[Fetal ECG extraction using temporal convolutional encoder-decoder network]

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Context:

• Maternal abdominal wall recordings contain mixed electrocardiogram (ECG) signals.
• Accurate fetal heart rate and ECG pattern analysis is crucial for fetal health monitoring.
• Existing methods struggle to isolate weak fetal ECG signals effectively.

Purpose:

• To develop a deep learning framework for extracting fetal ECG signals from mixed maternal abdominal ECG recordings.
• To improve the accuracy of fetal ECG analysis by removing maternal ECG interference.
• To enable more reliable fetal health monitoring during pregnancy.

Summary:

• A temporal convolutional encoder-decoder network (TCED-Net) was developed for nonlinear adaptive noise cancelling (ANC) of maternal ECG components.
• The TCED-Net estimates and subtracts maternal ECG from abdominal signals, isolating the fetal ECG.
• Performance was validated using synthetic (FECGSYNDB) and clinical (NIFECGDB, PCDB) datasets, demonstrating superior signal extraction capabilities.

Impact:

• The proposed method achieves high F1-scores (up to 99.1%) and improved R peak detection accuracy compared to existing algorithms.
• Extracted fetal ECG signals are significantly clearer, offering enhanced diagnostic potential.
• This advancement holds potential for more effective and non-invasive fetal health monitoring throughout pregnancy.