摘要
临床上常用的平均脑干听觉诱发电位(Brainstem Auditory Evoked Potential,BAEP)无法描述脑干功能的动态特性,从背景噪声中单次或少次动态提取的BAEP才是反映脑干功能的理想信号。径向基函数神经网络(RadialBasis Function Neural Network,RBFNN)已被用于BAEP的非线性动态提取,但是对于“淹没”在噪声中的信噪比很小的BAEP提取效果不好。本研究用移动窗口平均(Moving Window Average,MWA)先对含噪声的BAEP进行动态少次平均提高信噪比,然后再用RBFNN进行BAEP的非线性提取,在保留了绝大部分BAEP动态信息的前提下改善了RBFNN的提取性能。为了验证方法的可行性,构建了信噪比为-25dB的仿真BAEP-噪声序列,经MWA和RBFNN动态提取后相对均方误差约为19%,比仅用RBFNN时误差降低了6%左右。将上述方法用于实际测取的BAEP,可以得到每个子波形和平均BAEP波形波幅趋势大体相同的动态序列,这个BAEP动态序列为应用非线性动力学研究脑干功能动态特性打下了基础。
The common-used averaged brainstem auditory evoked potential （BAEP） can not describe the dynamic characteristics of brainstem, while single-trial BAEP extracted from background noises would be the ideal signal. Radial basis function neural network （RBFNN） has been used to extract BAEP nonlinearly and dynamically. However, its extraction performance is typically poor for the BAEP, which is emerged in the noises and has a very low signal-to-noise ratio （SNR）. In this paper, moving window averaging （MWA） is employed as a preprocessing method to raise the SNR in advance, and then RBFNN is used in the dynamic extraction of BAEP. The performance of RBFNN is improved and most useful dynamic information of BAEP is also maintained. Simulated BAEP-noise series with an SNR set to -25dB is constructed to evaluate the efficiency and feasibility of this combined method. The relative mean square error between the extracted BAEP and the original simulated BAEP is 19%, decreased 6% than only using RBFNN. When we use the MWA and RBFNN to extract the real BAEP recorded from a patient with brainstem functional disorder, a similar waveform tendency with the 300-trial averaged BAEP is obtained. This dynamic BAEP series is a foundation to the nonlinear dynamic analysis of brainstem＇ s function.
出处
《中国生物医学工程学报》
EI
CAS
CSCD
北大核心
2005年第6期681-684,共4页
Chinese Journal of Biomedical Engineering
基金
国家自然科学基金资助项目(60174041)
天津市应用基础研究基金资助项目(043800311)
关键词
脑干听觉诱发电位
动态提取
移动窗口平均
径向基函数神经网络
brainstem auditory evoked potential
dynamic extraction
moving window average
radial basis function neural network
