QRS pattern recognition using a simple clustering approach for continuous data

This Paper describes a clustering approach to be used for incoming data under computational constraints at an early stage of the signal processing chain. The algorithm is evaluated on the MIT-BIH Arrhythmia Database (MIT) and the European STT-Database (EDB) using a pseudo classification method to estimate the beat identification rates. The algorithm allows an extensive computational simplification, still providing reliable pattern recognition results for normal QRS beat types (Se=96.18 %; +P=99.61 % on MIT and Se=98.26 %; +P=99.95 % on EDB) as well as for ventricular ectopic QRS types (Se=97.61 %; +P=99.64 % on MIT and Se=99.07 %; +P=98.93 % on EDB). Besides its performance in terms of pseudo classification, the computational simplicity and few restrictions regarding its applicability render the proposed clustering method an interesting choice for online-clustering applications even apart from ECG processing.