Applying artificial intelligence to ensure high quality and equitable lung cancer screening

Introduction

Lung cancer is the 2^nd most frequently diagnosed cancer among US men and women and the leading cause of cancer death (1). Lung cancer incidence and mortality rates vary substantially by sex, age, race/ethnicity, and geographic region (2-4). Although contributors to disparities in lung cancer are multifactorial and multilevel, ensuring equitable access to high-quality lung cancer screening (LCS) programs is essential for reducing the burden of lung cancer across diverse populations.

The National Lung Screening Trial (NLST) reported in 2011 that LCS with low-dose computed tomography (LDCT) scans reduced lung cancer mortality by 20% compared to screening with chest X-rays (5). Subsequently, the US Preventive Services Task Force (USPSTF) recommended annual LDCT screening beginning at age 55 years for persons with a 30 pack-year (PY) smoking history who currently smoked or had quit within the past 15 years (6). Medicare & Medicaid Services (CMS) then issued a national coverage determination (NCD) mandating a shared decision-making visit to discuss the benefits and harms of screening and that LDCT be performed at accredited screening centers (where radiologists were expected to have experience interpreting chest computed tomography (CT) scans and results would be submitted to a national registry) (7). However, screening uptake has been low even though LCS is endorsed by multiple professional organizations (8-11). The Behavioral Risk Factor Surveillance System survey estimated that only 18% of eligible persons were screened in 2022 (8).

Disconcertingly, the initial guidelines also gave rise to geographic, racial, and socioeconomic disparities. The CMS requirement that LDCT be performed in accredited LCS centers impacted access, disadvantaging rural and low socioeconomic populations. Sixty million Americans (20%) live in rural areas (12), where lung cancer incidence and mortality rates may be elevated due to higher smoking prevalence and greater environmental exposures and there is less access to accredited screening centers (3). Five-percent of the population does not live within 40 miles of an American College of Radiology (ACR) accredited LCS registry and 18% of age-eligible adults live in a county where none of the population has access to a screening center (13,14).

Additionally, eligibility for LCS was based on the NLST enrollment criteria, a study that did not include representative samples of groups at high risk for lung cancer, including Black individuals, those with lower socioeconomic status, and younger women (5). Many Black individuals, who develop cancer at earlier ages and with less extensive smoking histories than White individuals, would not meet screening eligibility criteria (15,16). Surveillance, Epidemiology, and End Results (SEER) data further show that racial disparities in the burden of lung cancer, particularly the higher lung cancer incidence among Black men compared to non-Hispanic White and Hispanic men, increase with increasing rurality of residence, indicating the intersectionality of race and rurality in contributing to LCS disparities (17). Aside from screening eligibility and access to screening centers, multiple types of disparities have also been observed around adherence with annual screening and access to high-quality treatment (18-21).

To address the growing disparities in LCS, the USPSTF subsequently recommended lowering the starting age to 50 years and reducing the PY requirement to 20 (22). CMS then updated its NCD to align with the USPSTF eligibility criteria but removed requirements for experienced cardiothoracic radiologists and for submitting data to a national registry (23). The new guideline, which essentially doubled the number of eligible subjects, could help mitigate some of the racial and ethnic disparities created by the initial eligibility criteria (16). However, important sustainers of screening disparities among underserved populations persist, including limited access to high-quality imaging interpretations.

One promising strategy to enhance LCS and address existing challenges is to integrate artificial intelligence (AI) into the screening process. AI technologies have been applied across various stages of the screening process, including risk stratification, image acquisition, nodule detection, malignancy prediction, and the management of incidental findings. This review highlights the pivotal role of radiology in LCS and explores how incorporating AI into the screening workflow may help reduce disparities and improve outcomes. The authors initially reviewed a set of articles that came from their personal files; collectively, the authors have expertise in LCS (all), disparities in LCS (C.E.G., R.M.H.), and LDCT and AI (J.C.S., J.D.N). Authors additionally reviewed reference lists from these articles and conducted focused PubMed searches from 2011 through 2026 using the MeSH terms “artificial intelligence”, “early detection of cancer”, and “lung neoplasms” to identify potentially relevant review articles and clinical studies. We defined relevant to be addressing the use of AI to increase nodule detection, distinguish benign from malignant nodules, characterize incidental findings, and predict future cancers, even in the absence of nodules.

Performing and interpreting LDCT scans

The screening process begins with identifying persons eligible for screening, determining whether they are healthy enough to undergo screening and treatment, performing shared decision making, providing tobacco cessation counseling for individuals currently smoking, referring appropriate persons for an LDCT, and then interpreting the images. Follow-up will then be dictated by these interpretations.

The NLST results showed that LDCT scans improved the detection of lung cancer in at-risk smokers compared to chest radiography when scans were obtained in settings that assured high-quality image acquisition and interpretation. The ACR provides a mechanism to certify sites for performing LDCT LCS scans based upon their CT accreditation programs (24). LCS performed in a certified screening program requires radiologists to use the Lung-RADS system (25) to report LDCT scan results, including lung nodules and significant incidental findings. The initial CMS guidelines required the reading cardiothoracic radiologist to have recent experience with at least 300 chest CT acquisitions and that the radiology facility certify that the scans were performed and interpreted in accordance with guidelines from the ACR, Society of Thoracic Radiology and American Association of Physicists in Medicine (7). However, the updated guideline (23) removed the requirement for chest CT experience and markedly widened the pool of potential screening candidates, thus increasing radiologists’ workloads. These changes could exacerbate disparities if images are being performed and interpreted in settings with less experienced radiologists or limited radiology capacity. One potential strategy for addressing these challenges is to integrate AI to support the screening process (26).

AI solutions

AI may have a role in better identifying persons who can most benefit from LCS, assist radiologists in identifying nodules, determining whether nodules are more likely to be malignant, and addressing incidental findings (Figure 1). Applying AI solutions in under resourced settings, particularly those serving underrepresented racial/ethnic populations, may help address disparities by better identifying high-risk individuals eligible for screening and improving the quality of image interpretation.

