Research On Biological Signal Identification And Application Via Deterministic Learning

Human biological signal identification constitutes an important field of research in the last few years.As an important research direction in biometrics and biomedical engineering,biological signal identification aims at employing information science technology to signal collection,extraction,modeling,analysis and classification.Since biological signal is detected from human beings,it can be diverse and complicated due to physiological and psychological reasons.The development of biological signal identification plays an important role in life science research,prevention and treatment of diseases and medical instrument industry.From the viewpoint of information science,the identification of human biological signal belongs to the problem of feature extraction,pattern modeling and pattern recognition.There has been a lot of research on biological signal identification and many beneficial results have been obtained.However,most of the existing researches are transforming the pattern recognition problem into the static pattern modeling and recognition problem.It should be noticed that,most human biological signals are essentially spatialtemporal patterns with temporal trajectory,which are generated by complex nonlinear system.Therefore,if the dynamics within the complex nonlinear system can be accurately identified and used for the dynamical pattern recognition,it will be useful to provide a new powerful way for the human biological signal identification.Based on the deterministic learning theory,this thesis is mainly focused on the modeling of dynamics underlying the spatial-temporal human biological signal.The main contributions of the thesis are as follows:1.We propose a human biological signal identification method via deterministic learning theory.We introduce the scientific principles,applicable conditions and application schemes of the proposed method.Biological signals with recurrent trajectory are essentially spatial-temporal patterns generated by complex nonlinear systems.Deterministic learning theory can be applied to extract(model)the nonlinear dynamics underlying the spatial-temporal human biological signal through radial basis function(RBF)neural networks.The modeling results can be stored as the constant neural network weights,which is useful for the following individual identification and disease classification objectives.The nonlinear dynamics underlying the spatial-temporal human biological signal is expected to provide more-indepth information than the original signal.2.Human gait recognition via deterministic learning theory are investigated in this paper.Human gait belongs to a kind of spatial-temporal motion patterns,leading to the computation complexity problem in the process of human gait modeling and recognition.We achieve accurate gait system dynamics modeling and recognition via dynamical RBF neural networks through deterministic learning theory.Two important kinds of gait features are considered in the experiments on CASIA-B gait database.On basis of our previous work,this paper further investigate the gait recognition through feature fusion and view fusion.On one hand,we propose a robust gait recognition method using multiple views fusion and deterministic learning.Gaits collected under different views are synthesized as a kind of synthesized silhouette images,gait dynamics underlying different individuals' time-varying gait features is effectively modeled by using deterministic learning algorithm.Experimental results show that encouraging recognition accuracy can be achieved.On the other hand,this study focuses on gait features obtained by Kinect and proposes a new model-based gait recognition method by combining deterministic learning theory and data stream of Microsoft Kinect.Deterministic learning theory is employed to capture the gait dynamics underlying Kinect-based gait parameters.Addtionally,we discuss how to eliminate the effect of view angle on the proposed method.Experimental results indicate that encouraging recognition accuracy can be achieved.3.Early detection of myocardial ischemia via deterministic learning theory are investigated in this paper.Conventional electrocardiogram signals can be inaccurate in diagnosing coronary artery disease(CAD),particularly in stable and/or asymptomatic patients.This paper introduces a complementary diagnostic tool to conventional ECG,called Cardiodynamicsgram(CDG).We sought to evaluate the clinical utility of CDG for early CAD detection in suspected patients with CAD presenting with nondiagnostic ECGs in National Center for Cardiovascular Diseases.A total of 421 suspected patients with CAD presenting with nondiagnostic ECG were enrolled.Standard 12-lead ECG and CDG were performed simultaneously before invasive coronary angiography.Diagnostic accuracy of CDG for early CAD detection was assessed with reference to coronary angiography as the gold standard.Diagnostic accuracy of CDG at presentation for CAD was 84.6%,sensitivity 84.7%,and specificity 83.7%.In patients presenting with nondiagnostic ECGs,an abnormal status can be detected early through noninvasive CDG.Further experiments indicate that the combination of CDG and ECG yielded a sensitivity of 92.1%,a specificity of 84.2%,and an accuracy of 91.0%.