Back

Ejection fraction quantification from ungated chest CT by AI

Zhou, J.; Kwiecinski, J.; Shanbhag, A.; Pieszko, K.; Ramirez, G.; Lemley, M.; Killekar, A.; Hijazi, W.; Miller, R. J.; Kavanagh, P. B.; Liang, J. X.; Slipczuk, L.; Travin, M. I.; Alexanderson, E.; Carvajal-Juarez, I.; Packard, R. R.; Al-Mallah, M.; Einstein, A. J.; Acampa, W.; Knight, S.; Le, V. T.; Mason, S.; Rosamond, T. L.; Hiczkiewicz, J.; Wopperer, S.; Chareonthaitawee, P.; Berman, D. S.; Newby, D. E.; Di Carli, M. F.; Dey, D.; Slomka, P. J.

2025-11-19 radiology and imaging
10.1101/2025.11.14.25339933 medRxiv
Show abstract

Left ventricular ejection fraction (LVEF) is an important clinical metric, obtained by specialized imaging across the cardiac cycle. We present a novel AI approach to estimate LVEF from ungated chest CT. Using multicenter (11 sites) registry of 25,852 patients, AI-derived CT LVEF (AI LVEF) showed strong correlation with 3D gated positron emission tomography (r=0.84), area under the curve (AUC) of 0.96, negative predictive value of 95% for reduced LVEF (< 40%), and effectively stratified risk of heart failure, cardiovascular death, and all-cause death. In a separate large multicenter population (n=24,054) with lung CT scans, reduced AI LVEF was associated with a hazard ratio of 13.3 (95% confidence interval 9.7-18.4) for cardiovascular death. AI LVEF could also predict reduced echocardiographic LVEF (AUC=0.91). LVEF can be accurately estimated from non-contrast, ungated, low-dose chest CT scans, effectively stratifying patients for heart failure and mortality, with potential widespread clinical utility.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.