Investigating Useful Morphological characteristics of ECG for Predicting the Onset of Ventricular Arrhythmia

This work aims to introduce the most reliable morphological features of the electrocardiogram (ECG) signal for prediction ventricular tachycardia (VT) events. We focused on the ventricular activity part of the ECG signal and extracted features regarding morphological changes of this part. We discovered that our two newly introduced features: beat-to-beat correlation of T-wave and beat-to-beat correlation of ST-segment, along with some other useful features extracted from the heart's ventricular activity in previous researches, are significantly different between normal and pre-VT episodes. The results of maximum relevance-minimum redundancy (mRMR) feature selection method and Random Forest (RF) classifier well confirm our findings. Our database for evaluating the algorithm comprises 40 pre-VT and 40 normal records. We achieved a sensitivity of 95%, a specificity of 100%, and 96.7% accuracy in predicting VT events.