AECG-DecompNet: abdominal ECG signal decomposition through deep-learning model.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Accurate fetal electrocardiogram (FECG) extraction from abdominal recordings (AECG) is crucial for fetal health monitoring.
• Maternal ECG (MECG) and noise often interfere with FECG signals, complicating analysis.
• Existing methods struggle with precise FECG isolation and preserving signal morphology.

Purpose of the Study:

• To introduce AECG-DecompNet, a novel deep learning framework for efficient MECG and FECG extraction from single-channel AECG.
• To improve the accuracy and reliability of fetal health evaluation through enhanced FECG signal decomposition.
• To address the challenges posed by noise and maternal interference in AECG signals.

Main Methods:

• Developed AECG-DecompNet, a deep learning framework utilizing two series encoder-decoder networks.
• One network focuses on maternal ECG estimation, while the other eliminates interference and noise.
• Incorporated internal and external skip connections for improved signal reconstruction.

Main Results:

• AECG-DecompNet significantly outperforms single-network direct FECG extraction methods.
• Demonstrated superior performance in QRS detection and preservation of FECG morphological information compared to existing techniques.
• Achieved high precision (97.4%), recall (93.52%), and F1-score (95.42%), surpassing state-of-the-art approaches.

Conclusions:

• AECG-DecompNet provides a robust and accurate solution for decomposing AECG signals.
• The framework excels at preserving vital FECG morphological details, particularly in cases of weak fetal signals.
• This advancement offers a promising tool for non-invasive fetal health assessment and monitoring.