Invasive validation of novel 1D models for computation of coronary fractional flow reserve

Aims Computed virtual fractional flow reserve (vFFR), derived from invasive angiography, non-invasively quantifies coronary epicardial lesion physiology. Developments of 1-dimensional (1D) vFFR models have introduced methods of side-branch flow representation and reduced simulation time by several orders of magnitude vs. 3-dimensional (3D) alternatives. This study aimed to quantify agreement and diagnostic accuracy of 1D and 3D vFFR models, in a matched cohort, and compare results with established FFR alternatives. Methods and results We used five 1D models, which differed in their side-branch flow representation, to compute vFFR in 104 arteries. The simplest model ignored side-branch flow, the second and third models used vessel anatomy to homogenously distribute side-branch flow and regionalize this to bifurcations, respectively. The final two 1D models additionally used simulated pressure in the main vessel to modulate side-branch flow magnitude. To aid interpretability, diagnostic accuracy was also reported for 3D vFFR, visual assessment and resting invasive pressure assessment (Pd/Pa). Median FFR was 0.81 [0.73-0.88] and 46 (44%) lesions were haemodynamically significant. Optimal FFR agreement was achieved with the 1D model that regionalized side-branch flow to bifurcations (mean bias at diagnostic threshold -0.03, 95% agreement limits -0.23 to 0.20). Diagnostic accuracy did not differ significantly between the five 1D models, with area under the curve (AUC) values ranging 0.68 to 0.74. Diagnostic accuracy for 1D vFFR was superior to visual assessment, comparable to 3D vFFR and poorer than invasive resting pressure assessment. Conclusion 1D models of vFFR facilitate rapid in-silico assessment of epicardial lesion severity. Inclusion of anatomical side branch flow mildly improved agreement, but the additional inclusion of simulated pressure was not beneficial. Agreement of 1D models was comparable to 3D simulations. However, current 1D models are not sufficiently accurate to suggest they may entirely replace wire-based assessment.