Figure 1 Outline of the clinical workflow for LCS with LDCT and the complementary assistive opportunities to incorporate AI. AI, artificial intelligence; CMS, Medicare & Medicaid Services; CT, computed tomography; LCS, lung cancer screening; LDCT, low dose computed tomography.

Refining the cohort for CT-based LCS

Risk prediction models that target screening for those at highest risk for lung cancer could counterbalance the potential risks of the new USPSTF guidelines, which while increasing eligibility among underrepresented populations, also lead to screening many more low-risk patients. These patients, particularly those from economically and socially marginalized populations, have a lower expected absolute benefit from screening but face burdensome clinical and financial harms.

Pre-screening statistical models (not using AI), such as the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (PLCOm2012) model, incorporate demographic (age, race, education), medical (history of lung disease, body mass index), and behavioral (smoking status and intensity) factors to predict the likelihood of developing lung cancer within 6 years (27). The goal of using these models, which do not consider imaging data, is to provide risk thresholds to guide personalized decisions for undergoing LDCT. This approach could potentially limit screening among lower-risk individuals who are less likely to benefit but face clinical and financial harms from unnecessary procedures. Choi et al. explored the performance of the PLCOm2012 in a multiethnic cohort study and concluded that risk stratification could reduce racial and ethnic disparities for improved LCS compared to the USPSTF 2021 criteria (28). Incorporating imaging results from chest radiographs assessed by AI has been reported to improve predictive performance. Lu et al. developed a neural network model based on chest radiographs and data readily available in the electronic medical record to identify high-risk persons (29). The model had significantly better discrimination for lung cancer risk through 12 years than the CMS eligibility criteria. This AI approach was validated in a large cohort study (19,488 individuals), where applying USPSTF selection criteria resulted in selecting 7,835 individuals for LCS but missing 37 individuals with cancer (not selected for LCS). Incorporating the chest radiograph AI method, the number of subjects selected for LCS was reduced by 20.4% and only 3 individuals with cancer were missed (not selected for LCS) (30). However, this study did not include information about the racial diversity or socioeconomic status of the cohort and hence further validation of the approach is needed to determine performance stability across populations.

Increasing access to high-quality LCS

The most recent USPSTF guideline and CMS coverage determination may stress the US capacity to provide high-quality LCS. Based on the USPSTF criteria, the estimated number of eligible subjects has increased substantially from around 8 to 14 million (31). Many rural populations, though, are not near accredited centers and there are numerous geographic areas (counties in Maine, West Virginia, Georgia, Alabama, Oklahoma, Arkansas, Missouri, Mississippi Delta) with high prevalences of tobacco use, high lung cancer mortality rates, and limited access to screening (32-34). A study of LCS services in rural Oregon reported that many sites do not perform all the recommended components of a screening program, with 92% incorporating some sort of structured reporting (Lung-RADs), 21% having a tracking system for nodule management and 71% collected follow-up data for quality assurance (14). Rural residents and Black/African American individuals have lower adherence to follow-up imaging, and thus, the lack of tracking systems might increase disparities (35-37). A recent study shows that an interpretable AI model leveraging individual- and neighborhood-level social determinants of health can predict LCS non-adherence among racially and ethnically diverse high-risk populations, highlighting how community context drives adherence disparities (38). The electronic health records (EHRs) combined with automation and/or AI algorithms can be used to trigger reminders for patients and providers about recommended follow-up appointments (39-41). Automation in this context refers to rule-based systems that automatically perform tasks, such as triggering reminders, without manual intervention. AI algorithms advance upon these rule-based methods to analyze patterns in patient data to make more dynamic, complex or personalized predictions.

Diversity in CT scanner technology and acquisition protocol optimization exist across LCS programs. This may increase disparities because CT technology has evolved over time (42). Large, academic centers have many CT scanner systems, usually with state-of-the-art technology. Small rural LCS centers may have access only to an older model CT scanner. Although CMS provides some guidance on CT protocol for LCS (mainly limiting to low radiation dose levels), there is a large amount of clinical variability in image quality and quality control.

The potential benefits for AI software programs begin with ensuring that the proper LCS CT scanning protocol is used for a given make and model of CT scanner. AI software can help ensure lower radiation exposure and improve image reconstruction, particularly benefiting screening programs in under-resourced centers with limited radiology support and older model CT scanners. This is important for LCS safety as CT image quality at a given radiation exposure can have implications on lung nodule detection and longitudinal monitoring (43).

CT image reconstruction has advanced from filtered back projection to iterative reconstruction, and most recently, deep learning methods. Filtered back projection is a fast, traditional technique that mathematically reverses the projection process to form images but can be prone to noise and artifacts. Iterative reconstruction can improve image quality by repeatedly refining the image estimate to better match the measured data. Most recently, AI reconstruction techniques can further improve image quality and allow radiation dose reduction. Studies found that deep learning reconstruction for ultra-LDCT (<1 mGy) improved image quality, nodule detection, and measurement accuracy compared to iterative reconstruction methods (44,45).

Telehealth can be integrated with AI to further address disparities resulting from screening centers using less experienced radiologists. Local diagnostic testing facilities can perform LDCT scans and have images interpreted remotely by experienced radiologists (18). More accurate readings from centralized readers, particularly supplemented by AI, could reduce unnecessary resource utilization, costs, and risks to patients. Mobile screening units combined with centralized AI-supported readings may also support quality screening in rural or underserved areas (18,46). Pilot studies have shown the feasibility of targeting rural and minority populations with mobile CT screening units that transmit images for central review (47-50) or incorporate AI assistive tools for CT evaluation (51). A New Hampshire study found that individuals in low socioeconomic status communities preferred mobile CT screening to a hospital-based program, particularly for overcoming travel-related barriers (49).

