Quantitative Prediction of Right Ventricular Size and Function From the ECG

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Right ventricular ejection fraction (RVEF) and end-diastolic volume (RVEDV) are crucial metrics for assessing heart health but are challenging to measure accurately with conventional techniques.
• The application of deep learning (DL) for analyzing electrocardiograms (ECGs) to estimate right ventricular (RV) size and function remains an underexplored area in cardiovascular research.

Purpose of the Study:

• To develop and validate a DL-ECG model capable of predicting RV dilation and dysfunction.
• To assess the model's ability to estimate numerical RVEDV and RVEF.
• To investigate the association between predicted RVEF and patient survival.

Main Methods:

• A DL-ECG model was trained using 12-lead ECGs and cardiac magnetic resonance imaging (CMR) data from the UK Biobank (n=42,938) to predict RV dilation, RV dysfunction, RVEDV, and RVEF.
• The model was fine-tuned on data from a multicenter health system (MSH_original; n=3019) and prospectively validated (MSH_validation; n=115).
• Performance was evaluated using area under the receiver operating characteristic curve (AUC) for categorical predictions and mean absolute error (MAE) for continuous measures. Survival analysis was performed using Cox proportional hazards models.

Main Results:

• The DL-ECG model demonstrated good performance in predicting RV dysfunction (AUCs: 0.86-0.77) and RV dilation (AUCs: 0.91-0.92) across different cohorts.
• In the MSH_original cohort, the model achieved an MAE of 7.8% for RVEF and 17.6 mL/m² for RVEDV.
• Predicted RVEF was significantly associated with improved transplant-free survival (HR=1.40 per 10% decrease; P=0.031) over a median follow-up of 2.3 years.

Conclusions:

• Deep learning analysis of ECGs can effectively identify significant RV dysfunction and dilation, as validated against CMR standards.
• The DL-ECG model provides a non-invasive method for estimating RV parameters, offering valuable clinical insights.
• Predicted RVEF derived from ECGs shows a significant association with patient clinical outcomes, highlighting its prognostic value.