Detection of cardiac late potentials in nonstationary noise.

An 'instantaneous' optimal filter is presented for improving the signal-to-noise ratio in averaged electrocardiograms (ECGs). The filter design is based on a simple model that accounts for correlation across the ensemble of beats but ignores correlation in time. Another property of the model is that the noise level is allowed to change from beat to beat. Simulation results indicate that an improved performance is achieved by the new filter when compared to a filter with a similar structure but designed for processing beats with a constant noise level. Several ECG examples are included to demonstrate the performance of the filter for detecting late potentials.