Arrhythmia and ischemia classification and clustering using QRS-ST-T (QT) analysis of electrocardiogram

The objective is to propose a collective analytical model for QRS complex, ST segment and T wave (i.e., QT complex) of electrocardiogram to evaluate the onset or occurrence of cardiovascular abnormalities. The proposed methodologies also classify healthy subjects, arrhythmic and ischemic patients. The idea is to extract the QRS-ST-T features; where, QT interval and 99% occupied bandwidth (Hz) features are extracted from QT complex and QRS versus ST-T interval ratio (%) is also formulated after segmenting the QT complex into QRS complex and ST-T segment by localizing the inflection points. The evaluation of this proposed approach has been carried out using the selected 36 recordings (true positive (TP) beats) from each standard databases i.e., MIT-BIH arrhythmia database, FANTASIA and European ST-T database. The method is initiated with the preprocessing stage and then the inflection points (i.e., Q, S, T-offset) are detected using Pan-Tompkins method and curve analysis techniques. Then the time-frequency domain features (e.g. QT interval (s) and 99% occupied bandwidth (Hz)) are extracted from the segmented mean QT complex and the QRS versus ST-T interval (%) ratio is extracted from the segmented mean QRS versus ST-T segments simultaneously. These features are introduced to the classifier like decision tree, support vector machine and K-means for clustering operation. The classification success rate is 97.03% and resubstitution error rate is 2.97% among the arrhythmia, ischemia and healthy classes using QT interval and QRS versus ST-T interval ratio (%) features. The evaluations of other features are also analyzed along with graphical classification results. Allied evaluation of segments belonging to ventricular depolarization (QRS complex) and repolarization (ST segment and T wave) i.e., QT complex, will certainly improve the detection probability of ischemia and arrhythmia with further correlative parametric features. This also leads to automatic detection and classification of arrhythmia and ischemia by avoiding visual inspection and error free decison making.