Abstract
According to the data released by the Department of Health, Executive Yuan of Taiwan the heart disease was ranked number two among the top ten causes of death in Taiwan. Therefore, people pay more attention to heart disease now. The electrocardiogram (ECG) signal is the most commonly used condition to diagnose the heart disease. For long-term monitoring, the portable measuring instruments have become the mainstream. In addition to focus on the small size and low power of the development of portable measuring instruments, noise reduction is also an important topic. In this thesis, there are two main research topics about ECG signal proposed. One is noise reduction, and the other is R peaks detection. Both of the two algorithms are based on discrete wavelet transform (DWT). Wavelet transform (WT) is popular for signal processing recently which can supply time-frequency analysis. Thus, WT is efficient for analyzing non-stationary signals like ECG signal. Different bases, three thresholding algorithms and two kinds of wavelet transform are used to solve different kinds of noises in order to find the proper method of noise reduction. The signal-to-noise ratio (SNR) is used to evaluate the result. From the simulation results, the Symlets wavelets (sym5) and soft-thresholding are chosen as the wavelet function and thresholding method, respectively. Employ them to do noise correction at the first denoised stage. The second stage is R wave detection. The MIT-BIH arrhythmia database, the sampling rate is 360Hz, is used for example. As for QRS complex characteristics, the frequency range is from 5 to 40 Hz. Thus, we chose to reconstruct the decomposition level 3 to 5, because the frequency range of which is from 5.6 to 45 Hz. Choosing the adaptive threshold and window size is the key point for the result of error rate. Using two thresholds method leads to better performance, compared to using one threshold method. At the last stage, does noise correction again. In terms of R wave positions, the novel method is proposed for eliminating the electromyogram (EMG) signal. MIT-BIH arrhythmia database is utilized to verify our simulation results for R wave detection. The algorithm for R wave detecion has a sensitivity of 99.70% and a positive predictivity of 99.65%. The error rate is 0.65% under all kinds of situation (0.37% if ignoring 3 worst cases). For noise correction, the improvement SNR is achieved at least 9.5dB at SNR 5dB, and most of the improvement SNR are better than other methods at least 1dB. To apply presented algorithms for the portable ECG device, all R peaks can be detected no matter when people walk, run or go up and down stairs below 9km/hr. Thus, the proposed method can identify R wave correctly and reduce the noise effectively, so that the diagnosis of heart disease can be more accurate.