Automated quantitative myocardial perfusion cardiac magnetic resonance with artificial intelligence-based arterial input function correction versus fractional flow reserve

Abstract Funding Acknowledgements Type of funding sources: Public grant(s) – National budget only. Main source(s): Wellcome Trust. Background Quantitative myocardial perfusion cardiac magnetic resonance (QP-CMR) is gaining importance in clinical practice due to the increasing availability automated processing software. However, a significant obstacle widespread deployment QP-CMR accurate estimation arterial input function (AIF), as this precluded by non-linear relationship between observed MR signal and concentration contrast agent. Modern solutions favour dual-saturation acquisition sequence, acquiring specific low-resolution images with short saturation time minimise AIF. approach limited pulse sequence. Recently, AI-AIF has been proposed (1), which uses an artificial intelligence (AI)-based correction AIF, without need for dual-sequence or dual-bolus acquisition. Purpose In work, we compare values, quantified versus invasive reference standard: fractional flow reserve (FFR), identify diagnostically coronary artery disease (CAD), defined FFR < 0.80. Methods This study enrolled 18 patients (mean age 65 (± 11), 8 female) who underwent stress CMR (Philips Achieva 3T) assessment at least one artery. Perfusion was fully-automated manner motion compensation (2) AI-based segmentation (3), using Fermi function-constrained deconvolution. Fully pixelwise maps were generated, further subdivided into 16 segments. Segments assigned their respective territory (4) QP value given mean its two lowest Receiver operating characteristic (ROC) analysis performed cut-off thresholds report diagnostic accuracy approach. Results The overall 2.29 ml/min/g 0.32). Figure 1 shows values 3 example patients. 2.09 0.36) CAD 2.95 0.48) CAD. total, 28 vessels 12 10 being positive (FFR 0.80). FFR-positive territories 1.76 0.37). ROC curves are shown 2 on per-vessel (left) per-patient (right) level. optimal threshold level 2.11 1.96 ml/min/g. Conclusion feasible does not require dual injection. derived after AIF show good agreement epicardial disease. sample size, work required confirm these findings.