Lung-RADS evolves: new categories, volumetric debates and AI on the horizon

Lung cancer remains the leading cause of cancer mortality worldwide, and low-dose CT screening has been shown to reduce lung cancer deaths in high-risk populations. Since 2014, the American College of Radiology’s Lung Imaging Reporting and Data System, known as Lung-RADS, has served as the standardized lexicon and management framework for reporting screen-detected pulmonary nodules in the United States. It is the only classification system approved by the Centers for Medicare and Medicaid Services for lung cancer screening reporting, and it has been adopted internationally or used as a foundation for other reporting systems. A new review published in Cancer Imaging on 21 October 2026 summarizes the system’s major updates, the areas still under active debate within the Lung-RADS Committee, and the research directions that could reshape screening in the coming years.

Lung-RADS has evolved through iterative refinements rather than wholesale redesigns. Version 1.1, released in 2019, added classification criteria for perifissural nodules, revised size criteria for nonsolid nodules, and updated measurement criteria. The most recent version, v2022, introduced in November 2022, brought guidance on atypical pulmonary cysts, juxtapleural nodules, airway-centered nodules, potentially infectious findings, and a stepped management approach for nodules that are stable or decreasing in size. According to the review’s authors, these refinements have been guided by evolving knowledge of nodule behavior and input from multidisciplinary stakeholders, with the overarching goal of balancing early detection of clinically significant cancers against overdiagnosis, unnecessary imaging, and patient anxiety.

One of the most consequential debates concerns how nodules should be measured. Lung-RADS relies on mean diameter measurements, an approach inherited from the National Lung Screening Trial, whereas the Dutch-Belgian NELSON trial used volumetric assessment with volume doubling time. Volumetry offers potential advantages, including improved reproducibility and greater sensitivity to growth, particularly for small-to-intermediate, irregularly shaped nodules where a diameter may not reflect actual volume. Volume doubling time is a sensitive indicator of growth, with values greater than 30 days but less than 400 days generally indicative of malignancy. Yet volumetric approaches carry drawbacks: segmentation failures in subsolid or irregular nodules, susceptibility to differences in acquisition protocol and reconstruction kernel, variable performance at juxtavascular and juxtapleural locations, and inconsistent interoperability across vendors, all of which can affect classification and management recommendations.

The evidence base does not settle the question. Both measurement approaches were validated in randomized trials that established a mortality reduction from low-dose CT screening, but no randomized trial has directly compared diameter-based and volumetric assessment as its primary objective. The NELSON trial demonstrated a 24 percent reduction in lung cancer mortality at 10 years using volumetric classification, while extended follow-up of the National Lung Screening Trial at a median of 12.3 years reported an attenuated mortality reduction of 8 percent, or 11 percent after adjustment for dilution, figures lower than the 20 percent reduction in the original 2011 analysis. The review’s authors caution that cross-trial comparison requires considerable care: the National Lung Screening Trial compared low-dose CT against chest radiography, whereas NELSON was a screening-versus-no-screening trial, and the studies differed in population, nodule management, screening rounds, and era. The attenuation at extended follow-up is also partly explained by post-screening dynamics, as the lead-time advantage predictably diminishes once both arms return to standard care.

Retrospective studies show broadly comparable performance between the two methods in malignancy prediction, but with telling distinctions. A direct comparison of Lung-RADS with the NELSON volumetric algorithm found that clinical Lung-RADS, which incorporates radiologist judgment beyond size criteria alone, achieved the highest area under the curve at 0.96, while strict diameter-based Lung-RADS reached 0.81 and downgraded 24 percent of cancers to category 2. A systematic review found that for small indeterminate nodules in the 100 to 300 cubic millimeter range, diameter measurements may lack the sensitivity of volumetric assessment. One study of solid screen-detected nodules found that semiautomated volumetry assigned a lower Lung-RADS category to 36 percent of nodules compared with manual diameter measurement, with only 3 percent classified higher, suggesting manual measurement may systematically overestimate category, although clinical outcomes were not assessed. The review notes that most comparisons evaluate classification at a single time point rather than growth rate, and it concludes that the current literature does not demonstrate the superiority of either approach at trial endpoints. The Lung-RADS Committee has distributed a survey to American College of Radiology members to clarify current practice patterns, and mean diameters remain both validated and pragmatic for now.

Subsolid nodules, comprising nonsolid ground-glass and part-solid lesions, represent another challenging frontier. While these nodules can harbor malignancy, screening risks diagnosing and treating cancers that would never affect morbidity or mortality, given the generally low aggressiveness of the adenocarcinomas they most often represent. Long-term surveillance studies support a conservative approach. In one study of unresected subsolid nodules followed for approximately 77 months, the majority of prognostically significant lung cancers arose elsewhere in the lung, and no screened patient with lung cancer arising in a subsolid nodule experienced relapse or death attributable to that lesion. In the Multicentric Italian Lung Detection study, subsolid-derived cancers were never identified as the cause of death over nearly 10 years of follow-up. A study of 1,676 patients found that for subsolid nodules up to 2 centimeters with a consolidation-to-tumor ratio of 0.25 or less, five-year event-free survival was 100 percent with imaging surveillance versus 99 percent with surgical resection.

