Teemu Maaniitty, Sarah Bär, Takeru Nabeta, Jeroen J. Bax, Antti Saraste, Juhani Knuuti
Background Artificial intelligence–guided quantitative computed tomography ischemia (AI‐QCT ischemia ) is a novel machine‐learning method for predicting myocardial ischemia from coronary computed tomography angiography (CCTA). This observational cohort study aimed to compare the long‐term prognostic value of AI‐QCT ischemia with hybrid CCTA/positron emission tomography (PET) myocardial perfusion imaging in suspected coronary artery disease (CAD). Methods Symptomatic patients with suspected CAD underwent CCTA with selective downstream PET to detect ischemic CAD. Blinded reanalysis of CCTA images was done using the AI‐QCT ischemia algorithm, providing a binary result (normal versus abnormal). Results In the full analysis set (n=2271), hybrid CCTA/PET imaging was successful in 94% of the patients and AI‐QCT ischemia evaluation was feasible in 83%, resulting in a per‐protocol set of 1772 patients (19% with ischemic CAD on hybrid CCTA/PET and 25% with abnormal AI‐QCT ischemia ). There was moderate‐to‐substantial agreement between the methods (Cohen’s κ=0.61). During a median follow‐up of 7.0 years, 177 (10%) patients experienced the composite end point of all‐cause death, myocardial infarction, or unstable angina. Ischemic CAD on hybrid CCTA/PET was predictive of the composite end point (hazard ratio [HR], 2.35 [95% CI, 1.62–3.40]; P <0.001), after adjustment for clinical variables and early (6‐month) myocardial revascularization. Similarly, an abnormal (ischemic) AI‐QCT ischemia result was independently predictive of adverse outcomes (adjusted HR, 1.98 [95% CI, 1.39–2.80]; P <0.001). The adjusted models, including either hybrid CCTA/PET or AI‐QCT ischemia , demonstrated similar discriminative ability (C‐index 0.734 versus 0.729; P =0.53). Conclusions The AI‐QCT ischemia algorithm demonstrated long‐term prognostic value comparable to hybrid CCTA/PET perfusion imaging in suspected CAD.