Morphology-based detection of premature ventricular contractions

Premature ventricular contraction (PVC) is the type of ectopic heartbeat, commonly found in the healthy population and is often considered benign. However, they are reported to adversely affect the accuracy of R-R variability based electrocardiographic (ECG) algorithms. This study proposes a Principal Component Analysis (PCA) based algorithmic approach to detect the PVCs based on their morphology. The eigenvectors were derived from signal window around the R-peak, where signal window for the PVC (wPVC) and that of NSR (wNSR) were set to 0.55 seconds and 0.16 seconds respectively. We used 24 ECG recordings from MIT BIH arrhythmia database as training dataset and the remaining 24 ECG recordings as testing dataset. Using the derived eigenvectors and the Linear regression (LR) analysis; complexes corresponding to the wNSR and wPVC were estimated from training and testing datasets. Four different classification methods were employed to differentiate between wPVS and wNSR, namely, Root mean squared error (RMSE), Pearson product-moment correlation coefficient comparision, Histogram probability distribution and k-Nearest Neighbour (KNN). All four methods were implemented individually to classify the wPVC and wNSR. The performance of each of the classification approach was evaluated by computing sensitivity and specificity. With the sensitivity of 93.45% and specificity of 93.14%, KNN based classification method has shown the best performance. The method proposed in this study allows for an effective differentiation between NSR beats and PVC beats.