中潜伏期听觉诱发电位的提取与识别研究

Study on extraction and recognition of mid-latency auditory evoked potentials

目的：通过小波变换模极大值和人工神经网络对中潜伏期听觉诱发电位进行提取与识别,判断患者处于的麻醉深度。方法：利用信号和噪声在分解尺度上的不同特性来滤除中潜伏期听觉诱发电位中的强噪声成分。经过模极大值处理后,小波系数在每个尺度上只剩下有限个模极大值点。通过Mallat交替投影算法,根据这些模极大值点重构出中潜伏期听觉诱发电位。在此基础上,根据中潜伏期听觉诱发电位与麻醉深度之间的对应关系,建立适用于麻醉深度监测的BP（back propagation）神经网络模型。结果：通过小波变换模极大值提取后,中潜伏期听觉诱发电位的信噪比由-15.43~0 dB提升到10.05~20.18 dB,基本恢复了原始信号;麻醉深度监测的BP神经网络模型测试集精确度分别为90%和100%。约登指数和Kappa值均为0.900 0,大于0.75。该模型对中潜伏期听觉诱发电位具有良好的识别能力。结论：小波变换模极大值提取可以有效地去除信号中的强噪成分,还原信号;麻醉深度监测的BP神经网络模型对中潜伏期听觉诱发电位与麻醉深度之间的关系具有良好的识别性。

Objective To recognize and extract the mid-latency auditory evoked potentials （MLAEP） with wavelet transform modulus maxima method and artificial neutral network in order to judge the depth of anesthesia in the patients exactly. The difference of the signals and noises in decomposition scale was used to filter out the strong noises in MLAEP. After modulus maxima filtering, the wavelet coefficientat only remained limited modulus maximum points at all scales. Through Mallat alternating projection method, the real MLAEP was refactored by these modulus maximum points. According to the relationship between MLAEP and the depth of anesthesia, BP neural network model was established, which was used to monitor anesthesia depth on this basis. Results After wavelet transform modulus maxima extraction, the signal of MLAEP promoted from -15.43-0 dB to 10.05-20.18 dB, and the accuracy of the test set of BP neural network model for monitoring anesthesia depth were 90% and 100% respectively. Youden index and Kappa value were both 0.900 0, which were more than 0.75. The model behaved well in the recognition of MLAEP. Conclusion Wavelet modulus maxima extraction can effectively remove the strong noise ingredients and restore signal; BP neural network shows its advantage in recognizing the relationship between MLAEP and anesthesia depth.