基于多特征筛选的双频指数预测算法

Bispectral Index Prediction Algorithm Based on Multi Feature Screening

双频指数(Bispectral Index,BIS)是根据脑电信号得到的衍生指数,是目前镇静深度监测应用最广泛的指标。但是根据脑电信号计算得到BIS存在20 s~30 s的更新延迟,影响麻醉医师对患者镇静状态的判断,增大患者术中知晓的风险,影响患者术后恢复。论文结合脑电信号子参数、历史BIS和患者术中生命体征预测BIS值,为麻醉医师提供患者最新的镇静状态的变化趋势,进而方便其准确地调整麻醉用药。论文构造基于多特征筛选的BIS预测模型。首先通过经验模态分解算法对脑电信号噪声滤除,并对处理后的脑电信号进行子参数提取。然后采用随机森林算法筛选镇静特征参数,得到与BIS相关度最高的5个特征。最后将筛选的特征和对应的BIS值输入长短期记忆网络中进行预测。实验结果表明,该模型预测的BIS值与患者真实BIS值拟准确度达到0.93。与多层感知器和时间卷积网络进行比较,该算法预测的准确率分别提升了17.7%和12.9%。同时该算法预测30 s内BIS耗时0.32 s,比多层感知器多用了0.2 s,比时间卷积网络节省了2.12 s。

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.