QRS Complex Detection in Paced and Spontaneous Ultra-High-Frequency ECG

Background: Analysis of ultra-high-frequency ECG (UHF-ECG, sampled at 5,000 Hz) informs about dyssynchrony of ventricles activation. This information can be evaluated in real-time, allowing optimization of a pacing location during pacemaker implantation. However, the current method for real-time QRS detection in UHF-ECG requires suppressed pacemaker stimuli. Aim: We present a deep learning method for real-time QRS complex detection in UHF-ECG. Method: A 3-second window from V1, V3, and V6 lead of UHF-ECG signal is standardized and processed with the UNet network. The output is an array of QRS probabilities, further transformed into resultant QRS annotation using QRS probability and distance criterion. Results: The model had been trained on 2,250 ECG recordings from the FNUSA-ICRC hospital (Brno, Czechia) and tested on 300 recordings from the FNKV hospital (Prague, Czechia). We received an overall F1 score of 97.11 % on the test set. Conclusion: Presented approach improves UHF-ECG analysis performance and, consequently, could reduce measurement time during implant procedures.