Research On Several Algorithms For ECG Signal Auto-analysis

Cardiovascular disease is a major killer of human health and ECG plays an important role in diagnosis of these diseases. With the widespread using of ECG machines, the scholars have done extensive research on ECG auto-analysis algorithms. On this basis, the ECG auto-analysis algorithms are studied in this paper in order to meet the requirements of remote wireless multi-parameter monitoring system. Also, the system’s ECG auto-analysis module is designed and programed. This study makes up some deficiencies of the existing algorithms and software system.ECG auto-analysis techniques include ECG preprocessing, wave detection and characteristic point’s location, arrhythmia discrimination. In this paper, these three areas are studied, and some new algorithms are proposed. These new algorithms are used in the ECG auto-analysis module and get a good result that reaching the national criterion of detection. The main contents and achievements are as follows:1. For ECG preprocessing, a simple integer coefficients filter is designed for the requirements of the ECG real-time preprocessing, but some high-frequency signal is eliminated. In order to solving this problem, a new ECG signal preprocessing algorithm with median, mean filter and wavelet transform is proposed and it’s proved to be an excellent preprocessing algorithm by some experiments.2. For wave detection and characteristic point’s location, the QRS detection algorithms are studied in this paper. For real-time detection, a new algorithm is proposed which using filter, multi-order differential and Shannon energy transformation to process the ECG and using improved adaptive threshold method to detect the QRS waves. This method improves the accuracy of real-time QRS detection. For non-real-time detection, a new QRS detection algorithm based on equivalent wavelet packet filter is proposed. This method uses multiple filters to detect the QRS and comprehensive the results in the last. It decreases the QRS detection’s error rate.3. For arrhythmia discrimination, at first, waveform characteristics are selected for arrhythmia classification. There are ten characteristics are selected, such as: the different degree between adjacent RR intervals, RR interval, QRS wave width, Q wave amplitude and so on. Because of the deep belief networks have get good results inmultiple applications, so the author use the deep belief networks for arrhythmia discrimination, and the algorithm classification of arrhythmia with deep belief networks is proposed. Features that extracted by DBNs and RR intervals are used for arrhythmia discrimination and it improves the accuracy of arrhythmia discrimination.4. ECG auto-analysis module of remote wireless multi-parameter monitoring system is designed and programed. The author maps functional requirements modules to software modules, and programs them. At finally the module is verified by MLII data in the MIT-BIH database.