Enabling radiologists to increase the accuracy of nodule detection

AI-supplemented interpretations could improve the efficiency and accuracy of general radiologists in detecting and assessing nodules from screening images (52-57). Much research has been done on AI software tools for assisting with pulmonary nodule detection in CT imaging, also termed computer-assisted detection (CAD). Unfortunately, integrating early CAD systems into the clinical workflow was found to be more burdensome than assistive due to the high number of false nodule detections needing to be assessed by the radiologist (58). More recent, high-performing, deep learning approaches may feature concurrent detection with nodule segmentation (delineation of the boundaries of the detected nodule) and incorporate advanced nodule candidate assessment strategies capable of significantly reducing false positive detection rates while maintaining high detection sensitivity (55,59-61). Currently, though, the US Lung-RADS reporting scheme does not require nodule segmentation and assessment of nodule volume, as nodule diameter can be used (62). In Europe, the NELSON trial, which showed that LDCT screening reduced lung cancer mortality risk by 24% in men and 33% in women, incorporated nodule volume and volume doubling time (63). Investigators reported improvements over the NLST in specificity (98.3% vs. 73.4%) and slightly improved sensitivity (93.5% vs. 92.5%), supporting the potential advantage of nodule segmentation and volumetric assessment for LCS. Accordingly, the British Thoracic Society now recommends volumetry for LCS (64). An expert panel review article provides guidance for volumetry in the context of LCS, highlighting the importance of consistency in both CT acquisition settings and volumetric software for serial volumetric assessment (64).

The US Food and Drug Administration (FDA) has approved several chest CT nodule detection algorithms to serve as a second reader to radiologists (65). The functionality and use application beyond nodule detection vary across these commercially available tools, including two-dimensional or volumetric quantification, tracking over time, and/or radiomic or morphologic features of detected nodules (65). Successful clinical utility of nodule detection and segmentation AI software tools requires appropriate training and onboarding of radiologists (66).

A systematic review found that experienced radiologists using AI had increased sensitivity for nodule detection compared to unaided radiologists, though specificity was either unchanged or slightly decreased (53,57,67). The potential value for less experienced radiologists was evaluated in several studies. One study found that AI support increased interobserver agreement for nodule detection in NLST images when comparing experienced radiologists with radiology residents (68). Another study found that AI support significantly increased sensitivity for nodule detection with a non-significant increase in specificity among both experienced and less experienced radiologists (69).

Ensuring that AI can increase that specificity of nodule detection among less experienced radiologists is crucial for supporting expanded, equitable and high-quality screening efforts. Studies have shown that less experienced readers have higher false positive rates (70) and that community sites were more likely to report positive baseline screens than academic sites (32,71). False positives can increase costs and anxiety and lead to unnecessary diagnostic testing, including invasive procedures (33). Notably, data on over one million individuals captured by the ACR’s registry, 9.9% of baseline screens were classified as Lung-RADS 3 and 7.4% were classified as Lung-RADS 4. However, only 0.4% of those with a baseline Lung-RADS 3 were diagnosed with lung cancer compared to 2.6% to 19.9% for Lung-RADS 4 (34). Not reporting data to a central registry, though, will make it more difficult to determine whether screening centers are following expectations for appropriately performing and interpreting CT scans as well as tracking LDCT results, cancer diagnoses, and adherence with annual screening by geographic regions (17).

Enabling radiologists to better distinguish benign from malignant nodules

Most lung nodules detected with LDCT LCS are not cancerous. AI may be able to more accurately classify nodules to better guide biopsy decisions and to determine future lung cancer risk to personalize screening intervals (72). By improving the sensitivity and specificity of LDCT, AI models can lead to earlier cancer detection and to avoiding unnecessary invasive diagnostic procedures, thus reducing clinical harms, anxiety, and financial burdens.

AI approaches that provide lung cancer risk score for nodules detected with LDCT, also termed computer-aided diagnosis (CADx), are able to focus on the imaging features of the detected and segmented nodule, but also incorporate imaging features from beyond the detected nodule (73-75) or include the whole image (76,77). Approaches may provide a lung cancer risk score based on a CT-identified pulmonary nodule of interest or an overall risk score for the patient. Another way to utilize the lung cancer LDCT scan is to forecast the future risk of a lung cancer diagnosis, within a short (1 year) and extended period (2–6 years). Studies have reported high performance statistics for short-term prediction of lung cancer [area under the curve (AUC) of 0.86–0.95], which drops slowly for extended predictions (AUC of 0.74–0.75 at 6 years) and have included data in which no CT identifiable lesion was present (76-78). Such predictive capabilities could support personalized screening intervals, allowing low-risk patients to undergo less frequent scans, thereby reducing clinic visits, radiation exposure, and costs, without compromising timely diagnosis. High-risk individuals could be monitored more closely. However, to ensure equitable benefit, it is essential that personalized screening guidelines are applied consistently across populations. Without careful implementation, there is a risk that disparities could widen, particularly if minority patients are more likely to experience interval cancer due to inconsistent follow-up.

The Virtual Nodule Clinic (Optellum LTD) is the first FDA-cleared AI solution for lung cancer risk decision support for pulmonary nodules identified with CT. A recent small single-institution study (79) explored the performance of the Virtual Nodule Clinic lung cancer risk score compared to that of LungRADS and the Brock model [a multiparametric statistical risk tool based on clinical factors (80)]. This study incorporated 422 nodules (105 malignant), 5–30 mm, solid (95%), sub-solid (5%). The lung cancer risk score demonstrated superior AUC and sensitivity compared to LungRADS and the Brock model, but lower specificity. While these findings are promising, authors acknowledged the need to further validate performance in a more diverse cohort.

