Ultra-High Frequency ECG Deep-Learning Beat Detector Delivering QRS Onsets and Offsets

Abstract

Background: QRS duration is a common measure linked to conduction abnormalities in heart ventricles. Aim: We propose a QRS detector, further able to locate QRS onset and offset in one inference step. Method: A 3-second window from 12 leads of UHF ECG signal (5 kHz) is standardized and processed with the UNet network. The output is an array of QRS probabilities, further processed with probability and distance criterion, allowing us to determine duration and final location of QRSs. Results: The model was trained on 2,250 ECG recordings from the FNUSA-ICRC hospital (Brno, Czechia). The model was tested on 5 different datasets: FNUSA, a dataset from FNKV hospital (Prague, Czechia), and three public datasets (Cipa, Strict LBBB, LUDB). Regarding QRS duration, results showed a mean absolute error of 13.99 ± 4.29 ms between annotated durations and the output of the proposed model. A QRS detection F-score was 0.98 ± 0.01. Conclusion: Our results indicate high QRS detection performance on both spontaneous and paced UHF ECG data. We also showed that QRS detection and duration could be combined in one deep learning algorithm.