TCGAN: Temporal Convolutional Generative Adversarial Network for Fetal ECG Extraction Using Single-Channel Abdominal...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence in Healthcare

Background:

• Noninvasive fetal electrocardiogram (FECG) monitoring is crucial for assessing fetal development.
• Extracting FECG from abdominal ECG (AECG) is challenging due to maternal ECG (MECG) interference and noise.
• Accurate FECG waveform details are essential for reliable fetal health assessment.

Purpose of the Study:

• To develop an advanced method for extracting FECG from single-channel AECG.
• To improve the accuracy and detail preservation of FECG signals.
• To enhance the capabilities of noninvasive fetal monitoring.

Main Methods:

• A Temporal Convolutional Generative Adversarial Network (TCGAN) was designed for FECG extraction.
• An encoder-decoder architecture with temporal convolution blocks, transpose convolutions, and skip connections was utilized.
• The model was trained and validated using both synthetic (FECGSYDB) and real-world (ADFECGDB) datasets.

Main Results:

• TCGAN demonstrated outstanding performance in fetal QRS complex detection, achieving high positive predictive values (PPV) of 99.54% and 99.02% on the datasets.
• The method successfully extracted FECG signals with well-preserved waveform details.
• Comparative analysis showed TCGAN outperforms state-of-the-art methods in FECG extraction accuracy.

Conclusions:

• TCGAN is a highly effective deep learning model for FECG extraction from single-channel AECG.
• The preserved waveform details facilitate more accurate assessment of fetal development by clinicians.
• This technology holds significant promise for advancing noninvasive fetal monitoring.