Clustering Of Ecg Cycles

Abstract

The paper deals with application of cluster analysis to different ECG records in order to identify particular cardiac pathologies. The work is mainly focused on the detection of premature atrial and premature ventricular beats. Presented approach is based on the signal correlation and further beat type identification and beats clustering via specific ECG features and detection rules, including fuzzy expert rules. By evaluation the method on test data, we obtained Se 76,0 %, Sp 90,2 %, F1 43,8 %, Acc 89,5 %, and PPV 31,1 %. Pure F1 and PPV is due to high number of false positive detections mainly in noisy ECG or ECG with manifested atrial fibrillation.