Accelerated Sample-Accurate R-Peak Detectors Based on Visibility Graphs

The effective detection and accurate clinical diagnosis of cardiac conditions strongly relies on the correct localization of R-peaks in the electrocardiogram (ECG). Recently, demand for sample-accurate R-peak detection, which is essential to precisely reveal vital features, such as heart rate variability and pulse transit time, has increased. Therefore, we propose two novel sample-accurate visibility-graph-based R-peak detectors, the FastNVG and the FastWHVG detector. The visibility graph (VG) transformation maps a discrete signal into a graph by representing sampling locations as nodes and establishing edges between mutually visible samples. However, processing large-scale clinical ECG data urgently demands further acceleration of VG-based algorithms. The proposed methods reduce the required computation time by one order of magnitude and simultaneously decrease the required memory compared to a recently proposed VG-based R-Peak detector. Instead of transforming the entire ECG, the proposed acceleration benefits largely from building the VG based on a subset containing only the samples relevant to R-peak detection. Further acceleration is obtained by adopting the computationally efficient horizontal visibility graph, which has not yet been used for R-peak detection. Numerical experiments and benchmarks on multiple ECG databases demonstrate a significantly superior performance of the proposed VG-based methods compared to popular R-peak detectors.