Study On Eeg Feature Analysis And Diagnostic Identification Of Mild Cognitive Impairment

Mild Cognitive Impairment(MCI)is an intermediate state between normal brain aging and dementia,with MCI patients having a 10 times higher probability of developing Alzheimer’s Disease(AD)compared to normal elderly individuals.The electrophysiological characteristics of brain signals can effectively reflect abnormal changes in brain function.By analyzing MCI brain signals,meaningful features can be extracted to achieve MCI diagnosis and identification,which has important theoretical research significance and clinical application value.In this thesis,different features of the power spectrum density,fuzzy entropy,and multi-fractal characteristics of the EEG signals of MCI patients and healthy control groups were analyzed from different perspectives such as frequency domain and non-linearity.Independent sample t-tests were used to test the differences in different features between the two groups.Except for power spectrum density,all other features showed significant intergroup differences(P<0.01).Using directed transfer function as a brain functional connectivity index,the optimal threshold was determined by combining the threshold selection method,and the brain functional network was constructed.Based on this,the global efficiency,local efficiency,clustering coefficient,and node degree,which are characteristic quantities representing the brain functional state,were analyzed.Independent sample t-tests were used to analyze the correlations,and all features showed significant intergroup differences(P<0.01).To address the MCI diagnosis problem,a Bayesian optimization-based bidirectional long short-term memory network diagnostic model(BO-BiLSTM)was improved by combining the prior distribution with the posterior probability results to optimize the hyperparameters of the BiLSTM network.MCI single features and combined features were used as diagnostic model inputs for classification and recognition.The results showed that when using power spectrum density as input,the BO-BiLSTM network achieved the highest MCI diagnosis recognition rate of 93.95%±1.26%.When multiple characteristic quantities were fused as input to the model,the accuracy of the BO-BiLSTM model reached 98.64%±0.61%.Compared with traditional machine learning algorithms,the improved model has higher accuracy.