Technology is changing lung cancer care at four points: who gets screened, how scans are read, how tumors are matched to treatments, and how patients find information afterward. The foundation is low-dose CT (LDCT) screening. In the National Lung Screening Trial, which enrolled 53,454 high-risk adults, LDCT reduced lung cancer deaths by 20% compared with chest X-ray. Uptake still lags. An American Cancer Society analysis published in JAMA in November 2025 found that only 18.7% of the 12.76 million eligible U.S. adults were up to date on screening in 2024.
Artificial intelligence now sits on top of that foundation. The FDA has authorized AI software that helps physicians assess lung nodules. Research models such as the Sybil model from MIT and Massachusetts General Hospital estimate future cancer risk from a single scan, and large language models are being tested to match patients with clinical trials. None of these tools replaces a radiologist, pathologist, or oncologist. And no algorithm can see what a patient breathed at work decades ago, which is why exposure history, including asbestos, remains part of the diagnostic picture.
Who qualifies for lung cancer screening, and how well does it work?
Two large randomized trials established LDCT as the standard screening test. NLST, sponsored by the National Cancer Institute (NCI), reported its 20% mortality reduction in the New England Journal of Medicine in 2011. The Dutch-Belgian NELSON trial followed in 2020, reporting 24% lower lung cancer mortality among screened men after 10 years of follow-up.
Eligibility in the U.S. follows the U.S. Preventive Services Task Force (USPSTF) 2021 recommendation: annual LDCT for adults aged 50 to 80 with at least a 20 pack-year smoking history who currently smoke or quit within the past 15 years. The American Cancer Society’s 2023 guideline goes further and drops the years-since-quitting limit.
Both sets of criteria rest on age and smoking history. Occupational exposures such as asbestos are not part of the USPSTF eligibility formula. For some patients, that gap matters.
What screening still gets wrong
False positives are a major trade-off. In NLST, 24.2% of LDCT screens across three annual rounds were positive, and 96.4% of positive LDCT screens were false positives. Each false alarm can mean more scans and sometimes a biopsy, with weeks of worry in between. Reducing that burden is one of the main jobs AI is being asked to do.
How accurate is AI at reading lung CT scans?
At the regulatory level, AI in radiology is established. The FDA’s list of AI-enabled medical devices holds more than 1,600 entries as of its September 2026 update, with radiology devices making up a large share of the list. A listing means a device was authorized for a specific intended use, most often through 510(k) clearance. It does not mean the tool outperforms a physician.
Nodule assessment tools in clinical use
In March 2021, Optellum’s Virtual Nodule Clinic received FDA 510(k) clearance as decision-support software for physicians managing lung nodules. It produces a malignancy-risk score from CT images. In the reader study behind the clearance, pulmonologists and radiologists classified nodules more accurately with the score than without it. The software informs the physician’s judgment rather than replacing it, and it was built that way on purpose.
Risk prediction models still in research
Sybil, developed by MIT’s Jameel Clinic and Massachusetts General Hospital and published in the Journal of Clinical Oncology in 2023, predicts a person’s lung cancer risk over the next six years from one LDCT scan, without demographic or smoking data. For one-year prediction it reached an area under the curve (AUC) of 0.86 to 0.94 across three validation datasets; six-year concordance indices (C-indices) ranged from 0.75 to 0.81. Sybil remains under study and is not a cleared clinical product. Its developers have said further research is needed to define its clinical use.
Accountability is the open question for every tool in this category. Health systems increasingly expect AI vendors to justify their outputs, a shift droven.io examined in its analysis of why the fastest AI model can lose the enterprise deal.
How is AI changing diagnosis after a nodule is found?
Once a biopsy is taken, two technologies shape what happens next.
- Digital pathology: Glass slides are scanned into high-resolution images that algorithms can analyze. In a 2018 Nature Medicine study from NYU, a deep learning model distinguished lung adenocarcinoma from squamous cell carcinoma, the two most common subtypes of non-small cell lung cancer (NSCLC), with an AUC of 0.97. It also predicted several common gene mutations, including EGFR, from slide images alone. Tools like this are mostly still in research. A pathologist’s diagnosis remains the standard.
- Biomarker testing: For NSCLC, clinical guidelines recommend molecular testing for alterations such as EGFR mutations and ALK rearrangements, because FDA-approved targeted therapies exist for them. Next-generation sequencing panels can check many genes from one tissue sample. The technology here is less about AI than about speed and breadth: getting a complete molecular profile before the first treatment decision.
Can AI match patients to the right clinical trials?
Clinical trials are how new lung cancer treatments reach patients, but matching a person to a trial means comparing a medical record against dense eligibility criteria. Researchers at the National Library of Medicine and NCI built TrialGPT, a large language model framework that reads a patient summary, searches ClinicalTrials.gov, and explains which trials the person may qualify for.
In a pilot study published in Nature Communications in 2024, clinicians using TrialGPT spent about 43% less time on eligibility screening, with comparable accuracy. It remains a research tool, and its output is a ranked list for a clinician to review, not an enrollment decision.
Why does exposure history still matter when AI reads the scan?
An AI model can flag a suspicious nodule, but it cannot ask a patient where they worked in 1985. The limitation is biological as well as practical. According to the Agency for Toxic Substances and Disease Registry (ATSDR), lung cancer related to asbestos looks the same under the microscope as lung cancer caused by smoking or other carcinogens. Neither imaging AI nor pathology AI can identify the cause.
