CT Growth Measurements Reveal Which Fuzzy Lung Nodules Turn Invasive

One of the most common dilemmas in modern lung cancer screening is what to do with a subsolid pulmonary nodule, the hazy, ground-glass shadow that appears on a computed tomography scan but does not behave like a classic solid tumor. These nodules can represent anything from a harmless precursor lesion to a fully invasive adenocarcinoma, and telling them apart before surgery is one of radiology’s hardest problems. A new retrospective study published in BMC Medical Imaging by Xiaofeng Wu, Yang Tao, Silin Du, and Fajin Lv of the First Affiliated Hospital of Chongqing Medical University takes a systematic look at whether watching a nodule change over time, rather than judging it from a single scan, genuinely helps clinicians predict invasiveness. The answer, drawn from 354 surgically resected nodules, is nuanced: growth measurements track the biology of lung adenocarcinoma remarkably well, but their practical value depends on what is already visible on the most recent scan.

The research team assembled a cohort of 354 patients, each contributing exactly one baseline subsolid nodule that was later removed surgically, along with at least two CT examinations separated by a minimum of 365 days. The starting pool was far larger: 2,119 screened patients were narrowed down through strict eligibility criteria, excluding those with incomplete imaging, follow-up intervals shorter than one year, poor image quality, prior treatment of the target lesion, or nodules that could not be reliably tracked across scans. Thirty additional patients whose nodules were already solid at baseline were also excluded, ensuring the analysis focused purely on the subsolid spectrum. Pathology divided the final cohort into three biologically meaningful groups: 147 lesions classified as atypical adenomatous hyperplasia or adenocarcinoma in situ, the earliest preinvasive stages; 116 minimally invasive adenocarcinomas; and 91 invasive adenocarcinomas, the stage at which tumor cells have breached the surrounding tissue and gained the potential to spread.

The study’s central technical contribution lies in how it quantified change. Rather than relying on subjective radiologist impressions, the researchers calculated annualized volume growth, annualized mass growth, the change in consolidation-to-tumor ratio, and two doubling-time metrics, volume doubling time and mass doubling time. Volume growth captures how fast the nodule’s three-dimensional footprint expands; mass growth incorporates density, multiplying volume by tissue attenuation measured in Hounsfield units, so it reflects both enlargement and solidification. The consolidation-to-tumor ratio, a staple of Japanese lung cancer literature, compares the solid component of a nodule to its total diameter and serves as a proxy for how much of the lesion has transitioned from lepidic, surface-growing cells to a denser invasive core. Each of these metrics was computed across the full follow-up interval and expressed per year, allowing fair comparison between nodules monitored for different lengths of time.

The results reveal a striking biological gradient. Median annualized volume growth was 16.52 cubic millimeters per year for preinvasive AAH/AIS lesions, 91.34 for minimally invasive adenocarcinoma, and 345.10 for invasive adenocarcinoma, a more than twentyfold difference between the two extremes. Annualized mass growth followed the same pattern, rising from 0.01 grams per year in preinvasive lesions to 0.05 in minimally invasive disease and 0.28 in invasive tumors. Volume doubling time, the standard measure of how quickly a lesion doubles in size, shortened progressively across the spectrum: a leisurely 2,063.89 days for AAH/AIS, 934.65 days for minimally invasive adenocarcinoma, and 738.32 days for invasive disease. All of these comparisons were statistically significant at P less than 0.001, confirming that the pace of growth is tightly coupled to the pathological stage a nodule has reached by the time it is resected.

To translate these descriptive findings into clinical decision support, the team built a series of logistic regression models and validated them internally with 1,000 bootstrap resamples, a technique that estimates how much a model’s apparent performance is inflated by overfitting. Four model families were compared: a baseline model using only the initial CT findings, a dynamic model built on the longitudinal growth metrics, a primary parsimonious combined model, and a last-static CT model based on the most recent scan alone. The optimism-corrected area under the receiver operating characteristic curve, or AUC, was 0.871 for the baseline model, rising to 0.937 for the dynamic model, 0.935 for the combined model, and 0.941 for the last-static CT model. In other words, any model that incorporated the most recent imaging information, whether static or dynamic, substantially outperformed one anchored to the baseline scan.

