Related Experiment Video
Updated: Jun 24, 2025
10:35
Best Current Practice for Obtaining High Quality EEG Data During Simultaneous fMRI
Published on: June 3, 2013
32.7K
A Generalisable Heartbeat Classifier Leveraging Self-Supervised Learning for ECG Analysis During Magnetic Resonance
IEEE Journal of Biomedical and Health Informatics
|June 10, 2024
Summary
Self-supervised learning (SSL) effectively improves electrocardiogram (ECG) analysis during MRI scans. This technique enhances heartbeat classification accuracy, even with limited annotated MRI ECG data.
Area of Science:
- Biomedical Engineering
- Medical Imaging
- Artificial Intelligence
Background:
- Electrocardiogram (ECG) monitoring during Magnetic Resonance Imaging (MRI) is crucial for patient safety and image synchronization.
- ECG signals suffer significant distortion in MRI's electromagnetic environment, complicating automated analysis and pathological heartbeat classification.
- Existing deep learning models for heartbeat classification require large datasets, which are scarce for ECG acquired during MRI.
Purpose of the Study:
- To develop an efficient ECG signal representation using a Siamese network and a large unannotated ECG database.
- To create a robust heartbeat classifier for ECG signals acquired during MRI.
- To evaluate the impact of self-supervised learning (SSL) and data augmentation on classifier performance in the MRI context.
Main Methods:
- Utilized a Siamese network to learn ECG signal representations from a large external dataset.
- Applied various data augmentation techniques, including MRI-specific artifact simulation.
- Employed self-supervised learning (SSL) pretraining for the heartbeat classifier.
- Assessed classifier generalization to ECG signals acquired during MRI.
Main Results:
- SSL pretraining significantly improved the generalizability of heartbeat classifiers in MRI, achieving an F1-score of 0.75.
- Deep learning without SSL achieved an F1-score of 0.46, while a classical machine learning approach yielded 0.40.
- The study demonstrated the effectiveness of SSL in learning efficient ECG representations from limited annotated medical data.
Conclusions:
- Self-supervised learning (SSL) techniques are highly beneficial for developing deep learning models for ECG analysis in MRI.
- SSL enables the creation of accurate heartbeat classifiers even with scarce annotated data specific to the MRI environment.
- The proposed approach offers a promising solution for imaging arrhythmic patients in MRI settings.
