OptRPC: A novel and optimized recurrence plot-based system for ECG beat classification

• Devising an improved recurrence plot to solve information sparsity. • Application of the improved recurrence plot to ECG data. • Determination of embedding dimension using reconstruction score metric. • Parameters of recurrence plot optimized by maximizing entropy. • Real-time capability of ECG beat classification. Cardiac arrhythmia refers to irregularities in heartbeats. Left undiagnosed arrhythmias can cause severe and potentially fatal complications. As a result, early finding of such abnormalities is critical. Electrocardiogram (ECG) is regularly used by medical professionals to diagnose and differentiate cardiac arrhythmias. As a result, there have been many deep learning methods over the years in an attempt to automate this process. But traditional deep learning methods require big training data which often clearly do not reflect the age, weight and gender spectrum of patients and are prone to misclassification when data from different demographics is shown. Hence, temporal features extracted from these datasets are demographically biased. Consequently, in this paper, we intend to introduce Optimum Recurrence Plot based Classifier (OptRPC); a dynamical systems-based method of classifying ECG beats by embedding them in higher dimensions and devising an optimized recurrence plot. A Convolutional Neural Network architecture is then used to classify these recurrence plots. The proposed scheme accomplished an overall accuracy of 98.67% and 98.48% on two benchmark databases and delivered better performance than the previous state-of-the-art methods.