On-site xFFR showed 98% sensitivity, 81% specificity, 89% accuracy, and AUC 0.89 for CAD, outperforming CCTA and stress CTP, with 8±3.4 min analysis time.
Does an on-site deep learning and fluidodynamic-based FFR-CT algorithm (xFFR) accurately assess the functional significance of CAD compared to invasive FFR and stress CTP in intermediate to high risk symptomatic patients?
The novel on-site xFFR algorithm provides high diagnostic accuracy for functionally significant CAD, outperforming CCTA and stress CTP when compared to invasive FFR.
Absolute Event Rate: 0% vs 0%
Abstract Background Non-invasive computed tomography-derived fractional flow reserve (FFR-CT) has emerged as a promising alternative to invasive FFR (iFFR)1. However, the current reliance on off-site analysis at the core laboratories and the time-consuming nature of FFR-CT pose logistical challenges, especially in the emergency setting, highlighting the need for on-site tool to derive FFR-CT. Purpose This prospective, single-center study evaluated the diagnostic performance of an on-site mixed deep learning and fluido-dynamic-based fractional flow reserve derived from computed tomography algorithm2 (xFFR, figure 1) in assessing the functional significance of coronary artery disease (CAD), using iFFR as the reference standard. Additionally, the accuracy of xFFR was compared stress computed tomography perfusion (CTP). Methods The study included 289 intermediate to high risk symptomatic patients for CAD who were scheduled for non-emergent ICA. According to the exclusion criteria, 250 patients (mean age 65±9 years, 76% male) were enrolled in the final cohort. All patients received coronary computed coronary angiography (CCTA), stress CTP and xFFR by using ICA+iFFR as reference Results Obstructive CAD (≥50% stenosis) was identified in 71.6% of patients via CCTA and 48% through iFFR. Functionally significant CAD was detected by xFFR and stress CTP in 56.6% and 58% of patients, respectively. xFFR showed sensitivity, specificity, and diagnostic accuracy of 98%, 81%, and 89%, respectively, and an area under the curve (AUC) of 0.89. xFFR showed a strong correlation with iFFR (Cohen’s Kappa: 0.78; 95% CI: 0.70–0.85) outperforming CCTA alone (AUC: 0.75, p 0.001) and stress CTP (AUC: 0.82, p 0.033) (AUC: 0.87, p 0.509) in diagnostic accuracy. The xFFR analysis time averaged 8 ± 3.4 minutes, demonstrating its practicality and feasibility for clinical application. Per-patient and per-vessels comparison of ROC curves between xFFR and CTP using iFFR as reference is shown in figure 2. Conclusions This study establishes xFFR as a robust and efficient on-site tool for assessing CAD, demonstrating high diagnostic accuracy and good agreement with invasive methods, superior to CCTA or CTP. Its rapid processing and integration into clinical workflows position xFFR as a promising alternative to off-site FFR-CT solutions. Further studies are warranted to confirm its generalizability and optimize its implementation.Figure 1.xFFR algorithm Figure 2.Comparisons of ROC curves
Fazzari et al. (Sat,) reported a other. On-site xFFR showed 98% sensitivity, 81% specificity, 89% accuracy, and AUC 0.89 for CAD, outperforming CCTA and stress CTP, with 8±3.4 min analysis time.