Fetal ECG Extraction From Maternal ECG Using Attention-Based CycleGAN.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Non-invasive fetal electrocardiogram (FECG) monitoring is crucial for fetal well-being.
• Separating FECG from maternal ECG (MECG) is challenging due to low FECG amplitude, signal overlap, and noise.
• Traditional methods require complex tuning and pre-configuration, limiting their effectiveness.

Purpose of the Study:

• To develop an advanced signal processing technique for accurate FECG decomposition from MECG.
• To enhance the precision of signal generators in FECG extraction using attention mechanisms.
• To improve the fidelity of fetal heart electrical activity monitoring.

Main Methods:

• Utilized an attention mechanism to create a masking region of interest, enhancing signal generator precision.
• Employed a sine activation function to preserve signal details across different domains.
• Evaluated the method on Physionet datasets (A&D FECG, NI-FECG, NI-FECG challenge) and a synthetic dataset (FECGSYN).

Main Results:

• Achieved an average R-Square of 98% for mapping abdominal MECG to scalp FECG on the A&D FECG dataset.
• Attained high F1-scores for fetal QRS detection: 99.7% (A&D FECG), 99.6% (NI-FECG), and 99.3% (NI-FECG challenge).
• Signal distortion was rated as "very good" to "good", comparable to state-of-the-art methods.

Conclusions:

• The proposed algorithm demonstrates high performance in non-invasive FECG signal-to-signal conversion.
• The method shows significant potential for reliable and accurate fetal electrocardiogram monitoring.
• This technique offers a promising advancement in obstetric signal processing and diagnostics.