Artificial intelligence-based bi-ventricular systolic and diastolic volume, ejection fraction using non-contrast...

Ejection fraction (EF) and end-systolic volume (ESV) are prognostic markers in cardiovascular disease. While MRI provides accurate assessments, its cost limits widespread use. Non-contrast cardiac CT (NCCT), used for coronary artery disease screening, may offer additional functional information. To evaluate the accuracy of AI-derived ventricular volumes and EF from NCCT compared with contrast cardiac CT (CCT) and MRI.

This single center study included 205 patients who underwent cardiac CT for valve planning, divided into retrospective and prospective cohorts. A validated AI algorithm was applied to low-dose NCCT images at end-diastole and end-systole. Right (RV) and left ventricles (LV) volumes and their EFs were compared with CCT and MRI. In the prospective cohort (49 women, 53 men; mean age 73.9 ± 10.3 years), NCCT correlated strongly with CCT for LVEDV (152 mL; -14.2% relative difference; r = 0.91) and LVESV (96 mL; +32.6%; r = 0.84), with similar correlations for RVEDV (163 mL; -8.4%; r = 0.82) and RVESV (121.4 mL; +33.1%; r = 0.85). NCCT predicted LVEF <40% with 98% negative predictive value and 87% accuracy. LVEDV correlated strongly with MRI (n = 16) for CCT (240 mL; +4.2%; r = 0.99) and NCCT (197 mL; -14.3%; r = 0.97), as did LVESV for CCT (115 mL; -5%; r = 0.99) and NCCT (134 mL; +11%; r = 0.97).

AI-derived ventricular volumes from NCCT show moderate to strong correlations, but EF is underestimated. The derived EF can be a screening tool to rule out significant ventricular dysfunction.