Font Size: a A A

The Research Of Dynamic Single Extraction Algorithm Of Brainstem Auditory Evoked Potential Signal Characteristics

Posted on:2013-06-03Degree:MasterType:Thesis
Country:ChinaCandidate:H CengFull Text:PDF
GTID:2248330371481054Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
Brainstem auditory evoked potentials (BAEP) are the potential changes recorded on the scalp from the cochlea to the brainstem auditory neural pathways in about10ms after stimulation of specific sounds in the auditory nervous system, with the features of weak amplitude, strong noise interference and low signal to noise ratio. As a safe and non-invasive, simple and fast, real-time monitoring method, the BAEP check is widely used in hearing tests, neurosurgery disease diagnosis, intraoperative monitoring and the assessment of mental illness in clinical practice. The most widely used method to extract BAEP is the averaging method, which not only might lose dynamic information, but also make the subjects tired due to the excessive repeated stimulations required. To overcome the drawbacks of averaging method, fewer or single trial extraction of BAEP signal has become the main direction of current research.This thesis proposes a dynamic signal trial BAEP extraction method by combing signal spectral analysis, conventional wavelet transform, improved translation invariant wavelet thresholding denoising and radial basis function neural network (RBFNN). The application of wavelet transform and RBFNN to extract BAEP can drop the unreasonable assumptions made in the averaging method, such as the assumptions of additive relationship and normal distribution of the noise. However, due to the limitations of wavelet transform and RBFNN, these two methods perform poorly in the case of the low signal-to-noise ratio (SNR). Compared to the conventional wavelet thresholding denoising method, the translation invariant wavelet thresholding denoising method performs better in the case of low SNR. This thesis proposes an improved threshold calculation method based on the original one. The new threshold calculation method selects the length of the decomposition coefficients of each wavelet layer as the threshold calculation criterion; whereas the original method uses the length of the selected signal to calculate the threshold. The new method can better reflect the feature of each layer of the decomposed signal. Simulation results show that the improved method of noise cancellation outperforms the original method and the conventional wavelet thresholding denoising method.In this thesis, the main work on dynamic single trial extract of BAEP consists of three steps. Making use of the spectral characteristics of recorded single trial BAEP signal, Step1decomposes the signal by conventional multi-resolution wavelet method, and set the wavelet coefficients outside the frequency band (100~3000Hz) to zero when reconstructing the signal. The purpose of Step1is to remove the power-line interference (50Hz), the majority of the spontaneous EEG (0.5~100Hz) and the high frequency interference signals. Step2further processes the output of the first step using the improved transform invariant thresholding wavelet denoising method to enhance the SNR. Step3achieves the non-linear approximation of the target signal using an appropriate RBFNN. The results of clinical trials show that, after enhancing the SNR with the conventional wavelet transform filtering and improved translation invariant thresholding wavelet denoising method, the designed RBFNN with a convergence rate of0.00002and30nodes in the hidden layer needs only350iterations to achieve the same approximation effect as that of1500averaging. The shape of the extracted signal is much smoother, exhibiting five peaks whose positions relate well to the original ones.In the simulation design, the thesis constructs the simulation signals by a mixture of colored noises and real clinical BAEP signals extracted based on the EEG model. Compared to the conventional normal assumption, the simulated signal is more realistic in the aspect of reflecting the non-stationarity and non-linearity characteristics of the BAEP signal, thus can better verify the validity of the proposed algorithm.The proposed method is an exploratory study of dynamic extraction in a single BAEP signal. Compared with the averaging method, the new method can better retain the dynamic information of the target signal so that it can be more accurate to guide clinical practice. Combining the conventional wavelet multi-resolution decomposition with spectral characteristics can achieve precise filtering of noisy signals. The improved transform invariant wavelet thresholding denoising method proposed in this thesis can on its own achieve better noise cancellation with an SNR smaller than-10dB. In summary, the proposed method has certain reference value for the extraction of brain evoked potentials.
Keywords/Search Tags:Brainstem Auditory Evoked Potential, BAEP, Single-Trial Extraction, Wavelet, Radial Basis Function Neural Network, RBFNN
PDF Full Text Request
Related items