Research On The Extraction And Analysis Method Of Brainstem Auditory Evoked Potential

BAEP (Brainstem Auditory Evoked Potential) is a sort of CEP (Cognitive Evoked Potentials). The time and frequency information BAEP contained is significant to evaluate auditory function and diagnose neural diseases. At present, averaging methods are always used to obtain BAEP signals. But the averaging methods may have lost a mass of dynamic information. So these BAEP signals could not describe the dynamic characteristic of brainstem. Thus, the single-trial (or a few) extraction of BAEP is the focus and nodus in recent years.Many researchers extracted CEP using new mathematics methods. But majority of those methods were based on the assumption that the BAEP signal and noise were additional related. It's irrational and deficient to some extent. Along with the ceaseless progress of mathematics and computer science, some researchers brought WT (Wavelet Transform) into nonlinear extraction of BAEP. This method discarded anterior assumption and also obtained favorable effect in higher SNR condition. But in low SNR condition, the WT method have certain defects because of the characteristics of WT.The noise contained in BAEP signal has certain randomicity and disproportion. In order to enhance SNR and to be propitious to WT, after several trials averaging, the author brings RST (Rough Sets Theory) into the research of BAEP. According to the basic theory of RST, we class the signal sequence using characters of BAEP. Then preprocessing is implemented according to different parts to obtain a preprocessed sequence whose SNR would be higher. Then we decompose, analyze, de-noise and reconstruct it using WT. Finally we could obtain the extraction result of BAEP. The RST classification method and WT are nonlinear mathematical methods. So the research of the thesis is nonlinear dynamic extraction of BAEP. Major jobs of the thesis would be described as follows.(1) First, we add and average the BAEP sequence with noise 10 times, and then analyze it dynamically and identify its characteristics. We classify the sequence into 3 different parts according to the parameter threshold. They are N₁, N₂ and N₃.Then we put up relevant operations according to different parts in order to complete nonlinear preliminary filtering, where we search the optimal classification threshold using GA (Genetic Algorithm). So the classification preprocessing based on RST would be more scientific.(2) Afterwards we put up multi-resolution analysis on noising BAEP signal after preprocessing and decompose it into detail and approximation sequences at 7 levels using DWT (Discrete Wavelet Transform). We reserve all the detail coefficients and the highest level approximation coefficients, then analyze the relativity of detail coefficients, and check their correlation degree. We will reserve well correlated ones and eliminate the rest to de-noise and then reconstruct the BAEP signal using the highest level coefficients and detail coefficients reserved. So the reconstructed signal is the extraction result.(3) Experiments results contrasting and analysis: the author designs a series of simulation experiments. Compare the extraction result using RST and WT method with other extraction methods such as additional averaging method and ordinary WT method. We weigh for the superiority and insufficiency of the thesis's method., and educe the conclusion. The algorithm of the thesis approves to be effective and steady by experiments. Besides satisfactory extraction effect could be obtained in low SNR condition.Finally, the author summarizes the whole thesis and prospect for the subsequent research direction and major jobs in the field.