Feature Extraction And Classificationof Resting-state EEG Signal Dynamic Features In Mild Cognitive Impairment

At present,China has entered an aging society,and the number of patients with Alzheimers Disease(AD)is increasing.Mild Cognitive Impairment(MCI)is an early stage of AD,and the diagnosis of MCI has attracted more and more researchers at home and abroad.Electroencephalogram(EEG)signal analysis is an efficient way to diagnose MCI.Deep learning algorithms have achieved great success in various fields.Therefore,deep learning algorithms are also widely used in EEG signal analysis.Improving EEG-based MCI diagnostic results requires excellent feature extraction methods and appropriate deep learning algorithms.In this paper,the EEG signal feature extraction and deep learning are studied,and the weighted sorting mutual information is optimized and improved.Based on this,a suitable deep learning model is constructed.Firstly,after analyzing Weighted Permutation Mutual Information(WPMI)and Permutation Mutual Information(PMI),an optimized Weighted Permutation Mutual Information method was proposed to calculate the amplitude characteristics of brain signals from different perspectives as weights.According to the characteristics of eeg signals,the WPMI characteristic values of multi-band combination are mapped to RGB images,and Convolutional Neural Networks(CNN)are constructed to classify and identify MCI.Secondly,a multi-input convolutional neural network model is constructed.The model can accept multiple inputs,classifying WPMI-RGB feature images combined with various frequency bands or WPMI-RGB feature images of various weights as inputs.The multiinput method allows the convolutional neural network to learn more features and improve classification performance.Finally,using the EEG signals of aMCI patients and normal control(NC)as experimental data,the WPMI method and PMI eigenvalue calculation methods proposed in this paper are compared and analyzed,and the feasibility of the algorithm is analyzed validity and accuracy were verified.The classification results of the multi-input model and the single-input model are compared to verify the superior performance of the multi-input model.