Automated CT-FFR model for diagnosing hemodynamically significant coronary artery disease. A prospective validation study

Aims This study aims to assess the diagnostic performance of a novel computed tomography-derived fractional flow reserve (CT-FFR) algorithm and to compare its accuracy at three predefined sites: (i) at the location of invasive FFR measurements (CT-FFRatloc), (ii) at selected sites determined by an automated module integrated within the algorithm (CT-FFRauto), and (iii) distally in the vessel (CT-FFRdistal). Methods and results We prospectively recruited 108 consecutive patients with stable symptoms of coronary artery disease and at least one suspected obstructive lesion on coronary computed tomography angiography (CCTA). CT-FFR was validated against invasive FFR as gold standard using FFR ≤ 0.80 to define myocardial ischaemia. CT-FFRatloc showed good correlation with invasive FFR (r = 0.67) and improved the ability to detect myocardial ischaemia compared with CCTA at both lesion [area under the curve (AUC) 0.83 vs. 0.65, P < 0.001] and patient level (AUC 0.87 vs. 0.74, P = 0.007). CT-FFRauto demonstrated similar diagnostic accuracy to CT-FFRatloc and significantly improved specificity compared with CT-FFRdistal (86% vs. 49%, P < 0.001). High end CT quality improved the diagnostic performance of CT-FFRauto, demonstrating an AUC of 0.92; similarly, the performance was improved in patients with low-to-intermediate coronary artery calcium score with an AUC of 0.88. Conclusion Implementing an automated module to determine the site of CT-FFR evaluations was feasible, and CT-FFRauto demonstrated comparable diagnostic accuracy to CT-FFRatloc when assessed against invasive FFR. Both CT-FFRatloc and CT-FFRauto improved the diagnostic performance compared with CCTA and improved specificity compared with CT-FFRdistal. High end CT quality and low-to-intermediate calcium burden improved the diagnostic performance of our algorithm. ClinicalTrials.gov Identifier NCT03045601