Time-varying Frequency And Statistical Analysis Methods For Biomedical Signals

Biomedical signals generated by complex mechanisms are commonly with weak amplitude,strong noise and randomness.This thesis reviews several conventional signal processing methods.For time domain analysis,the filters,autocorrelation functions and power spectral density are described.A number of time-frequency analysis methods are also included,such as short-time Fourier transform,wavelet analysis,Wigner-Ville distribution,matching pursuit,and ensemble empirical mode decomposition.In addition,we propose a novel time-varying statistical analysis method for non-stationary biomedical signal processing,which can effectively com-pute the local signal statistical characteristics for further pattern analysis.For the time-varying frequency analysis,our major contribution is to combine the ensemble empirical mode decomposition(EEMD)method with appropriate pa-rameter settings with the prior knowledge,for processing the local field potential(LFP)signals of mice.Decomposition of LFP signals with different oscillatory rhythm is useful for analysis of various neuronal activities in mice.In the present work,we first removed the power-line interference with high signal fidelity by using a notch filter with infinite impulse response.Next,we applied the EEMD method to separate the LFP signal into low-frequency,Delta,Theta,Beta,Gamma,Rip-ple,and high-frequency oscillations.Then,normalized autocorrelation functions of the resting respiratory signal and the reconstructed Delta oscillations were com-puted to estimate the corresponding power spectral densities by using the Fourier transform.The results of LFP signal decomposition and oscillatory rhythm re-construction demonstrated the effectiveness of the EEMD analysis method.The coherence analysis results indicated that the primary periodicity peak of the Delta LFP component is definitely linked to that of resting respiration in an awake mouse.For the time-varying statistical analysis,we established the time-varying sta-tistical model and calculated several time-varying statistical characteristics for com-mon biomedical signals.In the voice signal processing applications,detection and assessment of dysphonia is useful for the diagnosis of Parkinson’s disease.In this thesis,we first computed the feature TSF by using time statistical snalysis.Then,we study the linear correlations between 22 voice parameters of fundamental frequency variability,amplitude variations,and nonlinear measures.The highly-correlated vocal parameters are combined by using the linear discriminant analysis method.Based on the probability density functions estimated by the Parzen-window tech-nique,we propose an inter-class probability risk(ICPR)method to select the vocal parameters with small ICPR values as dominant features,and compare with the modified Kullback-Leibler divergence(MKLD)feature selection approach.The ex-perimental results show that the generalized logistic regression analysis(GLRA),support vector machine(SVM),and Bagging ensemble algorithm input with the ICPR features can provide better classification results than the same classifiers with the MKLD selected features.The SVM is much better at distinguishing normal vo-cal patterns with a specificity of 0.8542.Among the three classification methods,the Bagging ensemble algorithm with ICPR features can identify 90.77%vocal pat-terns,with the highest sensitivity of 0.9796,and largest area value of 0.9558 under the receiver operating characteristic curve.The classification results demonstrate the effectiveness of our feature selection and pattern analysis methods for dysphonic voice detection and measurement.