A recent systematic review evaluated how AI assistance for lung nodule assessment affected the performance of radiologists in detecting and assessing benign and malignant lung nodules using commercially available AI tools (57). Authors identified 17 studies evaluating radiologists’ performances in detecting lung nodules and predicting malignancy without and with AI. These studies showed that a radiologist achieved a higher sensitivity and AUC in detecting lung nodules with AI but a slightly lower specificity. For predicting malignancy, radiologists using AI achieved a higher sensitivity, specificity, and AUC. The review provided only limited details about the radiologist workflows using AI. Consequently, the practicality of using the AI tools in a busy clinical practice remains uncertain. Furthermore, while AI may further increase the predictive accuracy of LDCT (81), additional evidence needs to be generated from diverse populations beyond the homogeneous clinical trial subjects (18). A recent study highlighted this issue. Investigators developed a machine learning model that accurately predicted lung cancer risk in the NLST cohort (82). However, the model performed relatively poorly in a University of Illinois Health cohort, which had a substantially greater proportion of Black participants than were enrolled in NLST. Retraining the model in a subset with a greater number of Black participants improved performance.

Incidental findings and opportunistic screening

LDCT scans detected significant incidental findings in 7.5% of NLST participants, such as coronary artery calcification, emphysema, liver and renal masses, and aortic aneurysms, many of which would be considered clinically actionable (5,83). Henderson et al. found that the reporting of significant incidental findings and follow-up recommendations can vary by facility type (academic vs. community), radiologist training (cardiothoracic vs. general), facility location, and the LungRADs assessment (84). One post-hoc analysis of NLST data showed that AI could both increase detection of significant extrapulmonary findings and integrate these findings to accurately predict all-cause mortality (85). Additionally, investigators have also shown that applying AI to LDCT findings can be used to support opportunistic screening that would normally be performed with different testing modalities and at additional cost. Examples include deriving coronary calcium scores to predict the risk of cardiovascular disease events (86,87), quantifying COPD (88,89), and characterizing body composition and bone mineral density to identify patients with sarcopenia and osteoporosis (90-92). Using AI for more accurately detecting significant incidental findings and providing opportunistic screening could reduce the healthcare costs of screening and potentially lead to health benefits by offering interventions that reduce morbidity and mortality. AI systems could also aid in increased standardization of detection and follow-up recommendations for significant incidental findings across LCS providers, regardless of facility type and/or location. However, using AI for opportunistic screening or detecting incidental findings could exacerbate disparities for patients lacking adequate insurance coverage for additional diagnostic testing and treatment arising from LDCT findings. Furthermore, as with nodule detection and cancer prediction algorithms, AI must be trained on diverse populations.

Implementation considerations and future development needs

Implementing AI for widespread use in LCS faces important challenges even before considering its potential role in addressing disparities. As with any new technology, there are significant hurdles for obtaining FDA regulatory approval, including creating a workflow that is conducive to clinical practice [integration with EHR, picture archiving and communication systems (PACS), dictation tools]. Clinical radiology departments interested in approved AI technology need substantial resources, time, and effort for installation, training, and competency of use. A recent systematic review conducted by the National Institute for Health and Care Research in the United Kingdom evaluated the diagnostic performance and economics of using AI technology for detecting and analyzing lung nodules (93). They found that AI-assisted readings appeared to be cost-effective for LCS because they could improve nodule detection, decrease radiologist reading time, and reduce variability in measuring and managing lung nodules. However, cost-effectiveness estimates were considered highly unreliable due to the poor quality of available studies. Appropriate onboarding of radiologists for the use of the AI tool is vital to ensure alignment with approved processes and anticipated performance levels. For example, AI can be integrated into the radiologic workflow as a first reader (providing initial interpretation), a second reader (offering secondary review after the radiologist’s assessment), or as a concurrent reader (working in real-time alongside the radiologist), each requiring tailored training and clear understanding of the tool’s role in clinical decision making. A further challenge in translating research-developed AI solutions into the clinical workflow, which may be particularly burdensome to under-resourced imaging centers, is the low reimbursement for LDCT LCS. The process pipeline for FDA’s approach to AI software solutions has been rapidly evolving over recent years (94). Commercial companies need to invest significant resources to pursue approvals to transition promising solutions into regulatory-approved commercial software products that hospitals can purchase. The lack of additional billing codes to support AI technologies within the clinical workflow, combined with the low reimbursement for LDCT LCS, limits hospital investment in AI solutions.

Using AI to address disparities is also hampered by the lack of minority participation in clinical trials (i.e., Black individuals made up only 4% of participants in the NLST) and representation in existing data used for AI development. AI development must ensure that training/testing data adequately captures minority populations. Systems developed on largely White populations may provide sub-optimal performance on minority groups (in certain instances, sub-models may be required). Additionally, although only shown in the analysis of chest X-ray data and not yet with LDCT scans, AI may use demographic data to take shortcuts in analyzing medical images that bias the ability to accurately diagnose individuals based on race or gender (95). Thus, it will be important to develop machine learning models that are grounded in “fairness-aware” frameworks that can mitigate racial biases (96).

Conclusions

LDCT scanning has demonstrated substantial benefit in detecting early-stage lung cancer and reducing lung cancer mortality in screened populations. Despite these advantages, LDCT uptake remains low, particularly among rural and minority populations that face a higher burden of lung cancer. Recent guideline updates have expanded screening eligibility and removed the requirement for radiologists to have prior chest CT imaging experience, raising concerns about maintaining high-quality screening implementation.

AI offers promising solutions to support radiologists in efficiently and accurately performing LDCT screening. AI tools can assist with nodule detection, malignancy prediction, and the characterization of incidental findings. These capabilities are especially valuable for under-resourced populations. These tools may also help reduce unnecessary diagnostic procedures and improve overall screening quality.

However, integrating AI into clinical workflows presents challenges. AI tools need to be trained on diverse datasets that reflect the full spectrum of socioeconomic and ethnic groups at risk and eligible for screening. They should be constructed using fairness-aware frameworks to avoid perpetuating demographic biases. Successful implementation will require additional clinical and technological infrastructure, regulatory oversight to ensure fairness and accuracy, universal access to AI tools, and broad acceptance by both clinicians and patients.

Addressing these considerations could enable AI to improve the overall quality of the screening process and help reduce disparities related to variation in performing and interpreting screening examinations. Future research and development efforts must focus on enabling and documenting the effective and equitable integration of AI-assisted LCS with LDCT into routine clinical practice.

