Combined cCTA and TAVR Planning for Ruling Out Significant CAD: Added Value of ML-Based CT-FFR

Abstract

Objectives: The purpose of this study was to analyze the ability of machine-learning (ML)-based computed tomography (CT)-derived fractional flow reserve (CT-FFR) to further improve the diagnostic performance of coronary CT angiography (cCTA) for ruling out significant coronary artery disease (CAD) during pre-transcatheter aortic valve replacement (TAVR) evaluation in patients with a high pre-test probability for CAD.

Background: CAD is a frequent comorbidity in patients undergoing TAVR. Current guidelines recommend its assessment before TAVR. If significant CAD can be excluded on cCTA, invasive coronary angiography (ICA) may be avoided. Although cCTA is a very sensitive test, it is limited by relatively low specificity and positive predictive value, particularly in high-risk patients.

Methods: Overall, 460 patients (age 79.6 ± 7.4 years) undergoing pre-TAVR CT were included and examined with an electrocardiogram-gated CT scan of the heart and high-pitch scan of the vascular access route. Images were evaluated for significant CAD. Patients routinely underwent ICA (388/460), which was omitted at the discretion of the local Heart Team if CAD could be effectively ruled out on cCTA (72/460). CT examinations in which CAD could not be ruled out (CAD⁺) (n = 272) underwent additional ML-based CT-FFR.

Results: ML-based CT-FFR was successfully performed in 79.4% (216/272) of all CAD⁺ patients and correctly reclassified 17 patients as CAD negative. CT-FFR was not feasible in 20.6% because of reduced image quality (37/56) or anatomic variants (19/56). Sensitivity, specificity, positive predictive value, and negative predictive value were 94.9%, 52.0%, 52.2%, and 94.9%, respectively. The additional evaluation with ML-based CT-FFR increased accuracy by Δ+3.4% (CAD⁺: Δ+6.0%) and raised the total number of examinations negative for CAD to 43.9% (202/460).

Conclusions: ML-based CT-FFR may further improve the diagnostic performance of cCTA by correctly reclassifying a considerable proportion of patients with morphological signs of obstructive CAD on cCTA during pre-TAVR evaluation. Thereby, CT-FFR has the potential to further reduce the need for ICA in this challenging elderly group of patients before TAVR.

Keywords: aortic stenosis; computed tomography coronary angiography; coronary angiography; coronary artery disease; diagnostic accuracy; machine learning; transcatheter aortic valve implantation.