Research Of Quality Assessment Of ECGs Collected Via Mobile Device

A telemedicine ECG monitoring system (TEMS) that can provide early discovery and timely treatment has been widely used in treatments for heart disease. Nevertheless ECGs collected by Mobile terminals of TEMS are often severely corrupted by body movement, breathing, noise, artifact and missing data, which make large errors in the estimation of signal parameters and a high incidence of false alarms in TEMS. These errors and false alarms lead to lack of trust of clinical staff to TEMS so that the real critical alarms are probably neglected, even medical misdiagnosis probably occur. So, there is an essential requirement to assess ECG quality before it is used for clinical applications.This paper tried to provide quality indice (QI) based on the linear and nonlinear features of ECG, and then employed support vector machine to assessment ECG quality. At present, further studies for QI of ECG need to be conducted to improve accuracy of quality assessment and it mainly include the follow aspect. First, expanding feature space. The nonliear analysis of signal including Lyapunov exponent, entropy and Lempel-Ziv (LZ) complexity should be performanced for detecting the nonlinear characters of signal instead of just time-frequency analysis. The nonlinear features extracted from ECG can reflect information of signal in a comprehensive and accurate way because the nonlinear character is a valued and inherent part of ECG. Second, the nonlinear feature extraction algorithms need further improvement. Both of entrony and LZ complexity algrothme are the measure of complexity of signal however the complexities of signal calculated by these algrothm are two different concepts. So these algorithms measure the different information contrained in signal, and are used for ECG quality assessment in diffent way. Although recent proofs have proven that the LZ algorithm is more linked with the irregularity of signal instead of the complexity of signal. However, the LZ values for the random signals usually overlap with the values for the chaos signals, inducing the inaccurate explanations for the LZ algorithms. So much more study is needed for improving the accuracy of these algorithms. Third, increasing the number of QI derived from ECG feature. Most of the current studies often propose a QI derived from an ECG feature. However a QI is insufficient to reflect the feature, even miss it, resulting in no effect for signal analysis. So many QIs should be derived from a feature of ECG for quality assessment because an ECG feature has many different kinds of form. For example baseline wander is a typical feature of ECG, and its form can be detected by not only the highest amplitude voltage of baseline but also the continuous amplitude voltage of baseline in a time window. Given these, the paper analysed the nonlinear feature of signal after extracting time-frequency features, i.e. multi-scale entropy (MSE) and LZ complexity algotithm, and provided application methods and precondition factors. A novel LZ complexity algorithm, namely encoding LZ algorithm was proposed in the paper. The paper provided 19 QIs of linear and nonlinear, and these QIs were employed by support vector machine (SVM) for ECG quality assessment. Many artificial ECGs were used in this study, so a good ECG model is necessary. An important issue that should be considered in the modeling of realistic ECG signals is to model realistic noise sources. So this paper firstly analysed two simple noise source models, and then proposed time varying Autoregressive (AR) models to generate realistic ECG noises which follow the non-stationarity and the spectral shape of real noise. The idiographic work is as follows:1. This paper analysed the time-frequency features including lead-fall, baseline wander, R wave detection and power spectrum, and proposed 6 QIs derived from these features for quality assessment. This study shows the QIs derived from wave form and frequency features can reflect the level of noise contrained in signal, and simple computation. However the generalization ability is poor, and the accuracy of quality assessment is difficult to increase.2. This paper studied MSE and LZ complexity for using in ECG quality assessment, and observed the sensitivity of these algorithms to the levels of noises contrained in signal. Application methods and precondition factors of these algrothms were proposed. This study shows that MSE and LZ complexity can reflect the levels of noises contrained in signal, but its sensitivities to different noises are different. The choice of scale factor of MSE is in a certain range when the algrothm is applied for ECG quality assessment, and ECG needs to be pretreated firstly.3. Because the classical LZ complexity algrothm confuse between the randomness and chaos character of signal. So this paper proposed a novel encoding LZ complexity algorithm. The algorithm transforms the original signal to 8-state symbol sequence, avoiding 2-state sequence miss large useful information, keeping relateively less computation. This algorithm can distinguish the randomness and chaos character of signal and reflect the level of noise in signal.4. The paper proposed 19 QIs derived from the linear and nonlinear characters of ECG collected by mobile device, and use these QIs and SVM for assessing ECG quality. This study shows that the QIs can relatively comprehensively reflect the information of signal. The QIs of nonlinear characters have good effect where the QIs of time-frequency characters cannot assess the quality of some signals.5. Because the modeling of realistic ECG signals is to model realistic noise sources in this study. So this paper proposed a new noise source model based on time vary AR model. The parameters of this model can be trained by using real noises such as the Noise Stress Test Database of the MIT database. Having trained the model, it can be driven by white noise to generate different instances of such noises, with almost identical temporal and spectral characteristics.