Model-based filtering, compression and classification of the ECG

Extensions are presented to a previously described realistic nonlinear model of the electrocardiogram to account for T-wave asymmetry. By fitting the parameters of this model using a nonlinear optimization, we demonstrate that an arbitrary ECG can be modeled and consequently in-band noise can be completely removed. We also show that the fitting procedure effects a compression at a rate of ( : ) per beat or ( : ), where is the reciprocal heart rate, is the sampling frequency, and is the number of symmetric features (or turning points) and the number of asymmetric features used to fit the beat morphology. Performance tests show that the algorithm can run in real time on a modern desktop PC. Finally we demonstrate that by clustering the parameters, waveform classification is possible.