An Efficient 1D Autoencoder-Based Approach for R-Peaks Detection in Electrocardiogram Signals

Electrocardiogram (ECG) signal which is composition of multiple segments such as P-wave, QRS complex and T-wave plays a crucial role in evaluating human heart cardiac diagnosis. For an analysis of cardiac diagnosis, it is required that clinicians scan the ECG signal for R-peaks (the highest peak of the QRS complex) detection, which relies on their expertise and takes enormous time. In order to provide more realistic treatment of cardiovascular diseases in the ECG signal with noises and different characteristics, computer-based R-peaks detection approaches have been actively developed in the research article for many years and generally compose of two major operations; preprocessing and decision. Contrary to previous approaches which require high computational costs and complexity, appropriate initial parameters, a novel CNN based approach is presented for R-peaks detection in this paper. Both qualitative and quantitative analysis are performed in the real ECG signal taken from typical MIT-BIH arrhythmia database. In order to validate the success of the recommended approach, total 18 different R-peaks detection approaches reported in the literature are compared by utilizing performance evaluation metrics. Average values of sensitivity, positive predictivity, error rate, accuracy and f-score are computed as 99,96%, 99,95%, 0,09%, 99,90% and 99,95% respectively, which are the best values among well-established studies for R-peaks detection. It is demonstrated with qualitative and quantitative results that better performance is achieved by our suggested approach.