EEG Signals From Mild Cognitive Impairment Based On Correlation Analysis

The diagnosis of Mild Cognitive Impairment(MCI)has become a hot topic in academic circles at home and abroad.Among them,EEG(electroencephalogram)signal analysis has been widely recognized by the academic community as the main method of MCI diagnostic research.In order to deeply study the diagnostic methods of MCI patients based on EEG signals,we need to focus on the correlation of neural activity in each brain region.In order to effectively quantify the correlation of MCI’s neural activity,this paper proposes a multi-channel EEG signal feature extraction method based on correlation analysis from two different perspectives.Firstly,this paper proposes a feature extraction method based on correlation coefficient matrix decomposition.The method first calculates the Pearson correlation coefficient between the two-channel EEG signals,and then uses the matrix decomposition method to reduce the coefficient matrix.From the perspective of characterizing the correlation between brain regions,this method can be well used for feature extraction of MCI EEG signals.Secondly,in view of the current EEG signal coupling relationship feature extraction without considering the difference between different combinations of brain regions,the source data form is single,this paper proposes Weighted Permutation conditional mutual information(WPCMI)method.Methods The frequency domain features of multi-channel EEG signals were used as weights,and the characteristics of the coupling relationship were improved by the combination of weighted average and time domain coupling characteristics.In order to characterize the significant difference between the MCI EEG signals and the normal control group EEG signals,the MCI EEG signals and the normal control EEG signals were classified and identified.Finally,the above two methods are applied to the feature extraction of MCI multichannel EEG signals.Firstly,the EEG signals of MCI patients and control groups were extracted from two angles,and the above two feature extraction methods were compared and compared with the existing methods.The experimental results show that the correlation coefficient matrix decomposition method and the weighted permutation condition mutual information method both show excellent performance.