Noninvasive Extraction of Maternal and Fetal Electrocardiograms Using Progressive Periodic Source Peel-off

Yao Li, Xuanyu Luo, Haowen Zhao, Jiawen Cui, Yangfan She, Dongfang Li, Lai Jiang, Xu Zhang

Monitoring a baby’s heart rate during pregnancy is a cornerstone of prenatal care, but the current gold standard—ultrasound—can be bulky, sensitive to movement, and carries minor safety concerns with prolonged use. The "holy grail" of fetal monitoring is the noninvasive fetal electrocardiogram (fECG), recorded directly from the mother’s abdomen.

However, there is a massive signal-to-noise problem: the mother’s heartbeat (mECG) is orders of magnitude stronger than the baby’s, and both are buried under muscle noise and interference. A new paper published by researchers at the University of Science and Technology of China introduces a "Progressive Periodic Source Peel-off" (PPSP) method that might finally crack the code for high-precision fetal heart monitoring.

TL;DR

Researchers have developed a "peel-off" algorithm that iteratively identifies and removes the dominant maternal heartbeat and background noise to reveal the faint fetal ECG signal. By combining physiological constraints with singular value decomposition, the method achieves over 99.5% accuracy, significantly outperforming current deep learning and filtering techniques in noisy, real-world conditions.

What Problem Does This Paper Address?

Extracting a fetal ECG from abdominal recordings is like trying to record a whisper in a crowded stadium while a drum kit (the maternal heart) is playing right next to the microphone.

Existing methods usually fall into two camps: Adaptive Filtering, which often leaves "ghost" maternal signals behind, and Blind Source Separation (BSS), which struggles when the fetal signal is too weak to be distinguished from random noise. Even modern Deep Learning models, while powerful, require massive amounts of labeled data and often struggle with the "black box" problem—they can’t explain why they identified a specific spike as a heartbeat.

Methods

The authors propose the Progressive Periodic Source Peel-off (PPSP) framework, which consists of three clever stages:

  1. Periodic Constrained FastICA (PCFICA): Standard Independent Component Analysis (ICA) tries to find statistically independent signals. PCFICA goes a step further by adding "physiology-informed" constraints. It looks specifically for signals that are periodic (like a heartbeat), ensuring the algorithm doesn't get distracted by random noise.
  2. SVD Waveform Estimation: Once the "spikes" of the heartbeats are found, the system uses Singular Value Decomposition (SVD) to reconstruct the actual shape of the ECG waveform. This helps smooth out artifacts that might be caused by the mother moving or muscle contractions.
  3. The Peel-off Strategy: This is the core innovation. Instead of trying to find the mother and baby simultaneously, the algorithm finds the strongest periodic signal (usually the mother), reconstructs it, and "peels" it off (subtracts it) from the raw data. It then repeats the process on the remaining signal to find the much weaker fetal heart.

Datasets

The researchers validated PPSP across three distinct types of data:

  • Public Databases: The ADFECG and NIFECGA (PhysioNet) databases, which are standard benchmarks in the field.
  • Synthetic Data: Generated using the fECGSYN simulator to test the algorithm's limits at extreme noise levels (down to -6 dB).
  • Clinical Data: The team collected their own data from five pregnant volunteers using a flexible 4×4 grid electrode array, providing a high-density look at real-world abdominal signals.

Key Insights and Ideas

  • Iterative Refinement Wins: The "aha moment" here is that the fetal signal is often so weak that it’s invisible to algorithms until the maternal signal is completely removed. By treating the extraction as a multi-pass "peeling" process, they prevent the maternal signal from "shadowing" the fetus.
  • Physiology as a Guide: By forcing the algorithm to look for periodic signals (heartbeats) rather than just independent ones, they drastically reduced false positives from muscle noise (EMG).
  • SVD vs. Least Squares: The paper demonstrates that SVD is superior for waveform reconstruction because it avoids the abrupt signal truncation artifacts common in simpler mathematical approaches.

Key Results

The PPSP method didn't just work; it set a new bar for the field:

  • Accuracy: Achieved an F1-score of 99.59% on public data and 99.50% on synthetic data, even at the highest noise levels.
  • Clinical Performance: On real-world clinical data, it yielded the lowest Root Mean Square Error (RMSE) for fetal heart rate at 6.20%, significantly beating out state-of-the-art methods like CycleGAN (a popular deep learning approach).
  • Statistical Significance: The PPSP method outperformed all five comparison methods (including FastICA, LMS, and PFP) with a p-value < 0.05.

Limitations and Future Work

While the results are impressive, the authors note a few hurdles:

  • Offline Processing: Currently, the algorithm is designed for offline analysis. For use in a delivery room, it would need to be optimized for real-time, low-latency monitoring.
  • Computational Complexity: The iterative "peel-off" nature of the algorithm is more computationally expensive than a single-pass filter, which might limit its use on low-power wearable devices.
  • Future Directions: The team aims to adapt the framework for real-time use and explore how it handles twins or higher-order multiple pregnancies, where the "peeling" would need to happen even more times.

Summary updated

Abstract

Abdominal electrocardiogram (AECG) gives a safe and non-invasive way to monitor fetal well-being during pregnancy using surface electrodes. However, it is challenging to extract weak fetal ECG (fECG) from the AECG recordings with larger maternal ECG (mECG) and external noises. In this study, we introduce a novel progressive periodic source peel-off (PPSP) method for extracting periodic ECG sources from multi-channel AECG recordings, including three main modules: 1) A periodic constrained FastICA (PCFICA) module with ECG physiology-informed constraints for extracting precise ECG spike trains, 2) A singular value decomposition module for estimating ECG waveforms, and 3) A peel-off strategy that facilitates to discern weak fECG source by eliminating previously separated sources or noises. The performance of the PPSP method was examined on two public databases, synthetic data and our clinical data. For extracting fECG spike trains, our PPSP method achieved an F1-scores of 99.59% on public data, 99.50% on synthetic data at the highest noise level. It further yielded the lowest RMSE of fetal heart rate of 6.20% on clinical data. It significantly outperformed other state-of-the-art methods on any set of data (p &lt; 0.05). This study demonstrated effectiveness of the PPSP method for extracting and separating mECG and weak fECG signals, with high precision especially at high noise levels. Our study promotes noninvasive measurement and intelligent monitoring of both fetal and maternal heart activities towards advanced healthcare in perinatal medicine.
BibTeX
@misc{li2024noninvasive,
  author = {Yao Li and Xuanyu Luo and Haowen Zhao and Jiawen Cui and Yangfan She and Dongfang Li and Lai Jiang and Xu Zhang},
  title = {{Noninvasive Extraction of Maternal and Fetal Electrocardiograms Using Progressive Periodic Source Peel-off}},
  year = {2024},
  eprint = {2406.01281},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2406.01281},
}
Figure