Self-supervised VICReg pre-training for Brugada ECG detection | JoVE Visualize

Area of Science:

Cardiology
Artificial Intelligence
Medical Diagnostics

Background:

Supervised deep learning for ECG classification requires extensive labeled data, limiting application to rare conditions like Brugada syndrome (BrS).
Accurate ECG labeling for rare cardiac diseases is challenging, hindering the development of robust diagnostic algorithms.
Existing methods struggle with data scarcity for identifying less common cardiac arrhythmias.

Purpose of the Study:

To develop a novel deep learning model for Brugada syndrome (BrS) ECG classification that overcomes limitations of supervised learning due to scarce labeled data.
To leverage self-supervised pre-training with Variance-Invariance-Covariance Regularization (VICReg) for improved rare cardiac disease identification.
To refine the understanding of BrS prevalence and patient outcomes by identifying previously missed cases.

Main Methods:

Developed a novel deep learning model incorporating Variance-Invariance-Covariance Regularization (VICReg) for self-supervised pre-training.
Applied the VICReg model to classify electrocardiogram (ECG) data for Brugada syndrome (BrS) detection.
Compared the performance of the VICReg model against a state-of-the-art neural network using key performance metrics.

Main Results:

The VICReg model demonstrated superior performance compared to a state-of-the-art neural network in all evaluated metrics.
Achieved an area under the receiver operating characteristic curve (AUC-ROC) of 0.88 and an area under the precision-recall curve (AUC-PR) of 0.82.
Successfully identified previously missed Brugada syndrome (BrS) cases, leading to a revised estimation of institutional prevalence and improved patient outcome assessment.

Conclusions:

Self-supervised learning with VICReg offers a powerful approach for rare cardiac disease identification from ECGs, overcoming data limitations.
The developed model enhances diagnostic accuracy for Brugada syndrome (BrS) and provides a framework for other rare cardiac conditions.
This study challenges existing Brugada syndrome (BrS) prevalence estimates and highlights the potential of advanced AI in uncovering underdiagnosed conditions.