Unsupervised Learning-Based Non-Invasive Fetal ECG Muti-Level Signal Quality Assessment

Area of Science:

• Biomedical Engineering
• Signal Processing
• Maternal-Fetal Medicine

Background:

• Fetal heart rate (FHR) monitoring is crucial for assessing fetal health and growth.
• Non-invasive fetal electrocardiogram (fECG) is widely used for FHR estimation but is susceptible to noise, compromising accuracy.
• Accurate FHR estimation requires reliable fECG signal quality assessment (SQA) to remove low-quality segments.

Purpose of the Study:

• To develop an unsupervised learning-based multi-level fECG SQA approach.
• To classify fECG signal segments into three quality levels (high, medium, low).
• To improve the accuracy of FHR estimation by removing low-quality fECG data.

Main Methods:

• Extracted signal quality features: entropy-based, statistical, and ECG signal quality indices.
• Calculated autoencoder-based features using reconstruction error of spectrograms.
• Classified fECG signal quality using a self-organizing map with extracted features.

Main Results:

• Achieved a weighted average F1-score of 90% for three-level fECG signal quality classification.
• Demonstrated a reduction in FHR estimation errors after removing detected low-quality signal segments.

Conclusions:

• The proposed unsupervised fECG SQA method effectively classifies signal quality without labeled data.
• This approach enhances the reliability of FHR estimation by improving fECG signal quality.
• The method offers a valuable tool for improving prenatal monitoring and fetal health assessment.