Classification Of Sedation States Based On Machine Learning

It is important to monitor the sedation depth of anesthesia to judge the state of brain consciousness during operation.Electroencephalogram(EEG)can reflect the excitation or suppression of neurons in anesthesia,and then reflect the degree of excitation or inhibition of brain consciousness.Therefore,the use of EEG signals can directly analyze the degree of cerebral sedation,and can be used as an intraoperative monitoring means to feedback the sedation state of anesthesia and adjust the dosage of anesthesia drugs in surgery.EEG dual-frequency index,as a relatively authoritative sedation condition monitoring index,is widely used in the monitoring of intraoperative anesthesia depth,but its calculation method and the combination relationship of various parameters are not open,and the anesthesia monitor with it as the index is expensive.However,other single feature index,such as entropy,Nacrotrend and spectral energy,are usually cannot satisfy the different sedation depth accurate monitoring.In this paper,based on EEG signals,a combined monitoring index of sedation characteristics and the sedation state recognition of image features are proposed.Firstly,a variety of denoising methods are adopted for the single channel anesthetic EEG signal to eliminate noises such as 50 Hz power frequency and baseline drift.On this basis,we extracted nine sedation monitoring feature index of electroencephalogram signals of anesthesia.These parameters are conventional characteristic parameters applied in anesthesia,and have been proved to be correlated with anesthesia characteristics and age characteristics in many literatures.Secondly,the random forest method was used to train multiple classification trees,and a variety of machine learning classification algorithms were compared to achieve accurate prediction of the test samples in the waking period,anesthesia period,recovery period and burst suppression period.Furthermore,a kind of characteristic index for monitoring sedation state with multiple combinations of anesthetic EEG features is put forward,which realizes the accurate classification of different sedation states of patients under the control of different anesthetic drugs for adults and children By using the supportvector machine regression method optimized by particle swarm optimization algorithm,the regression fitting of bispectral index is realized.Finally,based on EEG signals from adults and children under control of different anesthetic drugs,analyzes the spectrum feature of the electrical diagram and the recursive diagram under different anesthetic sedative state difference.Convolution of the neural network method,set up with multiple convolution and pooling layer neural network structure,the characteristics of four kinds of sedation photo collections feature learning and sedation recognition,the convolutional neural network are analyzed in terms of identification of anesthetic sedative state performance,deep is verified based on neural network,the spectrum characteristics of EEG is used in the depth of sedation.