Acknowledgments

None.

Footnote

Peer Review File: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-1-0224/prf

Funding: This work was funded in part by the National Institutes of Health National Cancer Institute (R01CA267820, 2022). The funding agency had no involvement in writing this report or in making the decision to submit this article for publication.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-1-0224/coif). J.C.S. reports financial support from the National Institutes of Health National Cancer Institute and a stock relationship with VIDA Diagnostics. J.D.N. reports serving as a paid consultant and medical advisor for VIDA Diagnostics with financial support, stock, and stock options, acting as an advisor for Sanofi with financial support, receiving book royalties from Elsevier Inc., and funding from NIH grants. C.E.G. reports equity or stocks in CRISPR Therapeutics Inc., Beam Therapeutics Inc., Intellia Therapeutics Inc., and Editas Medicine; consulting or advisory roles with the National Comprehensive Cancer Network; board memberships with Guardant Health Inc., Roche, and Natera, Inc.; and funding grants from Genentech Inc. R.M.H. reports financial support from the National Institutes of Health National Cancer Institute. The authors have no other conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The manuscript was review of published literature and is not considered to be human subjects’ research. No IRB approval or consent form was necessary.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Siegel RL, Kratzer TB, Giaquinto AN, et al. Cancer statistics, 2025. CA Cancer J Clin 2025;75:10-45. [Crossref] [PubMed]
2. Torre LA, Siegel RL, Jemal A. Lung Cancer Statistics. In: Ahmad A, Gadgeel S. editors. Lung Cancer and Personalized Medicine. Switzerland: Springer; 2016:1-19.
3. Haddad DN, Sandler KL, Henderson LM, et al. Disparities in Lung Cancer Screening: A Review. Ann Am Thorac Soc 2020;17:399-405. [Crossref] [PubMed]
4. Atkins GT, Kim T, Munson J. Residence in Rural Areas of the United States and Lung Cancer Mortality. Disease Incidence, Treatment Disparities, and Stage-Specific Survival. Ann Am Thorac Soc 2017;14:403-11.
5. National Lung Screening Trial Research Team. Reduced lung-cancer mortality with low-dose computed tomographic screening. N Engl J Med 2011;365:395-409.
6. Moyer VAU.S. Preventive Services Task Force. Screening for lung cancer: U.S. Preventive Services Task Force recommendation statement. Ann Intern Med 2014;160:330-8.
7. Screening for Lung Cancer with Low Dose Computed Tomography (LDCT): CAG-00439N: Centers for Medicare and Medicaid Services; 2014. Available online: https://www.cms.gov/medicare-coverage-database/view/ncacal-decision-memo.aspx?proposed=N&NCAId=274
8. Bandi P, Star J, Ashad-Bishop K, et al. Lung Cancer Screening in the US, 2022. JAMA Intern Med 2024;184:882-91. [Crossref] [PubMed]
9. Wolf AMD, Oeffinger KC, Shih TY, et al. Screening for lung cancer: 2023 guideline update from the American Cancer Society. CA Cancer J Clin 2024;74:50-81. [Crossref] [PubMed]
10. Mazzone PJ, Silvestri GA, Souter LH, et al. Screening for Lung Cancer: CHEST Guideline and Expert Panel Report. Chest 2021;160:e427-94. [Crossref] [PubMed]
11. Jaklitsch MT, Jacobson FL, Austin JH, et al. The American Association for Thoracic Surgery guidelines for lung cancer screening using low-dose computed tomography scans for lung cancer survivors and other high-risk groups. J Thorac Cardiovasc Surg 2012;144:33-8. [Crossref] [PubMed]
12. The United States Census Bureau. 2020 Census Urban Areas Facts 2023. Available online: https://www.census.gov/programs-surveys/geography/guidance/geo-areas/urban-rural/2020-ua-facts.html
13. Febbo J, Little B, Fischl-Lanzoni N, et al. Analysis of Out-of-Pocket Cost of Lung Cancer Screening for Uninsured Patients Among ACR-Accredited Imaging Centers. J Am Coll Radiol 2020;17:1108-15. [Crossref] [PubMed]
14. Slatore CG, Golden SE, Thomas T, et al. "It's Really Like Any Other Study": Rural Radiology Facilities Performing Low-Dose Computed Tomography for Lung Cancer Screening. Ann Am Thorac Soc 2021;18:2058-66. [Crossref] [PubMed]
15. Maki KG, Talluri R, Toumazis I, et al. Impact of U.S. Preventive Services Task Force lung cancer screening update on drivers of disparities in screening eligibility. Cancer Med 2023;12:4647-54.
16. Aldrich MC, Mercaldo SF, Sandler KL, et al. Evaluation of USPSTF Lung Cancer Screening Guidelines Among African American Adult Smokers. JAMA Oncol 2019;5:1318-24. [Crossref] [PubMed]
17. Houston KA, Mitchell KA, King J, et al. Histologic Lung Cancer Incidence Rates and Trends Vary by Race/Ethnicity and Residential County. J Thorac Oncol 2018;13:497-509. [Crossref] [PubMed]
18. Rivera MP, Katki HA, Tanner NT, et al. Addressing Disparities in Lung Cancer Screening Eligibility and Healthcare Access. An Official American Thoracic Society Statement. Am J Respir Crit Care Med 2020;202:e95-e112.
19. Kunitomo Y, Bade B, Gunderson CG, et al. Evidence of Racial Disparities in the Lung Cancer Screening Process: a Systematic Review and Meta-Analysis. J Gen Intern Med 2022;37:3731-8. [Crossref] [PubMed]
