A comparison of single channel fetal ECG extraction methods

Area of Science:

• Biomedical Engineering
• Signal Processing
• Maternal-Fetal Medicine

Background:

• Abdominal electrocardiogram (ECG) is a non-invasive method for monitoring fetal cardiac activity.
• Overlapping signals from fetal ECG (FECG), maternal ECG (MECG), and noise present a significant source separation challenge.

Purpose of the Study:

• To compare temporal extraction methods for fetal signal extraction and fetal heart rate estimation.
• To evaluate a novel echo state neural network (ESN) based filtering approach for maternal ECG cancelation.

Main Methods:

• Comparison of ESN-based filtering with Least Mean Square (LMS), Recursive Least Square (RLS), and Template Subtraction (TS) techniques.
• Analysis of real abdominal ECG signals from nine pregnant women (4h 22min data, 37,452 fetal beats).
• Empirical evaluation of signal preprocessing effects, including baseline wander high-pass cutoff frequency.

Main Results:

• The ESN-based algorithm achieved the highest F1 measure (90.2%), outperforming LMS (87.9%), RLS (88.2%), and TS (89.3%).
• Increased high-pass cutoff frequency for baseline wander significantly improved performance across all methods.
• Open-source code provided for benchmark methods to ensure reproducibility.

Conclusions:

• The ESN-based filtering approach demonstrates superior performance for FECG extraction and fetal heart rate estimation.
• Optimizing baseline wander filtering is crucial for enhancing the accuracy of FECG analysis.
• The study provides a reproducible benchmark for evaluating FECG signal processing techniques.