Diagnostic accuracy of quantitative flow ratio for assessment of coronary stenosis significance from a single angiographic view: A novel method based on bifurcation fractal law

Abstract

Objectives: We aimed to evaluate the diagnostic accuracy of computation of fractional flow reserve (FFR) from a single angiographic view in patients with intermediate coronary stenosis.

Background: Computation of quantitative flow ratio (QFR) from a single angiographic view might increase the feasibility of routine use of computational FFR. In addition, current QFR solutions assume a linear tapering of the reference vessel size, which might decrease the diagnostic accuracy in the presence of the physiologically significant bifurcation lesions.

Methods: An artificial intelligence algorithm was proposed for automatic delineation of lumen contours of major epicardial coronary arteries including their side branches. A step-down reference diameter function was reconstructed based on the Murray bifurcation fractal law and used for QFR computation. Validation of this Murray law-based QFR (μQFR) was performed on the FAVOR II China study population. The μQFR was computed separately in two angiographic projections, starting with the one with optimal angiographic image quality. Hemodynamically significant coronary stenosis was defined by pressure wire-derived FFR ≤0.80.

Results: The μQFR was successfully computed in all 330 vessels of 306 patients. There was excellent correlation (r = 0.90, p < .001) and agreement (mean difference = 0.00 ± 0.05, p = .378) between μQFR and FFR. The vessel-level diagnostic accuracy for μQFR to identify hemodynamically significant stenosis was 93.0% (95% CI: 90.3 to 95.8%), with sensitivity, specificity, positive predictive value, negative predictive value, positive likelihood ratio, and negative likelihood ratio of 87.5% (95% CI: 80.2 to 92.8%), 96.2% (95% CI: 92.6 to 98.3%), 92.9% (95% CI: 86.5 to 96.9%), 93.1% (95% CI: 88.9 to 96.1%), 23.0 (95% CI: 11.6 to 45.5), 0.13 (95% CI: 0.08 to 0.20), respectively. Use of suboptimal angiographic image view slightly decreased the diagnostic accuracy of μQFR (AUC = 0.97 versus 0.92, difference = 0.05, p < .001). Intra- and inter-observer variability for μQFR computation was 0.00 ± 0.03, and 0.00 ± 0.03, respectively. Average analysis time for μQFR was 67 ± 22 s.

Conclusions: Computation of μQFR from a single angiographic view has high feasibility and excellent diagnostic accuracy in identifying hemodynamically significant coronary stenosis. The short analysis time and good reproducibility of μQFR bear potential of wider adoption of physiological assessment in the catheterization laboratory.

Keywords: artificial intelligence; coronary angiography; fractional flow reserve; quantitative flow ratio.