What Is Lung-RADS and What Do the Categories Mean?

Lung-RADS, short for Lung CT Screening Reporting and Data System, is a standardized framework that radiologists use to classify findings on low-dose CT scans performed for lung cancer screening. Created by the American College of Radiology in 2014 and updated several times since, it assigns each screening scan a numbered category that tells the ordering physician how likely a detected nodule is to be cancerous and what should happen next. The system has dramatically reduced false-positive rates compared with earlier approaches, but its real-world performance depends on measurement consistency, nodule type, and the population being screened.

What the Categories Mean

Lung-RADS sorts every screening CT into one of several categories, each tied to a specific level of cancer suspicion and a recommended next step. Categories 1 and 2 are considered negative or benign. A category 1 scan is essentially clean, while category 2 covers nodules or findings that look benign, such as small calcified granulomas or perifissural lymph nodes. In a large population-based study, fewer than 0.2% of people assigned category 1 or 2 at baseline were diagnosed with lung cancer within the following year.

Category 3 sits in the “probably benign” zone. These are nodules large enough or shaped in a way that warrants a short-interval follow-up scan, typically in six months, rather than the standard annual return. The cancer yield for category 3 nodules is low but not negligible. One clinical practice study found that about 4% of category 3 nodules turned out to be malignant, while a separate screening consortium reported a 12-month cancer yield of roughly 1.7%.¹ The gap between those numbers reflects differences in study populations and how long patients were followed, but the message is the same: most category 3 nodules are harmless, yet they deserve closer watching.

Category 4 is where suspicion rises sharply. It is subdivided into 4A, 4B, and 4X, each carrying a progressively higher likelihood of malignancy. In clinical practice, the frequency of cancer has been reported at roughly 16% for 4A, 36% for 4B, and 77% for 4X.² Category 4 findings typically trigger additional workup: diagnostic CT, PET/CT imaging, or tissue sampling, depending on the subcategory and the patient’s overall risk profile. PET/CT has shown fair ability to separate malignant from benign nodules in this group, though its specificity is moderate, meaning some benign nodules still light up on the scan.³

How Lung-RADS Cuts Down on False Alarms

Before Lung-RADS existed, the landmark National Lung Screening Trial (NLST) demonstrated that low-dose CT screening saves lives, but it also flagged a staggering number of benign findings as “positive.” When researchers retroactively applied Lung-RADS criteria to the NLST data, the false-positive rate at baseline dropped from about 27% to roughly 13%, and after baseline it fell from about 22% to around 5%.⁴ That is a massive practical difference. Fewer false positives mean fewer unnecessary biopsies, less patient anxiety, and lower cost to the healthcare system.

These improvements have held up outside the controlled trial setting. In a diverse, underserved urban screening cohort, the Lung-RADS false-positive rate was about 10%, confirming that the system works in real community practice, not just on curated research populations.⁵ Still, the trade-off is worth noting. A stricter threshold for calling something positive also raises the possibility of missing an early cancer, so the system is tuned to balance sensitivity against specificity rather than maximize either one alone.

The 2022 Update

Lung-RADS has been revised twice since its original release. Version 1.1 came in 2019, and version 2022 arrived in November of that year.⁶ The 2022 update tackled several real-world headaches that radiologists had been running into. It added guidance on how to handle atypical pulmonary cysts, juxtapleural nodules (those sitting right against the chest wall, which are hard to measure accurately), and airway-centered nodules.⁷ It also addressed potentially infectious findings, a problem in regions where conditions like tuberculosis or fungal infections create nodules that mimic cancer on CT.

Another important change in version 2022 was the introduction of “stepped management” for nodules that are stable or shrinking over time. Under earlier versions, a nodule that stayed the same size through multiple follow-up scans could keep a patient locked in a cycle of short-interval CTs. Stepped management lets the follow-up interval gradually lengthen when a nodule shows no growth, eventually returning the patient to annual screening.⁸ The update also clarified how to define nodule growth, since different readers had been interpreting that criterion inconsistently.

