Deep Learning-Enabled Assessment of Left Heart Structure and Function Predicts Cardiovascular Outcomes

Area of Science:

• Artificial Intelligence in Medicine
• Cardiology
• Medical Imaging Analysis

Background:

• Echocardiography is crucial for assessing cardiac structure and function.
• Automated interpretation of echocardiographic images using deep learning could enhance clinical workflows.

Purpose of the Study:

• To develop and validate a deep learning model for echocardiogram interpretation.
• To assess the association of deep learning-derived cardiac measures with incident clinical outcomes.

Main Methods:

• A 3D convolutional neural network was trained on 64,028 echocardiograms for view classification and quantification.
• Model performance was validated using independent datasets.
• Cox models were used to evaluate associations between model-derived measures and outcomes.

Main Results:

• The deep learning model accurately classified echocardiographic views and quantified cardiac measures (R² 0.53–0.91).
• Model performance was consistent across validation datasets.
• Lower left ventricular ejection fraction predicted increased risk of heart failure and death.

Conclusions:

• Deep learning provides accurate quantification of cardiac structure and function from echocardiograms.
• AI-driven echocardiogram interpretation shows potential for automated disease prediction at scale.