Semi-simulated Data for Improving Fetal QRS Detection Using Deep Neural Networks

This paper presents a deep learning-based approach for reliable fetal QRS detection in abdominal ECG recordings. Fetal electrocardiography (fECG) enables non-invasive monitoring of fetal heart activity using surface electrodes on the mother’s abdomen. Identifying fetal QRS complexes is crucial for heart rate monitoring but challenging due to their low amplitude relative to maternal components and noise. Limited labeled datasets further hinder the development of advanced detection methods. To address this, we extended the FECGSYN toolbox to generate semi-simulated abdominal fECG datasets. Simulated signals were combined with real ECG noise, and additional vectorcardiograms were incorporated to increase variability. This process produced 1,200 records suitable for training deep learning models. As proof of concept, a 1D U-Net was trained to segment fetal QRS regions, achieving a 98% F1-score on the generated dataset. These results demonstrate the potential of this approach to improve non-invasive fECG analysis and monitoring accuracy.