Medical News Observer

Key Points:

• Researchers created a deep-learning ECG Model by integrating 12-lead ECG data with cardiac MRI measurements, accurately predicting right ventricular dysfunction and dilation.
• The AI model is in its early stages and is not intended to replace advanced diagnostics, which has inherent limitations in current ECG and MRI data.
• This study highlights AI’s potential in cardiac health, aiming for future validation of the deep-learning ECG model in diverse populations to establish its clinical utility in various heart conditions.

Overview

Researchers from the Icahn School of Medicine at Mount Sinai have explored deep learning–enabled electrocardiogram (ECG) analysis to estimate right ventricular (RV) size and function. The research involved training a deep-learning ECG (DL-ECG) model with data from 12-lead ECGs and cardiac magnetic resonance imaging (MRI) measurements. This AI model could offer a simpler and more accessible alternative to more complex imaging technologies, potentially improving patient outcomes. The Journal of the American Heart Association published the study results.

Deep Learning ECG model

Researchers created this by pairing 12‐lead ECG data with cardiac MRI measurements in a large UK Biobank sample and fine-tuning it in a multicenter health system at Mount Sinai with subsequent validation. The model performed well in predicting RV dysfunction and dilation with high accuracy. Using Cox proportional hazards models, the study also found that predicted right ventricular ejection fraction (RVEF) was significantly associated with transplant-free survival.

Limitations

The study assessed the accuracy of predicting heart conditions and their impact on patient survival rates. However, the AI application is in an early stage and is not meant to replace advanced diagnostics. Furthermore, there are inherent limitations in the existing ECG and MRI data and variations in predictions across populations. 

Significance and next steps

The study represents an advancement in cardiac assessment, highlighting AI’s potential role in predicting right ventricular function and size. Future research aims to validate the deep-learning ECG (DL-ECG) models in diverse populations and confirm their clinical usefulness.

Reference

Duong, Son Q., Akhil Vaid, Vy Thi Ha My, Liam R. Butler, Joshua Lampert, Robert H. Pass, Alexander W. Charney, et al. 2024. “Quantitative Prediction of Right Ventricular Size and Function From the ECG.” Journal of the American Heart Association 13 (1): e031671. https://doi.org/10.1161/JAHA.123.031671.

More info about UK BioBank here.