A model-based approach to QRS delineation.

QRS delineation is considered to be a problem of finding changes in the spectral behavior of an ECG. Based on statistical models for the "low-frequency" and "high-frequency" segments of the ECG, the points for QRS onset and end are obtained from a maximum-likelihood procedure. This procedure is based on the properties of the prediction errors obtained from two different Kalman filters. The accuracy of the delineation scheme is studied on a QRS complex material in terms of agreement with manually obtained delineations. The material was independently delineated by three different readers. The sensitivity to noise is studied by adding white noise to the QRS complexes. The performance is compared to those of envelope-based delineation and template waveform delineation.