Discriminative sparse coding of ECG during ventricular arrhythmias using LC-K-SVD approach

Ventricular tachycardia (VT) and ventricular fibrillation (VF) are two major types of ventricular arrhythmias that results due to abnormalities in the electrical activation in the ventricles of the heart. VF is the lethal of the two arrhythmias, which may lead to sudden cardiac death. The treatment options for the two arrhythmias are different. Therefore, detection and characterization of the two arrhythmias is critical to choose appropriate therapy options. Due to the time-varying nature of the signal content during cardiac arrhythmias, modeling and extracting information from them using time and frequency localized functions would be ideal. To this effect, in this work, we perform discriminative sparse coding of the ECG during ventricular arrhythmia with hybrid time-frequency dictionaries using the recently introduced Label consistent K-SVD (LC-K-SVD) approach. Using 944 segments of ventricular arrhythmias extracted from 23 patients in the Malignant Ventricular Ectopy and Creighton University Tachy-Arrhythmia databases, an overall classification accuracy of 71.55% was attained with a hybrid dictionary of Gabor and symlet4 atoms. In comparison, for the same database and non-trained dictionary (i.e the original dictionary) the classification accuracy was found to be 62.71%. In addition, the modeling error using the trained dictionary from LC-K-SVD approach was found to be significantly lower to the one using the non-trained dictionary.