20. Dwyer LL, Vadagam P, Vanderpoel J, et al. Disparities in Lung Cancer: A Targeted Literature Review Examining Lung Cancer Screening, Diagnosis, Treatment, and Survival Outcomes in the United States. J Racial Ethn Health Disparities 2024;11:1489-500. [Crossref] [PubMed]
21. Phulgirkar A, Kelso A, Ranganathan R, et al. The impact of rurality, socioeconomic status, and health literacy on lung cancer screening uptake: a systematic review protocol. Transl Lung Cancer Res 2025;14:4658-65. [Crossref] [PubMed]
22. US Preventive Services Task Force. Screening for Lung Cancer: US Preventive Services Task Force Recommendation Statement. JAMA 2021;325:962-70.
23. Screening for Lung Cancer with Low Dose Computed Tomography (LDCT): CAG-00439R: Centers for Medicare and Medicaid Services; 2022. Available online: https://www.cms.gov/medicare-coverage-database/view/ncacal-decision-memo.aspx?proposed=N&NCAId=304
24. Kazerooni EA, Armstrong MR, Amorosa JK, et al. ACR CT Accreditation Program and the Lung Cancer Screening Program Designation. J Am Coll Radiol 2016;13:R30-4. [Crossref] [PubMed]
25. Lung Cancer Screening Center Designation (Revised 7-22-2024): American College of Radiology; 2024. Available online: https://accreditationsupport.acr.org/support/solutions/articles/11000061040-lung-cancer-screening-center-designation-revised-7-22-2024
26. Meyer CA, Klein JS, Liubauskas R, et al. Cardiothoracic Radiologist Workload, Work Capacity, and Burnout Post-COVID: Results of a Survey From the Society of Thoracic Radiology. J Thorac Imaging 2023;38:261-9. [Crossref] [PubMed]
27. Tammemägi MC, Katki HA, Hocking WG, et al. Selection criteria for lung-cancer screening. N Engl J Med 2013;368:728-36. [Crossref] [PubMed]
28. Choi E, Ding VY, Luo SJ, et al. Risk Model-Based Lung Cancer Screening and Racial and Ethnic Disparities in the US. JAMA Oncol 2023;9:1640-8. [Crossref] [PubMed]
29. Lu MT, Raghu VK, Mayrhofer T, et al. Deep Learning Using Chest Radiographs to Identify High-Risk Smokers for Lung Cancer Screening Computed Tomography: Development and Validation of a Prediction Model. Ann Intern Med 2020;173:704-13. [Crossref] [PubMed]
30. Lee JH, Lee D, Lu MT, et al. Deep Learning to Optimize Candidate Selection for Lung Cancer CT Screening: Advancing the 2021 USPSTF Recommendations. Radiology 2022;305:209-18. [Crossref] [PubMed]
31. Henderson LM, Su IH, Rivera MP, et al. Prevalence of Lung Cancer Screening in the US, 2022. JAMA Netw Open 2024;7:e243190. [Crossref] [PubMed]
32. Sahar L, Douangchai Wills VL, Liu KK, et al. Using Geospatial Analysis to Evaluate Access to Lung Cancer Screening in the United States. Chest 2021;159:833-44. [Crossref] [PubMed]
33. Eberth JM, Bozorgi P, Lebrón LM, et al. Geographic Availability of Low-Dose Computed Tomography for Lung Cancer Screening in the United States, 2017. Prev Chronic Dis 2018;15:E119. [Crossref] [PubMed]
34. Kale MS, Wisnivesky J, Taioli E, et al. The Landscape of US Lung Cancer Screening Services. Chest 2019;155:900-7. [Crossref] [PubMed]
35. Lake M, Shusted CS, Juon HS, et al. Black patients referred to a lung cancer screening program experience lower rates of screening and longer time to follow-up. BMC Cancer 2020;20:561. [Crossref] [PubMed]
36. Sesti J, Sikora TJ, Turner DS, et al. Disparities in Follow-Up After Low-Dose Lung Cancer Screening. Semin Thorac Cardiovasc Surg 2020;32:1058-63. [Crossref] [PubMed]
37. Spalluto LB, Lewis JA, Samuels LR, et al. Association of Rurality With Annual Repeat Lung Cancer Screening in the Veterans Health Administration. J Am Coll Radiol 2022;19:131-8. [Crossref] [PubMed]
38. Lu KH, Xiao Y, Akhtar A, et al. Neighborhood matters: Predicting lung cancer screening adherence with explainable AI. Lung Cancer 2026;211:108882. [Crossref] [PubMed]
39. Liu TL, Hetherington TC, Stephens C, et al. AI-Powered Clinical Documentation and Clinicians' Electronic Health Record Experience: A Nonrandomized Clinical Trial. JAMA Netw Open 2024;7:e2432460. [Crossref] [PubMed]
40. Tai-Seale M, Baxter SL, Vaida F, et al. AI-Generated Draft Replies Integrated Into Health Records and Physicians' Electronic Communication. JAMA Netw Open 2024;7:e246565. [Crossref] [PubMed]
41. Kukhareva PV, Li H, Caverly TJ, et al. Lung Cancer Screening Before and After a Multifaceted Electronic Health Record Intervention: A Nonrandomized Controlled Trial. JAMA Netw Open 2024;7:e2415383. [Crossref] [PubMed]
42. Vonder M, Dorrius MD, Vliegenthart R. Latest CT technologies in lung cancer screening: protocols and radiation dose reduction. Transl Lung Cancer Res 2021;10:1154-64. [Crossref] [PubMed]
43. Avila RS, Krishnan K, Obuchowski N, et al. Calibration phantom-based prediction of CT lung nodule volume measurement performance. Quant Imaging Med Surg 2023;13:6193-204. [Crossref] [PubMed]
44. Jiang B, Li N, Shi X, et al. Deep Learning Reconstruction Shows Better Lung Nodule Detection for Ultra-Low-Dose Chest CT. Radiology 2022;303:202-12. [Crossref] [PubMed]
45. D'hondt L, Franck C, Kellens PJ, et al. Impact of deep learning image reconstruction on volumetric accuracy and image quality of pulmonary nodules with different morphologies in low-dose CT. Cancer Imaging 2024;24:60. [Crossref] [PubMed]
