Applying Deep Learning for Decoding of EEG and BFV about Ischemic Stroke Patients and Visualization

Abstract

In the Intensive Care Unit (ICU) environment, the electroencephalogram (EEG) and cerebral artery blood flow velocity (BFV) could help detect ischemia stroke. In our experiments, the three neural network models - - - LSTM, CNN and CNN-LSTM were designed which could help to evaluate the situation of the patient. It was found that CNN-LSTM showed the best performance in predicting the MRS with the accuracy 99.68%. In exploring the EEG, we found that the envelope of beta and gamma bands show higher correlation with the unit output of the neural network (p <0.01, Analysis of Variance). For BFV, the correlation of BFV is strongly related to the responsible arteries of ischemic stroke, with a correct rate of 75%. In conclusion, the CNN-LSTM can predict the state of ischemic stroke patients, and the envelope of EEG and PSD features of BFV are important features learned by the network model for classification.