Template-based compression of ECG signals

Abstract

A new approach for ECG data compression is proposed in this paper. Using a nonlinear least squares optimization procedure, the approach employs an algorithm based on template model fitting. Only 12 parameters are required to fully represent the ECG signal without diagnostic information loss. The effectiveness of our ECG compression technique is described in terms of high compression ratios, relatively low distortion values of less than 9%, and a low computational cost, thus demonstrating the beneficial use of our technique for ECG data storage and online transmission. Comparisons with other recent compression methods in the literature have shown that our method performs better.

A Noise Tolerant Method for ECG Signals Feature Extraction and Noise Reduction

“…Fitting is performed using a nonlinear least squares optimization. Detailed information regarding the algorithm can be found in [9].…”

“…In a previous work [9] we proposed to model the ECG signal with tabulated templates, which include the complete shape information of its waves. To detect and delineate the P-wave, QRS-complex and T-wave, three different templates are extracted from the patient-specific ECG in a calibration or training stage.…”

“…The algorithm may converge to a local minimum which gives an unacceptable match of the sampled data, if a poor initial guess is used. In a noise free environment, we showed in [9] that the parameter set computed after fitting a sub-wave in the previous cycle provided a sufficiently reliable initial guess for the parameter estimation of the same sub-wave. We exploited thereby the particular property of the ECG signal, which has a very high degree of similarity from cycle to cycle.…”

“…However, noisy ECG signal do not dispose of this distinct feature, since signal-overlapping noise is a non-stationary signal with time-varying characteristics, which can totally mask the signal information in the electrocardiogram. Moreover, maximum searching was performed in [9] in the learning phase to provide initial values for heights and locations in time of the sub-waves. Nevertheless, many noise spikes may attain up to 500 percent of the ECG signal's peak-to-peak amplitude (electrode motion artifact).…”

“…In this paper, we present an improvement in our recently proposed template model fitting technique [9], originally used for ECG signal compression, to allow for a low sensitivity to noise omitting the need for excessive signal filtering in a preprocessing stage. Depending on collected training data about the different ECG component waves, a greater performance in their detection is achieved in presence of noise.…”