Vessel-specific perivascular fat attenuation index derived from AI-CCTA and its association with impaired coronary flow reserve in patients with INOCA

Objectives: To investigate the association between artificial intelligence (AI)-derived coronary computed tomography angiography (CCTA) features and impaired coronary flow reserve (CFR) in patients with ischemia and non-obstructive coronary arteries (INOCA). Methods: Retrospective analysis of 101 suspected coronary artery disease (CAD) patients with non-obstructive stenosis (<50%) on CCTA who underwent cadmium-zinc-telluride single photon emission computed tomography (CZT-SPECT). Stratified by coronary flow reserve (CFR) into CFR < 2.0 and CFR ≥ 2.0 groups at patient and vessel levels. Compared AI-CCTA parameters between groups; identified predictors via logistic regression; diagnostic performance was evaluated using receiver operating characteristic (ROC) analysis with Bootstrap internal validation. Results: At the vessel-level, the CFR < 2.0 group had lower coronary artery calcium score (CACS) (62.12 vs. 142.40 AU, P = 0.021) and higher perivascular fat attenuation index (FAI) (-78.69 ± 8.34 vs. -82.03 ± 8.56 HU, P = 0.009). FAI was independently predictor of CFR < 2.0 (OR = 1.043, 95%CI: 1.005 ∼ 1.084, P = 0.028). A combined model integrating AI-CCTA and clinical features showed an apparent AUC of 0.807, but Bootstrap validation yielded a corrected AUC of 0.648. Inverse spatial distributions of CACS (RCA > LAD > LCX) and FAI (LCX > LAD > RCA). Vessels with CFR < 2.0 were characterized by lower calcification and higher FAI. Conclusions: FAI is independently associated with vessel-level CFR impairment in INOCA. The combined model demonstrates potential but requires external validation. The observed inverse spatial and functional relationship between CACS and FAI may reflect different stages or patterns of coronary atherosclerosis in non-obstructive CAD, warranting further investigation.