AI Detects Low Ejection Fraction on Non-Contrast Chest CT

A retrospective AI model identified abnormal LVEF (<50%) on non-contrast chest CT with external AUROC 0.762, surpassing radiologists. Trained on 25,948 studies, it may enable opportunistic heart failure screening.

• Multi-institutional diagnostic study using 34,058 paired CT and echocardiogram studies; trained on 25,948, externally validated on 8,110 from a separate institution.
• The model outperformed expert radiologists in accuracy and efficiency (exact metrics not reported).
• Retrospective design; patients required both CT and echo, introducing potential selection bias; not yet prospectively validated in a screening setting.

RadPigeon summaries are original and for information only. They are not clinical advice.