Artificial intelligence(AI) utilizing, coronary angiography-based fractional flow reserve (angio-basedAI-FFR)has emerged as a less invasive coronary physiological assessment. The presentstudy aimed to evaluate the accuracy of Angio-AI-FFR using wire-based FFR asthe standard reference.

We investigated 39 patients (39 coronary arteries: 33 left anteriordescending arteries, 1 left circumflex arteries, 4 right coronary arteries) whounderwent diagnostic coronary angiography and FFR measurement with pressureguidewire to evaluate moderately stenotic lesions. Computation of angio-basedAI-FFR was performed using dedicated software (AutocathFFR, MedHub.AI, LosAngeles, CA).

On the quantitative coronary angiography, the percent diameter stenosis of the target lesions was a medianof 55 ± 12% and the lesion length was 18 ± 5 mm. The wire-based FFR value ofthe target vessels was a median of 0.82 ± 0.09. The frequency of wire-based FFR≤ 0.80 was 15 (38%) vessels. There was an excellent correlationbetween angio-based AI-FFR and wire-based FFR (R = 0.70, p < 0.001). Angio-based AI-FFR showed goodagreement with wire-based FFR with a small mean difference (0.010 ± 0.055). angio-based AI-FFR ≤ 0.80 predictedwire-based FFR ≤ 0.80 with an accuracy of 73%, sensitivity of 86%, specificity of 89%,positive predictive value of 88%, negative predictive value of 86%. Discordance between wire-based FFR and angio-basedAI-FFR was observed in 6 (15%) vessels, including the group withwire-based FFR ≤ 0.80 and angio-based AI-FFR > 0.80 (n = 3 [7%])and the group with wire-based FFR >0.80 and angio-based AI-FFR ≤0.80 (n = 3 [7%]).

Angio-based AI-FFR showed sufficient accuracy when wire-based FFR was used as thestandard reference. The results of the present study support the applicabilityof angio-based AI-FFR in routine clinical practice.