State-of-the-art review: Commercial artificial intelligence applications in cardiopulmonary imaging—from coronary fractional flow reserve to pulmonary embolism triage
Artificial Intelligence in Health · 28.07.2026 · 026200044
DOI: 10.36922/aih026200044 ↗Kurz gefasst
Fachartikel in der Zeitschrift Artificial Intelligence in Health, online erschienen am 28.07.2026. Verfasst von Shun Dai und 2 weiteren Autor:innen.
Abstract (Originalsprache)
Commercial artificial intelligence (AI) applications in cardiopulmonary imaging have expanded rapidly—723 Food and Drug Administration (FDA)-authorized radiology AI/machine learning (ML) devices had been identified by June 2024, representing 76% of 950 total cleared AI/ML medical devices, yet evidence supporting clinical adoption remains heterogeneous. This state-of-the-art review synthesizes published evidence across four cardiopulmonary subdomains (coronary, cardiac, aortic, and pulmonary artery imaging), applying the Fryback–Thornbury diagnostic efficacy hierarchy as a unified framework for cross-domain comparison, and integrating regulatory, ethical, and equity considerations aligned to the European Union (EU) AI Act (Regulation 2024/1689) and FDA lifecycle management milestones through early 2026. Evidence maturity varies markedly. Computed tomography-derived fractional flow reserve is supported by randomized trials and prospective registry data addressing clinical decision-making (Level 4); preliminary, non-replicated cost data are also available (Level 6), but no trial has demonstrated improvement in major adverse cardiac events (Level 5) as a primary endpoint. Cardiac AI achieves high reproducibility for automated quantification (intraclass correlation coefficient 0.80–0.95) but remains predominantly at Levels 1–2, with a single non-inferiority randomized trial. Aortic AI evidence is almost entirely retrospective (Level 2). Pulmonary embolism triage systems have demonstrated prospective workflow improvements, including reductions in time-to-diagnosis and radiologist miss rates, without demonstrated mortality benefit. Across all domains, only 5% of FDA-cleared devices underwent prospective testing, and only four of 213 (2%) European regulatory-approved radiology AI devices were clearly labeled for pediatric use. Algorithmic bias, explainability limitations, and immature post-market surveillance frameworks represent consistent cross-domain concerns. Realizing patient benefit requires prospective outcome-driven evaluation, representative training datasets, and rigorous post-market monitoring, particularly as EU AI Act high-risk obligations apply from August 2026.
Zitieren
APA
Dai, S., Zhong, J., & Xiao, Z. (2026). State-of-the-art review: Commercial artificial intelligence applications in cardiopulmonary imaging—from coronary fractional flow reserve to pulmonary embolism triage. Artificial Intelligence in Health, 0(0), 026200044. https://doi.org/10.36922/aih026200044
Vancouver
Dai S, Zhong J, Xiao Z. State-of-the-art review: Commercial artificial intelligence applications in cardiopulmonary imaging—from coronary fractional flow reserve to pulmonary embolism triage. Artificial Intelligence in Health. 2026;0(0):026200044. doi:10.36922/aih026200044
BibTeX
@article{Dai2026,
author = {Dai, Shun and Zhong, Jingyu and Xiao, Zhengguang},
title = {State-of-the-art review: Commercial artificial intelligence applications in cardiopulmonary imaging—from coronary fractional flow reserve to pulmonary embolism triage},
journal = {Artificial Intelligence in Health},
year = {2026},
volume = {0},
number = {0},
pages = {026200044},
doi = {10.36922/aih026200044}
}
EUVIMED