An Efficient and Robust Deep Learning Method with 1-D Octave Convolution to Extract Fetal Electrocardiogram.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Maternal-Fetal Medicine

Background:

• Invasive fetal electrocardiogram (fECG) monitoring carries infection risks and is typically limited to labor.
• Non-invasive fECG extraction from abdominal signals is challenging due to complex algorithms and feature engineering.

Purpose of the Study:

• To develop a pure end-to-end deep learning model for non-invasive fetal QRS complex detection.
• To create a computationally efficient model for potential out-of-clinic fetal monitoring.

Main Methods:

• An end-to-end deep learning model utilizing a ResNet architecture with 1-D Octave Convolution (OctConv).
• The model learns temporal frequency features and highlights important signal regions for QRS detection.
• Evaluation used PhysioNet 2013 Challenge data, including noise-added datasets to simulate real-world conditions.

Main Results:

• Achieved a F1 score of 91.1% for fetal QRS complex detection.
• Reduced computational cost by over 50% with less than 2% performance degradation.
• Demonstrated the model's effectiveness in identifying prominent regions for accurate detection.

Conclusions:

• The proposed deep learning model offers an effective and computationally efficient solution for non-invasive fetal QRS detection.
• This advancement holds promise for enabling remote and continuous fetal monitoring.
• The model's ability to reduce computational load makes it suitable for real-world applications outside clinical settings.