A Multilayer LSTM Auto-Encoder for Fetal ECG Anomaly Detection.

Area of Science:

• Biomedical Engineering
• Cardiology
• Artificial Intelligence in Medicine

Background:

• Fetal electrocardiogram (ECG) analysis is crucial for monitoring fetal well-being.
• Detecting abnormalities in fetal ECG signals presents significant challenges.
• Automated methods are needed to improve the accuracy and efficiency of fetal ECG interpretation.

Purpose of the Study:

• To introduce a multilayer long short-term memory (LSTM) based auto-encoder network for detecting abnormalities in fetal ECG signals.
• To develop a semi-supervised anomaly detection method capable of reproducing ECG variability.
• To enhance the analysis of fetal ECG signals for practical, daily applications.

Main Methods:

• Utilized a multilayer LSTM auto-encoder network architecture.
• Employed time series pattern detection and error reconstruction for anomaly identification.
• Implemented a semi-supervised learning paradigm for filtering and classification.

Main Results:

• The proposed LSTM auto-encoder effectively detects anomalies in fetal ECG signals.
• The method demonstrated superior feature learning compared to traditional approaches without prior knowledge.
• The anomaly detection procedure successfully reproduced ECG variability.

Conclusions:

• The developed LSTM-based auto-encoder is a promising tool for identifying fetal ECG abnormalities.
• This approach facilitates the analysis of fetal ECG signals in real-world scenarios.
• The method offers a robust and data-driven solution for non-invasive fetal monitoring.