Performance of artificial intelligence-based coronary CT angiography-derived fractional flow reserve across cohorts with different prevalence of obstructive coronary artery disease.
Authors
Affiliations (6)
Affiliations (6)
- Department of Cardiology, Heart Center, Amsterdam University Medical Center, Amsterdam Cardiovascular Sciences, the Netherlands.
- Department of Radiology and Nuclear Medicine, Amsterdam University Medical Center, University of Amsterdam, Amsterdam Cardiovascular Sciences, the Netherlands.
- Department of Cardiology, Heart Center, Amsterdam University Medical Center, Amsterdam Cardiovascular Sciences, the Netherlands; Department of Radiology and Nuclear Medicine, Amsterdam University Medical Center, University of Amsterdam, Amsterdam Cardiovascular Sciences, the Netherlands.
- Department of Radiology and Nuclear Medicine, Amsterdam University Medical Center, University of Amsterdam, Amsterdam Cardiovascular Sciences, the Netherlands; Department of Biomedical Engineering and Physics, Amsterdam University Medical Center, the Netherlands; Department of Radiology, Mayo Clinic Rochester, Rochester, United States.
- Department of Cardiology, Heart Center, Amsterdam University Medical Center, Amsterdam Cardiovascular Sciences, the Netherlands. Electronic address: [email protected].
- Department of Radiology and Nuclear Medicine, Amsterdam University Medical Center, University of Amsterdam, Amsterdam Cardiovascular Sciences, the Netherlands; Department of Radiology, Mayo Clinic Rochester, Rochester, United States.
Abstract
Coronary computed tomography angiography (CCTA) provides accurate anatomical assessment of coronary stenosis severity but has limited ability to determine lesion-specific ischemia as assessed by invasive fractional flow reserve (FFR). We aimed to evaluate the diagnostic performance of an artificial intelligence-based deep learning CCTA-derived FFR model (CT-FFR<sub>AI</sub>) in cohorts undergoing CCTA and invasive FFR as part of a study protocol and a real-world cohort undergoing invasive FFR based on suspected coronary artery disease on CCTA in routine clinical care. Diagnostic performance of CT-FFR<sub>AI</sub> (CorEx, SPIMED-AI) was evaluated using an intention-to-diagnose approach and compared with CCTA using 50% and 70% stenosis thresholds, with invasive FFR as reference standard. In the PACIFIC-1 study cohort, 208 patients underwent CCTA and invasive FFR assessment of all major coronary vessels irrespective of CCTA findings. The real-world cohort included 202 patients with at least one CCTA-defined lesion ≥70% stenosis who underwent ICA with invasive FFR. In the PACIFIC-1 cohort (612 vessels), per-vessel diagnostic accuracy was comparable between CT-FFR<sub>AI</sub> (77%), when using a CCTA based ≥50% stenosis threshold (76%; P = 0.689), and when using a ≥70% stenosis threshold (74%; P = 0.183). In the real-world cohort (325 vessels), CT-FFR<sub>AI</sub> demonstrated higher per-vessel diagnostic accuracy (63%) compared with CCTA using ≥50% and ≥70% stenosis thresholds (52% and 57%, respectively; P < 0.001 and P = 0.034). Net reclassification improvement for CT-FFR<sub>AI</sub> beyond CCTA alone using ≥50% and ≥70% stenosis thresholds was 7.9% (P = 0.256) and 17.7% (P = 0.012) in the PACIFIC-1 cohort, compared with 23.2% (P < 0.001) and 13.6% (P = 0.070) in the real-world cohort, respectively. CT-FFR<sub>AI</sub> may outperform anatomic assessment alone, with diagnostic accuracy depending on the prevalence of obstructive coronary artery disease and the range of stenosis severities among vessels undergoing invasive FFR assessment.