Research On Feature Extraction And Data Compression Of Electrocardiogram Based On Wavelet Packet

Cardiac diseases have been thought as a kind of major diseases which became more and more serious to human's health. On one hand, patients expect for accurate results of heart examination on time, on the other hand, patients need 24-hours intensive care. So some valid and effective technique was demanded to deal with the ECG while compression arithmetic with high performance was also required.In this thesis, mathematic morphologic is used to eliminate the noise firstly from the raw electrocardiogram. Then taking advantage of the multi-resolution property of wavelet transform, we bring forward a new R-wave detection method based on fuzzy measure of the mergence information on multi-scales wavelet transform of ECG signals. After the locating of R-wave, we study the characteristics of Q-wave and S-wave distributed over wavelet subbands. And then by the use of the convolution-eleminating property of Marr wavelet, the characteristics of Q-wave and S-wave are reflected on the third scale detail. So QRS waves are removed from the raw electrocardiogram signals and the characteristics of P-wave and T-wave are obviously strengthened on the remainder. At last, all character points extracted are used to construct the eigenvector to reflect full information of the ECG.In the aspect of ECG data compression, we first try the arithmetic of zero-tree wavelet to encode the ECG and prove it valid in practice. In order to gain better compression rate, we switch the coding method to wavelet packet transformation. After comparing the cost function results, we find out the best wavelet packet basis for ECG compression. During this process, we solve the problem of parents conflict which is brought by the features of wavelet packet. Then considering the time-domain relativity of ECG, we introduce the wavelet packet decomposition on signal from frequency-domain to time-domain. The ECG eigenvector is used for time-domain decomposition, and the boundary error problem in this process is well solved. By practice, the zero-tree wavelet packet coding method of time-frequency double tree makes good use of the features of the ECG and more zero-tree appears in the code stream. It will be a great support for subsequent Run-Length coding.