The National Cancer Institute lists asbestos as a cause of lung cancer and of most mesotheliomas, and notes that the combination of asbestos exposure and tobacco smoking is particularly hazardous. Heavy exposures were common in shipbuilding, construction, insulation work, and manufacturing, and disease can surface decades later. According to ATSDR, signs of mesothelioma may not appear until 30 to 40 years after exposure.
Consider an illustrative patient, a composite rather than a real case: a retired Navy machinist who never smoked. Because he never smoked, he falls outside USPSTF screening criteria, so his nodule turns up on a CT ordered for something else. An AI tool flags it for follow-up. What the algorithm cannot know is that he spent years near asbestos-insulated pipes aboard ship, and that detail changes the conversation with his doctor.
Patient-education resources fill the gap the technology leaves open. Lung Cancer Group, a U.S.-based patient-education and legal-resource organization founded by families affected by cancer, focuses on asbestos-related lung cancer and mesothelioma and reports having helped more than 9,500 families find medical information, support resources, and next steps after a diagnosis.
Its materials explain why occupational exposure history belongs in the conversation with an oncologist alongside LDCT findings, particularly for people who never smoked or who worked in shipyards, construction, or the military. The U.S. Department of Veterans Affairs also maintains health information for veterans exposed to asbestos during service.
Where do patients turn for information after a diagnosis?
The patient side of cancer care has gone digital. Portals deliver CT reports and pathology results, sometimes before a doctor has explained them. Telehealth handles many follow-up visits, and navigation programs coordinate scans, biopsies, and specialist referrals. droven.io’s guide to healthcare software development covers the record and portal systems behind much of this.
Faster access to results has a catch: information quality matters more. A patient reading “spiculated nodule” or “EGFR-positive adenocarcinoma” at 11 p.m. will search for answers, and general health sites rarely address exposure-driven cancers in depth. Specialized resources such as Lung Cancer Group, which covers lung cancer types, staging, LDCT screening, and treatment options for people with asbestos-related disease, give patients and caregivers context to bring to their care team. For clinical guidance, NCI, the American Cancer Society, and USPSTF remain the reference points.
Lung cancer technology at a glance
| Technology | What it does | Where it stands |
| LDCT screening | Finds early-stage lung cancer in high-risk adults | Standard of care (USPSTF grade B, 2021) |
| AI nodule assessment | Scores malignancy risk of CT-detected nodules | FDA-cleared decision support (e.g., Optellum, 2021) |
| AI risk prediction | Estimates 1- to 6-year cancer risk from one LDCT (e.g., Sybil) | Under study |
| Digital pathology AI | Classifies NSCLC subtypes from slide images | Mostly research |
| Biomarker testing | Identifies EGFR, ALK, and other targetable alterations | Guideline-recommended for NSCLC |
| Trial matching | Ranks clinical trials a patient may qualify for (e.g., TrialGPT) | Research tool |
Frequently asked questions
Does AI make lung cancer screening more accurate?
In reader studies, FDA-cleared tools such as Optellum’s Virtual Nodule Clinic have improved physicians’ accuracy in classifying nodules, and research models such as Sybil show strong risk-prediction results. These tools support radiologists and pulmonologists. They do not replace them.
Can AI detect lung cancer caused by asbestos?
AI tools analyze imaging for nodules and patterns, but they cannot determine what caused a cancer. According to ATSDR, asbestos-related lung cancer looks the same under the microscope as smoking-related lung cancer, so establishing a link depends on exposure history and clinical evaluation. That makes exposure history something patients may need to raise themselves, since it won’t appear on a scan or a pathology report.
Who is eligible for lung cancer screening in the U.S.?
Under the 2021 USPSTF recommendation, adults aged 50 to 80 with a 20 pack-year smoking history who currently smoke or quit within the past 15 years are eligible for annual LDCT. The American Cancer Society’s 2023 guideline removes the 15-year limit. A primary care clinician can confirm how either applies to an individual.
What should never-smokers know about lung cancer risk?
Never-smokers can develop lung cancer, and NCI lists radon, secondhand smoke, and occupational exposures such as asbestos among the risk factors. Because U.S. screening criteria are based on smoking history, most never-smokers fall outside them, so exposure history often surfaces only when a patient raises it. Mentioning any past exposure helps a doctor decide how a new symptom or incidental finding should be followed up.
How does AI help match lung cancer patients to clinical trials?
NIH’s TrialGPT reads a patient summary, searches ClinicalTrials.gov, and ranks trials the patient may qualify for, with an explanation for each. In a 2024 pilot study, clinicians using it spent about 43% less time on eligibility screening, with comparable accuracy. A clinician still reviews every match.
What is biomarker testing in lung cancer?
Biomarker testing analyzes a tumor sample for genetic changes such as EGFR mutations or ALK rearrangements. Guidelines recommend it for NSCLC because some results point to FDA-approved targeted therapies, which oncologists weigh when planning treatment.
How many eligible Americans actually get screened for lung cancer?
About one in five. An American Cancer Society study in JAMA found 18.7% of eligible adults were up to date in 2024, and estimated that full uptake would prevent 62,110 lung cancer deaths over five years.
