Dual-embedding neural network predicts coronary FFR noninvasively with AUC 0.90 vs. invasive FFR

A dual-embedding neural network that integrates patient-specific hemodynamic boundary conditions with coronary geometry achieved AUC 0.90 for FFR prediction vs. invasive FFR in a proof-of-concept study of 288 patients; inference is real-time but not yet externally validated.

• Retrospective study of 288 patients; model trained on CFD-derived FFRCT data.
• Against invasive FFR, correlation r=0.87 and AUC 0.901; RMSE vs. FFRCT was 7.42%.
• Proof-of-concept only; single-center, no external validation, limited CFD training set.

RadPigeon summaries are original and for information only. They are not clinical advice.