Deep learning for estimating right ventricular function from routine coronary angiography.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Coronary angiography traditionally assesses coronary artery disease.
• It may offer additional clinical insights beyond its primary use.
• Right ventricular dysfunction is a critical indicator of cardiac health.

Purpose of the Study:

• To investigate the utility of deep learning (DL) for detecting right ventricular (RV) dysfunction.
• To analyze RV function from cine images acquired during coronary angiography.
• To evaluate the performance of DL models in classifying RV function.

Main Methods:

• Trained 3D-convolutional neural networks (CNNs) using cine angiograms (LAO and RAO projections) of the right coronary artery (RCA).
• Used transthoracic echocardiography as the ground truth for RV function.
• Evaluated model performance in detecting any RV dysfunction (≥mild) and significant RV dysfunction (≥mild-moderate) in a cohort of 10,336 patients.
• Assessed the impact of integrating an ECG-driven AI model with the angiography-driven model.

Main Results:

• The DL models achieved an AUC of 0.82 for detecting any RV dysfunction and 0.83 for significant RV dysfunction.
• Sensitivity and specificity were 0.75 and 0.74 for any dysfunction, and 0.82 and 0.70 for significant dysfunction.
• Combining the angiography model with an ECG-driven AI model improved AUC to 0.83 for any RV dysfunction and 0.87 for advanced RV dysfunction.

Conclusions:

• A novel DL algorithm demonstrated acceptable accuracy in identifying RV dysfunction from routine RCA cine angiography.
• The model's predictive capability was enhanced by incorporating ECG data.
• This approach offers a potential method for non-invasive assessment of RV function using existing angiographic data.