Management of ground-glass nodules shows the greatest variation across guidelines. Chinese, Japanese, British Thoracic Society, and Fleischner Society guidelines recommend early recall low-dose CT in 3 to 6 months for ground-glass nodules 6 millimeters or larger but less than 15 millimeters, whereas Lung-RADS recommends repeat imaging at 12 months for nodules from 20 to less than 30 millimeters and at 6 months for those 30 millimeters or larger, with no biopsy recommendation until a nodule becomes part-solid or solid. An analysis of archival National Lung Screening Trial data suggested that if ground-glass nodule categories were calibrated to malignancy probability thresholds similar to those for solid nodules, the size threshold for early recall should be revised lower rather than higher. The authors note, however, that the supporting data derive primarily from Asian guidelines and cohorts, where subsolid nodule epidemiology may differ from Western screening populations. For part-solid nodules, the picture is different: a solid component greater than 8 millimeters on lung-window settings predicts histopathologic invasiveness, prompting diagnostic workup, and 3 percent of part-solid nodules with a solid component of 1 to 2 centimeters and 11 percent with a solid component of 2 to 3 centimeters have N1 or N2 nodal metastases.

Version 2022 also introduced Category 0 for probable infectious or inflammatory findings, addressing a prior mismatch in which new large nodules suspected of being infectious were managed as Category 4B with a 1-month follow-up, contributing to unnecessary anxiety and resource use. In a study of 3,800 screening patients, 331, or 8.7 percent, met criteria for infectious or inflammatory lesions, with patterns including ground-glass opacities in 46.8 percent, tree-in-bud or branching nodules in 16.9 percent, consolidation in 12.4 percent, and clustered nodular opacities in 20.2 percent. The new guidance recommends follow-up low-dose CT in 1 to 3 months for new segmental or lobar consolidation, large new nodules of 8 millimeters or more appearing within a short interval, or multiple new nodules, with reassessment based on the most concerning findings. Juxtapleural nodule criteria were also expanded: v2022 extends the benign Category 2 classification previously reserved for perifissural nodules to all juxtapleural nodules under 10 millimeters with smooth margins, direct pleural attachment, and characteristic shapes, features associated with a high likelihood of benignity. Multiple studies support this expansion, including one finding no cancers among 234 noncalcified nodules near fissures with triangular or oval morphology.

Looking forward, the review outlines several research directions. Risk-based approaches are being distinguished carefully from category-based ones: eligibility determination concerns whether a cancer-free individual will develop lung cancer, while nodule management concerns whether an identified nodule is malignant, and the authors caution against assuming that lower-risk screening-eligible individuals warrant less aggressive nodule follow-up, since nodule characteristics dominate malignancy prediction models. Artificial intelligence is advancing rapidly, with recent studies suggesting AI systems can match or possibly surpass radiologist performance in detecting subtle nodules. In an external test of a deep learning algorithm across three European screening trials, at 100 percent sensitivity the algorithm correctly classified 68.1 percent of participants with benign indeterminate nodules as low risk, compared with 47.4 percent for the PanCan model, a 39.4 percent relative reduction in false positives, though the authors stress that calibration, prospective validation, and recalibration to real screening prevalence remain prerequisites for clinical use. Other future directions include a Lung-RADS Atlas modeled after the BI-RADS Atlas, potential incorporation of incidental nodules, noting that five to nine times as many lung cancers are diagnosed following incidental nodules compared with screening, and international harmonization of reporting standards, an ambitious goal the authors argue remains worthwhile despite substantial transatlantic and transpacific differences in eligibility and nodule management.

Subject of Research: Updates to the American College of Radiology's Lung-RADS system for lung cancer screening reporting and nodule management

Article Title: ACR lung-RADS: updates and future directions

Article References: Petranović, M., Prosper, A. E., Hunsaker, A. R., Lam, S., & Christensen, J. D. (2026). ACR Lung-RADS: updates and future directions. Cancer Imaging, 26(1), Article 147. https://doi.org/10.1186/s40644-026-01140-7

Image Credits: AI Generated

DOI: 10.1186/s40644-026-01140-7

Keywords: Lung-RADS, lung cancer screening, low-dose CT, pulmonary nodules, volumetric measurement, subsolid nodules, ground-glass nodules, artificial intelligence, deep learning, overdiagnosis, American College of Radiology, risk stratification