46. Lam S, Baldwin DR, Devaraj A, et al. A Game-Changing 20 Years: Progress and Future Directions in Lung Cancer Screening. J Thorac Oncol 2026;21:235-52. [Crossref] [PubMed]
47. Raghavan D, Wheeler M, Doege D, et al. Initial Results from Mobile Low-Dose Computerized Tomographic Lung Cancer Screening Unit: Improved Outcomes for Underserved Populations. Oncologist 2020;25:e777-81. [Crossref] [PubMed]
48. Headrick JR Jr, Morin O, Miller AD, et al. Mobile Lung Screening: Should We All Get on the Bus? Ann Thorac Surg 2020;110:1147-52. [Crossref] [PubMed]
49. Balata H, Tonge J, Barber PV, et al. Attendees of Manchester's Lung Health Check pilot express a preference for community-based lung cancer screening. Thorax 2019;74:1176-8. [Crossref] [PubMed]
50. Crosbie PA, Balata H, Evison M, et al. Implementing lung cancer screening: baseline results from a community-based 'Lung Health Check' pilot in deprived areas of Manchester. Thorax 2019;74:405-9. [Crossref] [PubMed]
51. Huang F, Lin X, Hong Y, et al. The feasibility and cost-effectiveness of implementing mobile low-dose computed tomography with an AI-based diagnostic system in underserved populations. BMC Cancer 2025;25:345. [Crossref] [PubMed]
52. Espinoza JL, Dong LT. Artificial Intelligence Tools for Refining Lung Cancer Screening. J Clin Med 2020;9:3860. [Crossref] [PubMed]
53. Geppert J, Asgharzadeh A, Brown A, et al. Software using artificial intelligence for nodule and cancer detection in CT lung cancer screening: systematic review of test accuracy studies. Thorax 2024;79:1040-9. [Crossref] [PubMed]
54. Huang G, Wei X, Tang H, et al. A systematic review and meta-analysis of diagnostic performance and physicians' perceptions of artificial intelligence (AI)-assisted CT diagnostic technology for the classification of pulmonary nodules. J Thorac Dis 2021;13:4797-811. [Crossref] [PubMed]
55. Gao C, Wu L, Wu W, et al. Deep learning in pulmonary nodule detection and segmentation: a systematic review. Eur Radiol 2025;35:255-66. [Crossref] [PubMed]
56. Forte GC, Altmayer S, Silva RF, et al. Deep Learning Algorithms for Diagnosis of Lung Cancer: A Systematic Review and Meta-Analysis. Cancers (Basel) 2022;14:3856. [Crossref] [PubMed]
57. Ewals LJS, van der Wulp K, van den Borne BEEM, et al. The Effects of Artificial Intelligence Assistance on the Radiologists' Assessment of Lung Nodules on CT Scans: A Systematic Review. J Clin Med 2023;12:3536. [Crossref] [PubMed]
58. Eadie LH, Taylor P, Gibson AP. A systematic review of computer-assisted diagnosis in diagnostic cancer imaging. Eur J Radiol 2012;81:e70-6. [Crossref] [PubMed]
59. Min Y, Hu L, Wei L, et al. Computer-aided detection of pulmonary nodules based on convolutional neural networks: a review. Phys Med Biol 2022;67: [Crossref]
60. Zheng S, Kong S, Huang Z, et al. A Lower False Positive Pulmonary Nodule Detection Approach for Early Lung Cancer Screening. Diagnostics (Basel) 2022;12:2660. [Crossref] [PubMed]
61. Liu B, Chi W, Li X, et al. Evolving the pulmonary nodules diagnosis from classical approaches to deep learning-aided decision support: three decades' development course and future prospect. J Cancer Res Clin Oncol 2020;146:153-85. [Crossref] [PubMed]
62. Christensen J, Prosper AE, Wu CC, et al. ACR Lung-RADS v2022: Assessment Categories and Management Recommendations. J Am Coll Radiol 2024;21:473-88. [Crossref] [PubMed]
63. Dawson Q. NELSON trial: reduced lung-cancer mortality with volume CT screening. Lancet Respir Med 2020;8:236. [Crossref] [PubMed]
64. Nair A, Dyer DS, Heuvelmans MA, et al. Contextualizing the Role of Volumetric Analysis in Pulmonary Nodule Assessment: AJR Expert Panel Narrative Review. AJR Am J Roentgenol 2023;220:314-29. [Crossref] [PubMed]
65. Milam ME, Koo CW. The current status and future of FDA-approved artificial intelligence tools in chest radiology in the United States. Clin Radiol 2023;78:115-22. [Crossref] [PubMed]
66. Ewals LJS, Heesterbeek LJJ, Yu B, et al. The Impact of Expectation Management and Model Transparency on Radiologists' Trust and Utilization of AI Recommendations for Lung Nodule Assessment on Computed Tomography: Simulated Use Study. JMIR AI 2024;3:e52211. [Crossref] [PubMed]
67. Quirk J, Mac Donnchadha C, Vaantaja J, et al. Future implications of artificial intelligence in lung cancer screening: a systematic review. BJR Open 2024;6:tzae035. [Crossref] [PubMed]
68. Peters AA, Wiescholek N, Müller M, et al. Impact of artificial intelligence assistance on pulmonary nodule detection and localization in chest CT: a comparative study among radiologists of varying experience levels. Sci Rep 2024;14:22447. [Crossref] [PubMed]
69. Hsu HH, Ko KH, Chou YC, et al. Performance and reading time of lung nodule identification on multidetector CT with or without an artificial intelligence-powered computer-aided detection system. Clin Radiol 2021;76:626.e23-32.
70. Tailor TD, Tong BC, Gao J, et al. A Geospatial Analysis of Factors Affecting Access to CT Facilities: Implications for Lung Cancer Screening. J Am Coll Radiol 2019;16:1663-8. [Crossref] [PubMed]
71. Tailor TD, Gutman R, An N, et al. Positive Screens Are More Likely in a National Lung Cancer Screening Registry Than the National Lung Screening Trial. J Am Coll Radiol 2025;22:644-52. [Crossref] [PubMed]
