Automated AI-Based Aortic Measurements From Attenuation Correction CT as an Adjunctive Cardiovascular Risk Biomarker: An International Multicenter Study
Anna M. Marcinkiewicz, Aakash Shanbhag, Panithaya Chareonthaitawee, Wenhao Zhang, Hasanian Al-Jilaihawi, Ryan Zaid, Sebastien Cadet, Giselle Ramirez, Mark Lemley, Jirong Yi, Waseem Hijazi, Valerie Builoff, Joanna X. Liang, Vinicius F. Calsavara, Andrew J. Einstein, Edward Miller, Attila Feher, Terrence D. Ruddy, Viet T. Le, Steve Mason, Stacey Knight, Erick Alexanderson, Isabel Carvajal-Juarez, Leandro Slipczuk, Mark I. Travin, Thomas L. Rosamond, Samuel Wopperer, Daniel S. Berman, Damini Dey, Marcelo Di Carli, Robert J.H. Miller, Piotr J. SlomkaBACKGROUND:
Aortic enlargement is a powerful predictor of dissection and rupture, yet it is rarely evaluated during routine myocardial perfusion imaging, despite the widespread availability of computed tomography (CT) attenuation correction scans. The aim of this study was to determine whether fully automated, opportunistically derived, artificial intelligence-based aortic measurements from myocardial perfusion imaging CT attenuation correction scans are associated with adverse outcomes in a large multicenter cohort.
METHODS:
Computed tomography attenuation correction scans from patients undergoing positron emission tomography/CT and single-photon emission CT/CT myocardial perfusion imaging across 10 centers were included. A deep learning model automatically segmented the thoracic aorta, and a postprocessing algorithm extracted maximum ascending and descending diameters. The aortic size index (1) was calculated by indexing the diameter to body surface area.
RESULTS:
A total of 29 339 patients (56% male; median age, 66 years [interquartile range, 58–75 years]) were included. Over a median follow-up of 3.5 years (interquartile range, 1.9–5.0 years), 5083 (17.3%) patients died. Median ascending and descending aortic size index values were 1.8 cm/m
2
(interquartile range, 1.6–2.0) and 1.5 cm/m
2
(interquartile range, 1.4–1.6), respectively, with an increase with age and higher values in females. Elevated aortic size index thresholds (ascending >2.2 cm/m
2
; descending >1.6 cm/m
2
) were significantly associated with increased all-cause mortality (ascending: adjusted hazard ratio, 1.16 [95% CI, 1.07–1.26],
CONCLUSIONS:
Artificial intelligence can unlock previously unused information within routine myocardial perfusion imaging CT attenuation correction scans by rapidly and automatically quantifying aortic size at scale. Opportunistic aortic measurements derived from CT attenuation correction may serve as an adjunctive risk biomarker and could add prognostic value to standard myocardial perfusion imaging without additional imaging or radiation.