Residual convolutional autoencoder combined with a non-negative matrix factorization to estimate fetal heart rate

Area of Science:

• Biomedical Engineering
• Signal Processing
• Maternal-Fetal Medicine

Background:

• Fetal heart rate (fHR) monitoring is crucial for assessing fetal well-being.
• Non-invasive fetal electrocardiography (fECG) is an emerging alternative to traditional Doppler ultrasound.
• Extracting clean fECG from abdominal ECG (aECG) is challenging due to maternal ECG and noise.

Purpose of the Study:

• To develop an advanced method for extracting high-quality fECG from single-channel aECG recordings.
• To accurately estimate fHR using the extracted fECG signal.
• To provide a clinically relevant tool for non-invasive fetal health assessment.

Main Methods:

• A deep residual convolutional autoencoder network was trained on synthetic aECG data.
• Transfer learning was applied using real aECG recordings for fECG extraction.
• Non-negative matrix factorization was employed to estimate fHR from the cleaned fECG.

Main Results:

• The proposed method demonstrated significant performance improvements over existing techniques.
• The approach successfully extracted clean fECG and accurately estimated fHR.
• Evaluation on three public databases confirmed the method's efficacy.

Conclusions:

• The developed deep learning approach effectively extracts fECG and estimates fHR from single-channel aECG.
• This method offers a non-invasive and robust solution for fetal health monitoring.
• It eliminates the need for prior knowledge of noise sources or maternal R-peak locations.