AI-derived CT-FFR accuracy compared with CCTA stenosis across cohorts with differing CAD prevalence

An AI-based CT-FFR model achieved 63% per-vessel accuracy in a real-world cohort, outperforming CCTA ≥50% (52%, p<0.001) and ≥70% (57%, p=0.034). In a lower-prevalence trial cohort, accuracy was similar (77% vs 76% and 74%). Performance varied with CAD prevalence.

• In the real-world cohort, CT-FFRAI improved net reclassification by 23.2% (P<0.001) beyond CCTA ≥50% stenosis, and by 13.6% (P=0.070) beyond ≥70%.
• The PACIFIC-1 cohort (n=208) had universal invasive FFR regardless of CCTA findings; the real-world cohort (n=202) only had invasive FFR for lesions with ≥70% stenosis.

Automated summary

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