Validation and diagnostic performance of a fast on-site deep learning-based CT-FFR algorithm

Abstract Background/Introduction CT-based fractional flow reserve (CT-FFR) has been extensively studied and established as a valuable tool for clinical decision making over the past decade. Nevertheless, implementation not systematically adopted due to economic technical reasons. Among latter, turn-around time computation analysis' results potentially plays an important role. Purpose To evaluate feasibility diagnostic accuracy of CT-FFR computed on-site with novel, deep learning-based algorithm using invasive hemodynamic indices reference standard. Methods Sixty-one patients who underwent clinically indicated coronary tomography angiography FFR (iFFR) and/or instantaneous wave-free ratio (iFR) measurements were retrospectively included. analysis was performed in 77 arteries prototype software based on learning algorithms anatomy segmentation prediction pressure drop under rest hyperemia. The performance detect significant lesions assessed iFFR (≤0.8) iFR (≤0.89) standard (60 iFFR, 11 iFR, 3 both) receiver operating characteristic area curve (AUC) calculated. Furthermore, correlation Bland-Altman (BA) performed. Time including processing manual edits lumen recorded. Results successful 59 (97%) 74 (96%) arteries. In arteries, 31 invasively found be hemodynamically significant. Total mean per patient 7 minutes 55 seconds. Compared indices, per-lesion sensitivity specificity 90%, 98%, respectively. AUC vs. significance 0.94, (95% confidence interval: 0.86–0.98). correlated well (r=0.77) only very small bias (0.02) narrow BA limits agreement (−0.14 0.17). accuracy, 96%, 93%, 100%, Conclusion A novel yields excellent compared lesion-specific ischemia offers potential readily implemented into practice given that it can fast on-site. Funding Acknowledgement Type funding sources: None.