Premature Ventricular Contractions Detection by Multi-Domain Feature Extraction and Auto-Encoder-based Feature Reduction

Cardiovascular disorders are known to be among the most severe diseases and the leading causes of mortality all over the globe. Premature ventricular contractions (PVC) are one of the most prevalent types of cardiac arrhythmia. Recording and analyzing electrocardiogram (ECG) signals is one of the most popular methods (the least intrusive and least expensive) for investigating cardiac disorders. In this study, a new supervised approach for the automated detection of PVC has been developed through a combination of handcrafted feature extraction from ECG signals and deep-learning-based feature reduction. The proposed approach utilized multiple methodologies, namely statistical and chaos analysis in the time domain and time–frequency domain, and morphological assessment, to extract numerous features from ECG signals. Then, a variational autoencoder network is developed as a deep learning-based feature reduction technique to reduce the number of extracted features and obtain the most discriminating features. Finally, a support vector machine, k-nearest neighbor, and neural network classifiers with fivefold cross-validation are utilized to classify ECG signals. The MIT-BIH database is used to evaluate the proposed approach. The numerical results show that the proposed approach performs better than the current state-of-the-art studies.