The Research Of ECG Signals Automatic Analysis Based On Wavelet And Neural Network

The methods of signal preprocessing, waveforms detection, classification and diagnosis for dynamic electrocardiogram (ECG) are studied in this dissertation. The algorithm of automatic ECG analyzing is also accomplished.Firstly, we analyze the interferences of ECG,and presented the method of interferences rejection, such as power line interference, muscle electricity and baseline wander; Then a synthetic denoising algorithm of ECG signal based on wavelet transform is provided, and the wavelet decomposing and reconstructing method and wavelet thresholding method are studied in this dissertation. After analysis the characteristic of the methods applied to ECG denoising,we combined them into a complete algorithm. This method can effectively suppress the 50Hz interference, baseline drift and myoelectric interference; at the same time, it can remain the geometrical characteristics of the original ECG signal perfectly.After that, difference threshold algorithm for detecting QRS wave is studied. For the limitation of the difference threshold, a precise real-time QRS detection algorithm based on 2-order B-Spline wavelet transforms is provided. Using the relationship between wavelet transforms and signal singularity point, we detect the peak of R wave in scale 23,the begin and the end of QRS wave in scale 21. Upon part of data of MIT/BIH arrhythmia database, the algorithm correctly detects up 99.64%. It improves the accuracy of the detection and the orientation of QRS complex.At last,on the bases of standards that used in most present cardiogram monitors, a automatic diagnosis technology—artificial Neural Network (ANN) model is mainly discussed in this paper. This method employed the characters of the QRS waves to realize the detection, including R-R distance, amplitude of R waves, and distance of QRS waves. The designation and training of the Neural Network is discussed in detail. The validities of the algorithm were certified by applying it to part of MIT/BIH standard arrhythmia database. The experimental results confirm that the network which based on L-M algorithm can detect APC,PVC and arrhythmia accurately. Also the proposed method is capability of good robustness and efficiency.