用脑电互信息及复杂度预测异氟醚麻醉时患者对切皮刺激的反应

PREDICTION OF RESPONSE TO INCISION USING THE MUTUAL INFORMATION AND COMPLEXITY OF ELECTROENCEPHALOGRAMS DURING ANAESTHESIA

自从麻醉应用于临床以来 ,麻醉深度的可靠监测是十分必要的。但到目前 ,尚没有一个公认可靠准确的方法。本文提出一种麻醉深度监测的新方法 ,即用脑电的互信息序列及其复杂度分析来反应异氟醚麻醉条件下患者的麻醉情况。首先计算出四导脑电的互信息时间序列 ,然后计算该序列的复杂性测度 ,借助于神经网络可实现用脑电来监测麻醉深度。神经网络的输入是复杂度值和对应的MAC水平 ,输出即是麻醉深度状况的结果。从 98个自愿患者进行的实验中得到 98个不同程度异氟醚麻醉时切皮前脑电片断 ,同时监测血液动力学参数和患者的呼吸模式。切皮后 ,仔细观察每个患者两分钟 ,以检查患者对切皮的反应 ,把有反应时的脑电标上 0 .0 ,无反应时的脑电标上 1.0。训练和测试神经网络用“去掉一个”方法。从患者对切皮的反应和神经网络的输出结果可检测系统的预测情况。实验表明 ,系统对切皮后患者反应的平均正确判断率为 91.84 % ,该方法比传统脑电分析方法如边缘频率法、中心频率法、双谱分析法有更高的准确性。另外 ,该方法计算时间短 ,适合临床实时使用。

The need for a reliable method of estimating depth of anaesthesia has existed since the introduction of anesthesia. This paper presents a new approach to predict response during isoflurane anaesthesia by using mutual infomation time series of electroencephalograms and their complexity analysis. The mutual infomation time series between four leads were first computed using the EEG time series. The Lempel-Ziv complexity measures, C(n)s, were extracted from the mutual infomation time series by complexity analysis. Prediction was made by means of neural network(ANN). The input to the neural network was the C(n) values and the MAC level, the output was results of the prediction. From 98 consenting patient experiments, 98 distinct EEG recordings were collected prior to incision during isoflurane anaesthesia of different levels. Hemodynamic variables and respiration pattern were also monitored. After skin incision, each patient was observed carefully for 2 min to detect purposeful responses and then the EEG was labelled as 0.0 for responder or as 1.0 for non - responder. Training and testing the ANN used the drop one person method, response prediction was tested by monitoring the response to incision and the result given by the ANN. The system was able to correctly classify subsequent response with an average accuracy of 91.84%. The results showed that the method has a better performance than others, such as spectral edge frequency, median frequency, and bispectral analysis. This method is computationally fast and acceptable real-time clinical performance could be obtained.