An Autoencoder-Based Fetal Heart Rate Detector for Noninvasive Recordings

Area of Science:

• Biomedical Engineering
• Signal Processing
• Maternal-Fetal Medicine

Background:

• Antenatal fetal health monitoring relies on noninvasive abdominal electrocardiogram (ECG) for fetal heart rate (HR) estimation.
• Maternal ECG, motion artifacts, and bio-electric potentials significantly degrade fetal ECG signal quality.
• Accurate fetal HR monitoring is crucial for assessing fetal well-being during pregnancy.

Purpose of the Study:

• To propose an improved framework for accurate fetal HR estimation from abdominal ECG recordings.
• To effectively remove maternal ECG contamination using a neural network autoencoder.
• To enhance the diagnostic quality of noninvasive fetal ECG signals.

Main Methods:

• Utilized a neural network autoencoder to reconstruct maternal ECG from abdominal ECG signals.
• Employed blind-source separation (BSS) algorithm for post-processing residual signals.
• Developed a framework combining autoencoder-based maternal ECG removal and BSS.

Main Results:

• The proposed framework demonstrated superior performance in fetal HR estimation compared to conventional methods.
• Effective reconstruction and removal of maternal ECG signals were achieved.
• Preservation of inherent fetal ECG artifacts was noted during maternal ECG reconstruction.

Conclusions:

• The novel framework significantly improves fetal HR estimation accuracy from abdominal ECG.
• Neural network autoencoders offer a promising approach for maternal ECG suppression in noninvasive fetal monitoring.
• The method enhances the reliability and diagnostic value of antenatal fetal health assessments.