Study On Dynamic Electrocardiograph Signal Processing Method Of A Wearable Physiological Parameters Monitoring System

Recently, many factors affect human health, such as acceleration of life pace, increasement of work pressure, and decrease of physical activities, et al. Meanwhile, people's awareness in health is growing, so the medical equipements should be developed in varity styles, not only the larg-scale and complex equipements in hospital, but there is a need to develop small, portable and wearable monitoring devices for families, individuals, and community, therefore, it's significance and valuable to study on wearable medical devices with monitoring and early-warning functions.A wearable monitoring physiological parameters system is developed by Jilin University, from a key project'wearable physiological parameters Non-invasive continuous monitoring device (20070333)', which is sponsored by Jilin Province Science and Technology Department. This device can do the long-time, real-time and non-invasive measurement of physiological temperature, blood pressure, oxygen values, and dynamic electrocardiogram (DCG). DCG automatic analysis is particularly important among these parameters, so it's crutial to study on DCG analysis.DCG de-noising, waveform detection and assistant diagnosis are studied, respectively, and due to the limitations of current processing methods in DCG automatic domain, this study proposed some applicable algorithms for our system. Firstly, determined to use wavelet analysis method for DCG de-noising and waveform detection by anlysis the characteristics of DCG signals.Secondly, studied the wavelet thresholding theory systemtically, and summarized five key factors which influence the de-noising effect: wavelet transform(WT) techniques, wavelet basis, decomposition levels, shrinkage functions and thresholds, and applied this theory for ECG De-noising.For WT techniques, using classical WT, wavelet packet, lifting wavelet and stationary wavelet transform (SWT), respectively, through a comprehensive comparison, determined using SWT;For wavelet basis, through analysis the mathmetical characteristics of eight kinds of wavelet bases, determined the suitable wavelet bases for ECG de-noising theoretically, and then through the experiments, determined that the optimal basis for de-noising ECG signals is Symmlet10 by quantitative assessment;For wavelet decomposition levels (scales), considered two aspects, one is separation of the valuable information from the noise, the other is reduction of the reconstruction error, and determined the optimal decomposition level is five, in addition, using the changing SNR ECG signals to test its stability;For shrinkage functions, compared with Soft, Hard, Firm and Garrot functions, each function has its own advantages and disadvantages. However, Hard shrinkage function is the most suitable one for keeping the singular point of the original signal; For thresholds, compared with the commonly used thresholds, such as Minimaxi, Universal, SURE, and Hybrid, in addtion, due to the GCV estimation doesn't need any priori knowledge of noise energy, and Ebayes estimation is good to access the minimum mean square error, that is, the threshold is close to the optimal threshold.Through the comprehensive analysis and comparison of the above methods, proposed using SWT with 5-scale decomposition, Sym10 wavelet basis, and Hard with Ebayes thresholding method for de-noising ECG signals, and through testing, the algorithm can reduce muscle noise, power line interference and electrode motion artifacts, but can not remove the baseline wander (BW).Thirdly, analysis of the bandwidth of BW, and comparison of two efficient methods to remove BW: using median filtering and using high-scale wavelet decomposition estimation (WTSE). The results showed that the WTSE method is better than median filtering method by comparison of SNR improvement and singular points maintaining.If using high-scale decomposition in single thresholding method, directly, it can also achieve the purpose of BW removal through eliminating the high scale coefficients, but, on the one hand, following the increase of decomposition scales, except the BW, other noise removal will be affected, on the other hand, the execute time will increase.Therefore, proposed a combined algorithm, using WTSE method to remove BW, and then using SWT threhsolding algorithm to reduce other noise. The de-noising results proved that the combined algorithm is better than the single de-noising methods, it not only removes noise effectively, but better protects the feature points.Fourthly, QRS complex detection is crucial in ECG automatic analysis, including the R-wave peak and the QRS onset and offset detection. There are three typical R-wave detection algorithms based on differential, digital filter and wavelet transform, respectively. No matter what kind of algorithm was used, all need the appropriate threshold, maybe due to the different selection, even though the same method, can lead to the different detection rate. Through analysis the threshold selection rules, adopted a grouped calculation strategy to select threshold, then decrease the false detection rate because of the sudden waveform variations. Using the typical data from MIT-BIH Arrhythmia database to assess the effectiveness, the anti-interference ability, and the execute time, the above algorithms have their own advantages and disadvantages. Wavelet modulus maximum (WMM) method obtained the highest detection accuracy and the best anti-interference ability, but, it is too complicated, and the execute time need to be improved. For these limitations, this study improved the traditional WMM method, proposed a fast algorithm based on WMM, which adopted two new thresholds,and eliminated the redundant steps from the traditional WMM algorithm. In addition, using MIT-BIH Arrhythmia database to assess the effectiveness and the execute time, the results proved that the proposed algorithm not only obtained the better detection accuracy, but improved the execute time. Based on this algorithm, in scale 2, implemented the QRS complex onset and offset dectection.Fifthly, P wave and T wave have the following properties, such as smaller amplitude, lower frequency, and morphological diversity, therefore, the existing detection algorithms have great limitations, meanwhile, due to the lack of a unified, open standard evaluation database, so far, P wave and T wave detection are still difficult in the automatic analysis of DCG. Based on the targeted QRS position, according to the frequency characteristics of P wave and T wave, selected the appropriate scale, threshold and time window, detected the peak, the onset and offset of the P-wave and T-wave, respectively.Finally, designed the structure diagram of the wearable physiological parameters monitoring system, the system includes three main parts: the wearable device for monitoring physical parameters, the user terminal, and the supporting analysis and diagnosis system. According to the DCG diagnostic criteria, listed the mathmetical presentations of the common arrhythmia classification rules, compared the calculated ECG parameters with the above rules, and realized the auto classification of normal electrocardiogram and common arrhythmia beats. This tool can assistant the clinicians to diagnose.