When Radiologists Disagree

One of the less visible challenges with Lung-RADS is that two radiologists looking at the same scan don’t always assign the same category. A systematic review and meta-analysis of inter-reader agreement found that overall agreement was substantial, with a pooled kappa of 0.72, and that agreement on nodule size measurement was nearly perfect.⁹ Those sound like reassuring numbers, but the averages hide meaningful variation depending on how measurements are made and what types of nodules are involved.

When radiologists measured nodules manually with electronic calipers, agreement on Lung-RADS categories was noticeably lower (kappa of 0.66) compared with computer-aided semiautomated diameter measurements (0.91) or volume-based measurements (0.95).¹⁰ Nodule type also matters. Studies evaluating scans with all solid nodules had much higher agreement (kappa 0.85) than those with a mix including many subsolid nodules (kappa 0.55 when fewer than 30% were solid).

An earlier multi-reader study illustrated the real-world consequences. Among seven experienced radiologists reviewing the same 160 screening CTs, disagreement occurred in about 29% of reading pairs, and 8% of all pairs resulted in a management-level discrepancy, meaning one reader would recommend routine annual screening while another would send the patient for a diagnostic workup.¹¹ Surprisingly, most of the disagreements weren’t about how big a given nodule was. Three-quarters of the discrepancies happened because the two readers picked entirely different nodules as the most important one on the scan. This suggests that the biggest source of inconsistency is not measurement error but perceptual judgment about which finding matters most.

Measurement Methods and Semiautomated Tools

Because nodule size is the primary driver of Lung-RADS categorization, how you measure matters enormously. A nodule that measures 5 mm sits in a very different category than one that measures 7 mm, yet a millimeter or two of variation is easy to introduce when a radiologist eyeballs the diameter on a screen and places calipers manually. Semiautomated CT volumetry, which uses software to trace the three-dimensional boundaries of a nodule and calculate its volume, substantially improves consistency. In one study, over a third of nodules were classified into a lower Lung-RADS category when volumetry was used instead of manual diameter measurements, while only 3% moved to a higher category.¹²

The effect is even more pronounced for subsolid nodules, which are partly or entirely ground-glass in appearance and notoriously difficult to measure by hand. Computer-aided measurement of the solid component within a part-solid nodule raised the kappa for Lung-RADS classification from 0.65 (manual) to over 0.95 (semiautomated).¹³ Subsolid nodules carry a high risk of malignancy when they do turn out to be cancer, but the cancers they produce tend to grow slowly, which is why Lung-RADS allows longer follow-up intervals for them compared with solid nodules of the same size.¹⁴

Growth Assessment and What Volume Doubling Time Tells You

Growth is one of the strongest signals of malignancy in a lung nodule, and Lung-RADS uses it to bump a stable-looking nodule into a higher suspicion category. But growth is not an absolute indicator. In one screening study, the volume doubling time (VDT) of malignant solid nodules that were growing averaged about 204 days, compared with 386 days for benign growing nodules.¹⁵ Among growing solid nodules, 92% of malignant ones had a VDT under 400 days, but so did 58% of benign ones. That overlap means a growing nodule with a fairly fast doubling time could still turn out to be benign, and a handful of cancers grow slowly enough to fall outside the typical malignant window.

When the researchers looked at all growing nodules without separating by type, the difference in VDT between malignant and benign was not statistically significant, largely because subsolid cancers tend to double their volume much more slowly than solid ones. This underscores why Lung-RADS considers nodule type alongside size and growth rate rather than relying on any single criterion.

The S Modifier and Incidental Findings

Lung-RADS includes an “S” modifier that radiologists can append to any category to flag clinically significant incidental findings, things that are not lung nodules but were spotted on the scan and might need attention. In one screening program, 45% of participants received the S modifier at their baseline scan, with coronary artery calcification accounting for nearly half of the flagged findings.¹⁶ Other common incidentalomas included thyroid nodules, aortic abnormalities, and adrenal or renal masses.

The problem is consistency. In the same study, inter-radiologist agreement on whether the S modifier should be used was poor, with a kappa of only 0.2. Many findings that one reader flagged as clinically significant were described but left untagged by another. The researchers estimated that if reporting had been uniform, up to 69% of participants could have received the S modifier.¹⁷ Version 2022 attempted to provide clearer criteria for this modifier, but the underlying challenge remains: a low-dose chest CT reveals a lot of anatomy, and deciding which incidental findings are worth mentioning is inherently subjective.

