Abstract
Biomedical systems have expanded markedly in recent years, spreading to most aspects of human life. In promoting rapid advancements in biological science, which have led to the creation of novel electrical circuits and signal processing methods for developing tools for diagnosing and treating human diseases, several researchers in biomedical engineering have developed new tools for specific medical conditions. Electronic instruments provide an interface between biology and electronics. Such interfaces enable quantifying and characterizing biological phenomena, which can then be investigated to elucidate biological processes. A typical interface comprises a sensor or electrode for detecting biological parameters, the signals of which can then be amplified and converted into the digital domain. These digital data can be processed by hardware or transferred to a personal computer for long-term storage and more accurate signal processing. Depending on the application requirements, the data can be transferred through a wired or wireless link. For instance, we propose a real-time closed-loop neurostimulation system and off-line bioinformatics data-analysis method. A neurostimulation system can be measured on the basis of neural phase synchrony by using real-time electrical processing. In addition, the closed-loop phase synchrony detection requires 70.1% electrical energy to be delivered to the deep brain, which can reduce the system power consumption and extend the battery life. The electrical signals output from an analog circuit are processed by signal processing through a data acquisition unit, signal conditioning block, and digital signal processing program. In this dissertation, the digital signal processing program features a short-time Fourier transform, phase coherence analysis, linear analysis of heart rate variability, multiscale entropy analysis, and detrended fluctuation analysis. The short-time Fourier transform and phase coherence analysis were used to determine the bioelectrical activity of an object. Analysis of linear heart rate variability, which represents one of the most promising autonomic activity markers, was performed in the time and frequency domains. In addition to conducting a linear analysis, we also propose two innovative analysis methods derived from nonlinear and nonstationary processes. The first method is a multiscale entropy measurement that provides the regularity pattern of a time series by analyzing the interplay between quantitative connotations and the correlations among individual electrical signals. The second method, detrended fluctuation analysis, is employed to evaluate the fractal correlations causing heart rate fluctuations that originate from the interactive regulatory mechanisms. Moreover, most processed biometric signals can be statistically analyzed to estimate effects and predict outcomes. For example, this dissertation presents a method requiring purely biometric values derived from a regression model and machine learning technique, which are referred to as change-score analysis and generalized additive models (GAMs), to evaluate relaxing states. Furthermore, considering the practical applications for wearable sensor nodes, EEG data obtained before stress were not used in the change-score analysis. A biomedical system comprising electronic instruments, signal processing modules, and bioinformatics is discussed in this dissertation. The analog front end and digital circuits for detecting and acquiring electrical signals were developed using integrated circuits. The guidelines for developing the signal processing algorithms were low complexity, short latency, high sensitivity, and accurate characterization. Additionally, a microprocessor was used to ensure that the electronic algorithm design is flexible and adaptable. Finally, a statistical analysis method for estimating the correlations between biometric values and clinical informatics is presented. Even when the parameters are irregular and change in the external environment, the variable selection method can accurately extract the signals of interest. In this dissertation, the experimental results of the biomedical system were compared with various performance criteria. The comparison showed that the proposed real-time closed-loop neurostimulation system attains the required low energy for stimulating deep brain activity. In the bioinformatics data analysis, the change-score analysis and GAMs respectively achieved an overall accuracy of 80.7% and 86.5% in recognizing changes in brain waves.