Lung cancer resulted in 1.8 million deaths globally in 2022 and remains the leading cause of cancer-related mortality worldwide. Low-dose CT screening has been shown to reduce lung cancer mortality, and many countries have now implemented CT-based screening programs.
James L. Mulshine, MD
Bruce S. Pyenson
In a perspective recently published in Frontiers in Oncology, James L. Mulshine, MD, of Rush University Medical Center in Chicago and Bruce S. Pyenson, an actuary who recently sserved on the Medicare Payment Advisory Commission, an agency of the U.S. Congress, explained that CT scans contain vast amounts of anatomic and quantitative information that could be leveraged to identify risk of multiple thoracic and cardiac conditions.1 They propose that the millions of low-dose CT scans obtained during routine lung cancer screening could become a public research resource to accelerate AI development, improve workflow, and enable earlier detection of emphysema, coronary artery disease, atrial fibrillation, and other thoracic conditions.
Further, the authors wrote that coronary calcification detected on screening CT is a biomarker for cardiac mortality risk, and that CT-based coronary calcium assessment may outperform low-density lipoprotein (LDL) cholesterol or other lipid biomarkers for cardiovascular risk assessment.
High-Volume Image Datasets
The authors suggest that, soon, as many as 2 million lung cancer screening scans could be performed annually in the United States.
The National Lung Screening Trial (NLST) released over 48,000 CT images which proved valuable for research and training. Nevertheless, these images stemmed from scans beginning in 2002 and they are considered technologically outdated and inappropriate for many studies today, the authors noted.
Traditional image collection is slow and costly, and many AI tools are trained on scans from a single institution or scanner type, the authors explained, adding that proprietary image repositories create barriers and favor well-capitalized private firms.
Dr. Mulshine and Mr. Pyenson contend that, with regulatory clearance permissions, images acquired routinely during CT screening for early lung cancer could serve multiple additional purposes if made publicly available for strategic analyses and AI tool development.
The authors propose that a cloud-based public utility or “public library” model of storing deidentified images would foster transparency, equitable access for researchers, validation of AI tools, improved public trust, and cost savings. Importantly, they propose that images would remain stored at individual screening centers. Rather than transferring images to a centralized repository, the authors envision a federated cloud-based model in which deidentified imaging data remain at participating screening centers, AI algorithms are deployed to those local datasets, and only analytic results are transmitted back to the central cloud hub.
Clinical Opportunities Beyond Lung Cancer
The authors argue that information already captured on low-dose CT scans could support interventions aimed at reducing cardiac mortality risk, strengthening smoking cessation efforts, and identifying other thoracic conditions at a stage when earlier intervention may reduce future morbidity and mortality. In particular, the authors reference newer studies that demonstrate that coronary calcification detected on screening CT scans is a highly infomrative biomarker for cardiac mortality risk.Actionable measures to lower this risk are available, including lifestyle changes and statin administration, thereby potentially avoiding later cardiac mortality. Also, it has already been shown in retrospective analyses that emphysema detection could support earlier smoking cessation interventions. Other information garnered from a CT chest scan could lead to early intervention addressing other conditions as well.
AI as Workflow Necessity
The role of AI in supporting this effort is critical. Current radiology workflows would not be able to manage the expanding screening volume, and as such AI will be needed to review serial annual scans, identify subtle interval changes, identify no-change cases, flag high-risk imaging patterns, and support earlier interventions.
The authors identify progress on privacy and governance challenges, and heterogeneous data quality, among others. Despite these challenges, the authors conclude that low-dose CT screening images already contain clinically actionable information that extends significantly beyond lung nodule detection.
In conclusion, the authors wrote: “[Lung cancer screening] images include a remarkable amount of currently useful, clinically relevant information but there is likely even more information of clinical and operational value that is not being used. The imminent availability of millions of [lung cancer screening] images and associated clinical information creates logistical challenges, but the information has huge value. We offer a novel, public utility approach that will accelerate both operational and clinical breakthroughs efficiently and at sustainable, low cost.”
DISCLOSURE: Bruce Pyenson is employed by Pyenson Healthcare Analytics. Dr. Mulshine reported no conflicts of interest.
REFERENCE
1. Mulshine JL, Pyenson BS: Lung cancer screening with AI can discover cures for many early diseases. A public utility can make sure it happens. Front Oncol. Published online March 10, 2026. Available at www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2026.1797777/full.