Infectious Mimics in TB-Endemic Regions

Lung-RADS was developed and validated primarily in populations where tuberculosis is uncommon. When the system was applied in the Korean Lung Cancer Screening Project (K-LUCAS), researchers found that TB sequelae, the scarring and calcified nodules left behind by old infections, were associated with more positive screening results. Specificity for participants without TB sequelae was about 85%, compared with 80% for those who had evidence of prior TB on their scans.¹⁸ TB scarring was not itself associated with an increased risk of actual lung cancer in this cohort, so the additional positive results represented pure false positives driven by old infection residua mimicking suspicious nodules.

Similar issues arise in regions where fungal infections like histoplasmosis are common, particularly parts of the midwestern and southeastern United States. Granulomas from these infections can look indistinguishable from early-stage lung cancers on CT. Version 2022’s new guidance on potentially infectious findings was a direct response to this pattern, giving radiologists a framework for flagging a nodule as likely infectious without necessarily escalating it to a category 4 workup.

What Happens After a Positive Finding

For patients, receiving a “positive” Lung-RADS result, typically category 3 or above, is the beginning of a process rather than a diagnosis. Most of these nodules are not cancer, even in category 4A. The downstream workup can include short-interval CT follow-up, PET/CT scanning, or a tissue biopsy, and each step carries its own risks and costs. In one community-based screening program, about 10% of the invasive procedures performed led to a complication, with pneumothorax being the most common, occurring in 10 of the 226 procedures tracked.¹⁹ Six of those pneumothorax cases required a chest tube. Viewed as a fraction of all screened individuals, the procedural complication rate was well under 1%, but the number is not zero, and it is part of why keeping the false-positive rate low matters so much.

Cost-effectiveness analyses have compared different algorithms for managing category 4 nodules. Lung-RADS and the British Thoracic Society (BTS) guideline were both found to be cost-effective, with the BTS approach yielding slightly more quality-adjusted life-years, though at a higher incremental cost per year gained.²⁰ In a separate analysis comparing various triage approaches, thoracic radiologist visual assessment was the most accurate overall for distinguishing malignant from benign nodules, while the BTS algorithm outperformed both the American College of Chest Physicians guideline and a mathematical risk calculator used alone.²¹

Patient Anxiety and the Category 2 Problem

Even a “benign-appearing” result can be surprisingly stressful. Patients who receive a Lung-RADS category 2, which is technically a negative screen with a recommendation to return in 12 months, sometimes experience confusion, anxiety, and distress at learning that any nodule was found at all.²² There is some evidence that patients process a category 2 result very differently from a category 1, which may affect whether they return for their next annual scan. If the anxiety of “we found something but it’s probably nothing” drives a patient away from follow-up screening entirely, the system has inadvertently made them less safe. Clear, plain-language communication from the care team, explaining what a benign-appearing nodule actually means and why routine follow-up is all that is needed, appears to be one of the most important and most frequently underdelivered parts of the screening process.

Artificial Intelligence and Where the System Is Heading

AI-assisted tools are beginning to work alongside Lung-RADS rather than replace it. One approach combines a machine-learning malignancy risk score with the existing Lung-RADS category to refine the system’s predictions. Combining the AI score with Lung-RADS also appears to reduce unnecessary follow-up investigations. A cost analysis of a combined AI and Lung-RADS strategy at baseline screening found it could result in fewer follow-up procedures and substantial savings.²³ Meanwhile, the Lung-RADS system itself reduces false positives but introduces prolonged periods of uncertainty for people whose nodules fall into indeterminate categories, and some researchers have explored whether blood-based biomarkers could help shorten that limbo for patients stuck in surveillance cycles.²⁴

The meta-analysis on inter-reader agreement points in the same direction: computer-aided detection tools already bring category agreement close to near-perfect when used consistently.²⁵ Whether AI ultimately changes the category thresholds themselves or simply provides a second opinion that keeps radiologists more consistent, the trajectory seems to be toward layering computational tools on top of the existing Lung-RADS framework rather than starting from scratch. For patients, that could mean fewer ambiguous results and faster resolution of the question everyone wants answered: is this nodule something to worry about, or not.