Improving PVC Detection in ECG Signals: A Recurrent Neural Network Approach

Premature ventricular contractions (PVCs) are abnormal ventricle heartbeats that disrupt the normal QRS rhythm. Classification of PVCs is crucial for the diagnosis and management of cardiac conditions. Detecting anomalies in an electrocardiogram (ECG) is straightforward, but determining the specific number of anomaly classes related to PVCs remains challenging despite Holter monitoring. This work explores the potential of Recurrent Neural Networks (RNNs), specifically Long Short-Term Memory (LSTM) and bidirectional LSTM (BLSTM) models, to learn and reproduce the wave patterns present in ECGs without abnormalities. The working hypothesis is that these trained RNNs can accurately analyze new ECGs with normal QRS complexes by minimizing the root mean square error (RMSE) between their predictions and the actual ECG values. Applying the trained RNNs to ECGs with anomalies aims to identify intervals where the RMSE is high, indicating the presence of PVC pattern classes. Only regular ECG data in the dataset is used for training and evaluation. At the same time, both classes are included in the testing phase to assess the ability of the network to discriminate between the two ECG classes. Also, we used trained models on another dataset composed of PVCs to extract the interval with the highest RMSE. The results obtained from this study suggest the promising performance of LSTM-based approaches in PVC pattern classification.