AI-augmented ECG for pre-echocardiography triage: a tool to optimize cardiac imaging utilization

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Echocardiography is crucial for diagnosing cardiac dysfunction but is frequently overused.
• Variability in clinical assessment leads to unnecessary referrals, increasing healthcare costs.
• A scalable, cost-effective screening tool is needed to optimize echocardiography utilization.

Purpose of the Study:

• To develop and validate an AI tool for predicting major echocardiographic abnormalities using standard 12-lead ECG images.
• To serve as a triage tool, reducing unnecessary echocardiography referrals.
• To identify reduced ejection fraction, valvular heart disease, and elevated pulmonary artery pressure.

Main Methods:

• Utilized a dataset of 52,817 patients who underwent both ECG and echocardiography.
• Trained an ensemble of 3 deep learning models on ECG images.
• Assessed model performance using AUROC, PRAUC, sensitivity, specificity, PPV, and NPV, with internal and external validation.

Main Results:

• The AI model achieved an AUROC of 0.87 and PRAUC of 0.66 on the internal test set.
• Achieved 80% sensitivity and 95% negative predictive value at the Youden threshold.
• External validation showed an AUROC of 0.84, indicating strong generalizability.

Conclusions:

• The AI model accurately identifies major echocardiographic abnormalities from ECG images.
• The tool demonstrates high negative predictive value and strong generalizability.
• This AI can serve as an effective triage tool to optimize echocardiography referrals.