Abstract :
This paper presents a new event classification method. First a pre-filtering is carried out, followed by a (ECG) beat detector and classifier. When it is possible, the characteristic points (P,Q,R,S,J,T,U) are recognized and classified. After that an event recognition method is performed, which can extract the most important information helping doctors to build up a quick and reliable diagnosis. Tested with MIT/BIH database, we observed (ECG) beat detection rate above 99.85%, but the beat classification algorithm needs more development for both methods (parametrical and transformation). To evaluate the performance of the characteristic points detection algorithm, we used our evaluated samples (16-bit resolution and 500 Hz sampling rate). The main result of this work is, that although in many cases the ECG signal contains in itself enough information to build up a diagnosis and the program can determine many useful information for the doctor, the developed algorithm is not able to realize by itself a safe diagnosis
Keywords :
adaptive filters; electrocardiography; medical signal detection; medical signal processing; signal classification; signal sampling; smoothing methods; wavelet transforms; ECG beat detector; ECG recordings; QRS beats; adaptive smoothing; beat classification algorithm; beat detection rate; characteristic points; entropy coding; event classification method; event recognition; event separation; optimal filter; parameter extraction; pre-filtering; real-time adaptive algorithm; self-learning method; wavelet transform; Character recognition; Classification algorithms; Data mining; Databases; Detection algorithms; Detectors; Electrocardiography; Sampling methods; Signal resolution; Testing;
