The Study Of Anesthesia Emergence Classification And EEG Modeling

Human brain and its related physiological changes has been the hot and difficult topic in scientific studies.Currently,the mechanisms of general anesthesia is still not clear.The patients will appear to various electroencephalogram emergence patterns during the recovery period,because of the different biological processes of individual patients.However,these patterns can not be quantitatively identified using commercially available depth of anesthesia.This is a huge problem for clinicians to judge patient's postoperative brain status effectively.In order to solve about problems,this paper provide a index which is used as feature to classify the EEG emergence pattern by machine learning algorithm,and builds an anesthesia electroencephalographic model to realize these pattern by adjusting parameter.Firstly,we find four EEG emergence patterns by frequency spectrum from 52 patients.State entropy,SynchFastSlow,permutation entropy and approximate entropy,as well as the relative power spectrum density of five frequency sub-bands of clinical interest are used as the feature to classify the EEG emergence pattern by the genetic algorithm support vector machine.The result show that the relative power spectrum density is the best feature.The machine learning methods of grid-search based SVM and the BP neural network were compared.The performance was reported in terms of sensitivity,specificity and accuracy.The result shows that the GA-SVM had a better performance in small sample classification.To analysis the potential inner relationship of emergence patterns and physiological information,we had done a statistics for all class subjects classified.We find the age is a important factor,the EEG emergence patterns may correlate with underlying neural substrate which related with patients' age.Secondly,for the modeling study of anesthetic mechanisms,we built a model which can describe the interrelationship between the thalamus and cortex and reproduce the macroscopic EEG.In order to simulate the EEG signal during the general anesthesia,the grid search algorithm is used for tuning parameters based on permutation entropy and correlation coefficient.In our study,the EEG data from the propofol induced general anesthesia was compared with the simulated EEG data in respect to log spectrum,permutation entropy and relative power spectral density.The result shows that the simulated EEG data could reflect some main features of real EEG data.For the recover period,we adjusted the parameter by relative power spectral density,and the parameters range of two kinds of patterns is given.Finally the model was extended to two-channel,and analyze the influence with the increasing of couple strength.