Benchmarking open-source algorithms for QRS detection and RRI editing in textile electrocardiography

The increasing availability of textile electrocardiography sensors integrated into T-shirts raises the question whether they could be used to monitor clinically-relevant parameters in posthospital care. A proven family of parameters for this task is heart rate variability which is based on quantifying the variations in time intervals between heartbeats. This analysis requires the detection of individual heart beats in electrocardiography signals using QRS detectors, with the interval between successive beats referred to as RR interval (RRI). In case of inaccurate detections, editing algorithms are used for correcting erroneous RRIs. Since a thorough benchmark of QRS detectors and RRI editing in textile electrocardiography is lacking, we evaluate 12 open-source QRS detectors on a publicly-available dataset, containing data from 13 healthy volunteers performing different activities while wearing a commercially-available T-shirt. We perform another experiment by adding additional noise from the MIT-BIH Noise Stress Test Database to the signals before processing. The detectors xqrs (mean sensitivity 99.6%, mean precision 99.1%), kalidas2017 (97.0%, 95.4%) and a low-complexity algorithm Rodrigues2021 (95.7%, 95.3%) achieved best results on the noisy data. Results show that their performance is not much degraded due the additional noise or subject activities. Multiple other QRS detectors resulted in low sensitivity and/or precision, indicating that they are not well suited for textile ECG data. In addition, we apply a state-of-the-art method for RRI editing on the results of xqrs and demonstrate that it is able to reject false positive detections in 25% of subjects.