Bispectral Index Prediction Algorithm Based on Multi Feature Screening
Bispectral Index(BIS)is a derivative index obtained from EEG signals.It is the most widely used index for depth of sedation monitoring at present.However,according to the calculation of EEG,there is a 20 s～30 s update delay in BIS,which af-fects the anesthesiologist's judgment of the patient's sedation state,increases the risk of the patient's intraoperative awareness,and affects the patient's postoperative recovery.In this paper,it combines the brain signal parameters,the historical BIS and the vital signs of patients to predict the BIS value,so as to provide anesthesiologists with the latest trend of sedation and facilitate their accu-rate adjustment of anesthetic drugs.This paper constructs a BIS prediction model based on multi feature screening.Firstly,the EEG signal is denoised by empirical mode decomposition algorithm,and the sub parameters of the denoised EEG signal are extracted.Then,the random forest algorithm is used to screen the sedation feature parameters,and the five features with the highest correla-tion with BIS are obtained.Finally,the screened features and the corresponding BIS values are input into the long-term and short-term memory network for prediction.The experimental results show that the predicted BIS value and the real BIS value of the model are 0.93 accurate.Compared with multilayer perceptron and time convolution network,the prediction accuracy of the algo-rithm is improved by 17.7%and 12.9%respectively.At the same time,the algorithm predicts that the BIS takes 0.32 s in 30 s,which is 0.2 s more than that of multi-layer perceptron and 2.12 s less than that of time convolution network.
bispectral indexlong short-term memoryEEG signal denoisingfeature selection
万方数据
维普
