Classification of ECG ventricular beats assisted by Gaussian parameters’ dictionary

Automatic processing and diagnosis of electrocardiogram (ECG) signals remain a very challenging problem, especially with the growth of advanced monitoring technologies. A particular task in ECG processing that has received tremendous attention is to detect and identify pathological heartbeats, e.g., those caused by premature ventricular contraction (PVC). This paper aims to build on the existing methods of heartbeat classification and introduce a new approach to detect ventricular beats using a dictionary of Gaussian-based parameters that model ECG signals. The proposed approach relies on new techniques to segment the stream of ECG signals and automatically cluster the beats for each patient. Two benchmark datasets have been used to evaluate the classification performance, namely, the QTDB and MIT-BIH Arrhythmia databases, based on a single lead short ECG segment. Using the QTDB database, the method achieved the average accuracies of 99.3% ± 0.7 and 99.4% ± 0.6% for lead-1 and lead-2, respectively. On the other hand, identifying ventricular beats in the MIT-BIH Arrhythmia dataset resulted in a sensitivity of 82.8%, a positive predictivity of 62.0%, and F1 score of 70.9%. For non-ventricular beats, the method achieved a sensitivity of 96.0%, a positive predictivity of 98.6%, and F1 score of 97.3%. The proposed technique represents an improvement in the field of ventricular beat classification compared with the conventional methods.