The most provocative and clinically consequential finding, however, is what happened when the researchers tested whether growth metrics added anything on top of the last-static CT model. Adding annualized mass growth and consolidation-to-tumor ratio change to the model built from the final scan did not significantly improve either model fit or discrimination, with a likelihood-ratio test yielding P equal to 0.148 and a DeLong comparison of AUCs yielding P equal to 0.668. This is a sobering result for the longitudinal-monitoring paradigm. It suggests that by the time a nodule has been followed for a year or more, its most recent appearance already encodes most of the diagnostically relevant information, and the trajectory that led there contributes little that is statistically independent. For screening programs weighing the cost and anxiety of repeated scans, this implies that a well-characterized single late scan may be nearly as informative as the full time series.

Yet the story is not entirely closed. In an exploratory subgroup analysis restricted to 210 patients whose nodules showed a low solid-component burden at the last follow-up CT, a population where static appearance is least informative, annualized mass growth remained independently associated with invasive adenocarcinoma after adjustment for consolidation-to-tumor ratio change. For every 0.1 gram per year increase in mass growth, the odds of invasiveness rose by a factor of 2.478, with a 95 percent confidence interval of 1.530 to 4.013 and P less than 0.001. This makes physiological sense: in nodules that still look predominantly ground-glass, density accumulation may be the earliest visible signature of cells infiltrating the alveolar framework, detectable before any solid component becomes measurable on diameter-based assessment. The authors are careful to label this subgroup finding exploratory, noting that it involved only 11 invasive events and requires external validation before it can guide practice.

The methodological rigor of the study deserves attention. The team reported their prediction model analysis against the TRIPOD+AI checklist, assessed events-per-variable ratios to guard against overfitting, examined collinearity among predictors, performed sensitivity analyses around doubling-time calculations and alternative codings of consolidation progression, and quantified interreader agreement for both quantitative measurements and categorical imaging features. Supplementary analyses included integrated discrimination improvement and net reclassification improvement metrics, calibration curves corrected via bootstrap resampling, and decision curve analysis for the primary combined model. This level of transparency is increasingly expected in the radiomics and prediction-model literature, where optimistic performance estimates have historically been a persistent problem, and it strengthens confidence in the reported AUC values as realistic rather than inflated.

For clinicians managing the growing stream of subsolid nodules detected by lung cancer screening programs, the study offers both reassurance and a caution. The reassurance is that quantitative growth metrics are biologically meaningful: a nodule gaining 345 cubic millimeters per year and doubling in under two and a half years is very likely to harbor invasive disease, while one gaining a mere 16 cubic millimeters per year with a doubling time exceeding five years almost certainly does not. The caution is that these dynamic numbers should not be treated as a substitute for careful assessment of the current scan, because the last-static CT model performed at least as well as any growth-informed alternative in the full cohort. The practical implication may be that longitudinal monitoring is most valuable for deciding when a nodule has changed enough to warrant intervention, while the final preoperative judgment about invasiveness should rest on a meticulous read of the most recent examination, supplemented, perhaps, by mass-growth trends in nodules that still appear predominantly ground-glass.

Limitations frame the scope of these conclusions. The study was retrospective and single-center, conducted at a Chinese tertiary hospital under institutional review board approval with waived informed consent, and it included only nodules that were surgically resected, a selection that inherently enriches the cohort for suspicious lesions and excludes the many indolent nodules managed by surveillance alone. The authors themselves emphasize that the subgroup finding for annualized mass growth is exploratory and requires external validation. Nonetheless, by quantifying the growth signature of the entire lung adenocarcinoma continuum in 354 pathologically confirmed nodules, and by honestly demonstrating where longitudinal data do and do not add diagnostic value, the Chongqing team has provided a rigorous empirical foundation for the next generation of nodule-management algorithms, one that future prospective studies across multiple centers will be able to build upon and test.

Subject of Research: Longitudinal CT growth metrics for predicting invasive lung adenocarcinoma in subsolid pulmonary nodules

Article Title: Longitudinal CT-derived growth metrics associated with invasive lung adenocarcinoma in subsolid pulmonary nodules

Article References: Longitudinal CT-derived growth metrics associated with invasive lung adenocarcinoma in subsolid pulmonary nodules. (n.d.). https://doi.org/10.1186/s12880-026-02874-3

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02874-3

Keywords: lung adenocarcinoma, subsolid nodule, pulmonary nodule, computed tomography, volume doubling time, mass doubling time, consolidation-to-tumor ratio, nodule growth, invasiveness, prediction model, radiology, lung cancer screening