Evaluation of clinical performance of artificial intelligence-assisted measurement of thoracic aortic diameter on non-contrast CT

Abstract

This study aimed to evaluate the clinical performance of artificial intelligence (AI)-assisted measurement of thoracic aortic diameter on non-contrast computed tomography (NCCT). This single-center prospective study included 79 patients (mean age = 70 ± 13 standard deviation; 46 men) who underwent NCCT and arterial phase contrast-enhanced computed tomography (CECT) for evaluating aortic diseases between 2021 and 2022. Maximum diameters of 9 American Heart Association landmarks and the maximum ascending/descending aorta were measured. Upper limits of the 95% confidence interval (CI) for the absolute difference in aortic diameter between the AI-assisted/manual method on NCCT and the manual method on CECT (Diff AI-assisted /Diff manual ) were calculated and compared using a predefined cutoff of 1.5 mm. Evaluation times were also compared. No significant difference was found between per-patient Diff AI-assisted and Diff manual (1.10 mm; 95% CI = 1.03–1.18 vs 1.03 mm; 95% CI = 0.95–1.11, P = .058); the upper limit of the 95% CI was <1.5 mm. The upper limit of the 95% CI of the geometric mean Diff AI-assisted and Diff manual was also <1.5 mm at each landmark. The median evaluation time of the AI-assisted method was shorter than that of the manual method (expert radiologist = 272 seconds [interquartile range (IQR) = 229–313] vs 496 seconds [IQR = 438–550]; P < .001, training radiologist: 304 seconds [IQR = 257–404] vs 710 seconds [IQR = 563–855]; P < .001). AI-assisted thoracic aortic diameter measurement on NCCT was as accurate as manual measurement on CECT; its evaluation time was shorter than that of manual measurement on NCCT.