72. Silvestri GA, Jett JR. The Intersection of Lung Cancer Screening, Radiomics, and Artificial Intelligence: Can One Scan Really Predict the Future Development of Lung Cancer? J Clin Oncol 2023;41:2141-3. [Crossref] [PubMed]
73. Uthoff J, Stephens MJ, Newell JD Jr, et al. Machine learning approach for distinguishing malignant and benign lung nodules utilizing standardized perinodular parenchymal features from CT. Med Phys 2019;46:3207-16. [Crossref] [PubMed]
74. Masquelin AH, Alshaabi T, Cheney N, et al. Perinodular Parenchymal Features Improve Indeterminate Lung Nodule Classification. Acad Radiol 2023;30:1073-80. [Crossref] [PubMed]
75. Calheiros JLL, de Amorim LBV, de Lima LL, et al. The Effects of Perinodular Features on Solid Lung Nodule Classification. J Digit Imaging 2021;34:798-810. [Crossref] [PubMed]
76. Ardila D, Kiraly AP, Bharadwaj S, et al. End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography. Nat Med 2019;25:954-61. [Crossref] [PubMed]
77. Mikhael PG, Wohlwend J, Yala A, et al. Sybil: A Validated Deep Learning Model to Predict Future Lung Cancer Risk From a Single Low-Dose Chest Computed Tomography. J Clin Oncol 2023;41:2191-200. [Crossref] [PubMed]
78. Niu C, Lyu Q, Carothers CD, et al. Medical multimodal multitask foundation model for lung cancer screening. Nat Commun 2025;16:1523. [Crossref] [PubMed]
79. Piskorski L, Debic M, von Stackelberg O, et al. Malignancy risk stratification for pulmonary nodules: comparing a deep learning approach to multiparametric statistical models in different disease groups. Eur Radiol 2025;35:3812-22. [Crossref] [PubMed]
80. McWilliams A, Tammemagi MC, Mayo JR, et al. Probability of cancer in pulmonary nodules detected on first screening CT. N Engl J Med 2013;369:910-9. [Crossref] [PubMed]
81. González Maldonado S, Delorme S, Hüsing A, et al. Evaluation of Prediction Models for Identifying Malignancy in Pulmonary Nodules Detected via Low-Dose Computed Tomography. JAMA Netw Open 2020;3:e1921221. [Crossref] [PubMed]
82. Zakkar AJ, Perwaiz N, Harikrishnan V, et al. Hybrid Computer Vision Model to Predict Lung Cancer in Diverse Populations. JCO Clin Cancer Inform 2026;10:e2500041. [Crossref] [PubMed]
83. Gareen IF, Gutman R, Sicks J, et al. Significant Incidental Findings in the National Lung Screening Trial. JAMA Intern Med 2023;183:677-84. [Crossref] [PubMed]
84. Henderson LM, Chiles C, Perera P, et al. Variability in Reporting of Incidental Findings Detected on Lung Cancer Screening. Ann Am Thorac Soc 2023;20:617-20. [Crossref] [PubMed]
85. Marcinkiewicz AM, Buchwald M, Shanbhag A, et al. AI for Multistructure Incidental Findings and Mortality Prediction at Chest CT in Lung Cancer Screening. Radiology 2024;312:e240541. [Crossref] [PubMed]
86. de Vos BD, Lessmann N, de Jong PA, et al. Deep Learning-Quantified Calcium Scores for Automatic Cardiovascular Mortality Prediction at Lung Screening Low-Dose CT. Radiol Cardiothorac Imaging 2021;3:e190219. [Crossref] [PubMed]
87. van Velzen SGM, Lessmann N, Velthuis BK, et al. Deep Learning for Automatic Calcium Scoring in CT: Validation Using Multiple Cardiac CT and Chest CT Protocols. Radiology 2020;295:66-79. [Crossref] [PubMed]
88. Yeom JA, Kim KU, Hwang M, et al. Emphysema Quantification Using Ultra-Low-Dose Chest CT: Efficacy of Deep Learning-Based Image Reconstruction. Medicina (Kaunas) 2022;58:939. [Crossref] [PubMed]
89. Jin H, Heo C, Kim JH. Deep learning-enabled accurate normalization of reconstruction kernel effects on emphysema quantification in low-dose CT. Phys Med Biol 2019;64:135010. [Crossref] [PubMed]
90. Lenchik L, Barnard R, Boutin RD, et al. Automated Muscle Measurement on Chest CT Predicts All-Cause Mortality in Older Adults From the National Lung Screening Trial. J Gerontol A Biol Sci Med Sci 2021;76:277-85. [Crossref] [PubMed]
91. Fang Y, Li W, Chen X, et al. Opportunistic osteoporosis screening in multi-detector CT images using deep convolutional neural networks. Eur Radiol 2021;31:1831-42. [Crossref] [PubMed]
92. Pan Y, Shi D, Wang H, et al. Automatic opportunistic osteoporosis screening using low-dose chest computed tomography scans obtained for lung cancer screening. Eur Radiol 2020;30:4107-16. [Crossref] [PubMed]
93. Geppert J, Auguste P, Asgharzadeh A, et al. Software with artificial intelligence-derived algorithms for detecting and analysing lung nodules in CT scans: systematic review and economic evaluation. Health Technol Assess 2025;29:1-234. [Crossref] [PubMed]
94. Artificial Intelligence and Machine Learning in Software as a Medical Device: US Food and Drug Adminsitation (FDA). Available online: https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-and-machine-learning-software-medical-device
95. Yang Y, Zhang H, Gichoya JW, et al. The limits of fair medical imaging AI in real-world generalization. Nat Med 2024;30:2838-48. [Crossref] [PubMed]
96. Conahan P, Robinson LA, Le T, et al. Easy ensemble classifier-group and intersectional fairness and threshold (EEC-GIFT): a fairness-aware machine learning framework for lung cancer screening eligibility using real-world data. JNCI Cancer Spectr 2025;9:pkaf030. [Crossref] [PubMed]