Artificial Intelligence Evaluation of Coronary Computed Tomography Angiography

An artificial intelligence (AI) based evaluation of coronary computed tomography angiography (CTA) in close agreement to blinded, core lab-interpreted quantitative coronary angiography, enables an accurate, rapid identification and exclusion of high-grade stenosis, according to findings published in the Journal of the American College of Cardiology: Cardiovascular Imaging.

Clinical reads of coronary CTA sometimes result in overestimation of the severity of coronary disease stenosis. AI-based solutions may exist. Researchers sought to compare the performance of AI-enabled quantitative coronary computed tomography angiography (AI-QCT) for detection and grading of coronary stenosis with core lab-interpreted coronary CTA, core lab quantitative coronary angiography (QCA), and invasive fractional flow reserve (FFR).

To accomplish this, researchers conducted a retrospective study of 303 patients (64±10 years of age, 71% men) from the multinational Computed Tomographic Evaluation of Atherosclerotic Determinants of Myocardial Ischemia (CREDENCE; ClinicalTrials.gov identifier: NCT02173275) trial with coronary CTA, FFR, and QCA data analyzed with a Food and Drug Administration-cleared, cloud-based software that performs AI-enabled coronary segmentation, lumen and vessel wall determination, plaque quantification and characterization, and stenosis determination.

Stenosis of 50% or greater was found in 0, 1, 2, and 3 coronary vessel territories, with a prevalence of 32%, 35%, 21%, and 13%, respectively. Average AI-QCT analysis time was 10.3±2.7 minutes. For patients with stenosis of 50% or greater, AI demonstrated per-patient sensitivity of 94%, specificity of 68%, positive predictive value of 81%, negative predictive value of 90%, and accuracy of 84%. For detection of stenosis of 70% or greater, per-patient sensitivity was 94%, specificity was 82%, positive predictive value was 69%, negative predictive value was 97%, and accuracy was 86%.

Correlation was high between stenoses detected by AI-QCT vs QCA on a per-vessel and per-patient basis (intraclass correlation coefficient = 0.73 and 0.73, respectively; P <.001 for both). False positive AI-QCT findings were noted in 7.3% of vessels.

This study was limited by using post hoc analysis instead of a prospective clinical trial.

“In this analysis of the multinational CREDENCE trial, an AI-based evaluation demonstrated high diagnostic performance for the identification, exclusion, discrimination, and correlation to a QCA reference standard,” the study authors wrote. “Given the rapid turnaround time of this AI-QCT and its superior performance to previous coronary CTA core lab and site readers, this approach may augment clinical coronary CTA interpretation.”

Disclosure: Some study authors declared affiliations with biotech, pharmaceutical, and/or device companies. Please see the original reference for a full list of authors’ disclosures.

Reference

Griffin WF, Choi AD, Riess JS, et al. AI evaluation of stenosis on coronary CT angiography, comparison with quantitative coronary angiography and fractional flow reserve: A CREDENCE trial substudy. JACC Cardiovasc Imaging. Published online February 16, 2022. doi:10.1016/j.jcmg.2021.10.020