A Deep Learning-Guided Ensemble Empirical Mode Decomposition Method for Single-Channel Fetal Electrocardiogram...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Fetal electrocardiogram (FECG) is vital for monitoring fetal well-being and detecting distress.
• Abdominal electrocardiogram (AECG) offers home monitoring but suffers from weak signals and maternal interference.
• Accurate selection of intrinsic mode functions (IMFs) is crucial for Ensemble Empirical Mode Decomposition (EEMD) analysis of AECG.

Purpose of the Study:

• To develop an automated method for extracting FECG signals from AECG data.
• To enhance the accuracy and reliability of FECG analysis for home-based fetal monitoring.
• To improve the detection of fetal arrhythmias and distress using advanced signal processing and deep learning.

Main Methods:

• A deep learning approach using a 1D convolutional neural network (1D CNN) to score and select EEMD-derived IMFs.
• Maternal QRS complex template subtraction and secondary EEMD purification for automated FECG extraction.
• Leave-one-subject-out (LOSO) cross-validation on simulated and real AECG datasets (ADFECGDB, DaISy, NIFEA).

Main Results:

• The IMF classifier achieved a mean AUC of 0.9282 ± 0.0189.
• The CNN-2×EEMD method demonstrated high correlation coefficients (0.94-0.96) and F1-scores (0.8372-0.9565) for fetal R-peak detection.
• Significant SNR improvements (13.39-15.88 dB) were achieved, outperforming conventional methods and matching manual selection.

Conclusions:

• The proposed deep learning-guided method enables accurate and automatic FECG extraction from AECG signals.
• The technique shows significant potential for real-time, wearable fetal monitoring systems.
• This advancement can improve the early detection of fetal cardiac issues and enhance prenatal care.