Pacing Electrocardiogram Detection With Memory-Based Autoencoder and Metric Learning.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Remote electrocardiogram (ECG) diagnosis is crucial for clinical workflows.
• Detecting pacemakers in ECGs is challenging due to limited patient history.
• Automatic pacing ECG detection can reduce cardiologist workload and misdiagnosis rates.

Purpose of the Study:

• To propose a novel autoencoder framework for automatic pacing ECG detection from remote ECG signals.
• To enhance the autoencoder with a memory module for improved feature representation and retrieval.
• To introduce a metric learning-based objective function for robust pacing ECG identification.

Main Methods:

• Developed a memory-enhanced autoencoder framework for pacing ECG detection.
• Integrated a memory module to record and query typical pacing ECG features.
• Utilized a metric learning-based objective function to differentiate pacing from non-pacing ECGs.
• Introduced a new pacing ECG database with 800 patients and 8,000 heartbeats.

Main Results:

• The proposed method achieved an average F1-score of 0.918 on the pacing ECG database.
• The framework demonstrated effective generalization on the MIT-BIH arrhythmia database.
• The memory module enabled accurate reconstruction of pacing ECGs.

Conclusions:

• The novel autoencoder framework with a memory module is effective for automatic pacing ECG detection.
• The metric learning objective function provides a reliable indicator for pacing ECG identification.
• This approach can significantly aid cardiologists in remote ECG diagnosis.