A two-channel adaptive filtering approach for recognition of the QRS morphology

An approach for QRS morphology classification is proposed. The cross-correlation structure between two electrocardiogram (ECG) channels is used to characterize different QRS morphologies. The classifier consists of a bank of adaptive filters, one for each QRS morphology; each filter has the QRS samples sequence of an ECG channel as reference input, and attempts to predict the QRS samples sequence of the other channel. A cumulative error and a prefixed threshold are used to decide the class assignment or a new class generation. The classifier was tested on the MIT-BIH database, showing satisfying sensitivity and positive predictive accuracy values. The method can be used to add useful information and to improve the performance of a template matching classifier