Fetal ECG extraction using short time Fourier transform and generative adversarial networks

Area of Science:

• Biomedical Engineering
• Signal Processing
• Fetal Monitoring

Background:

• Abdominal ECG (AECG) signals are crucial for fetal monitoring but are often corrupted by noise.
• Extracting clean fetal ECG (FECG) from noisy AECG presents significant challenges.

Purpose of the Study:

• To develop and validate a novel approach for FECG extraction from AECG signals.
• To utilize short-time Fourier transform (STFT) and generative adversarial networks (GANs) for improved FECG signal isolation.

Main Methods:

• Transformed 1D AECG signals into 2D time-frequency domain using STFT.
• Employed a GAN model to estimate FECG components in the time-frequency domain.
• Reconstructed the FECG signal in the time domain via inverse STFT.

Main Results:

• The proposed method demonstrated high effectiveness on two public databases (PCDB and ADFECGDB).
• Achieved high performance metrics including Sensitivity (SE), Positive Predictive Value (PPV), and F1-score.
• Reported SE, PPV, and F1 scores above 90% on both datasets.

Conclusions:

• The novel method successfully extracts FECG directly in the 2D time-frequency domain, unlike traditional 1D time-domain methods.
• This approach offers a promising advancement for FECG extraction and fetal monitoring.
• The technique provides a new perspective on addressing FECG extraction challenges.