Preoperative CT-based artificial intelligence-derived quantitative parameters and imaging features for predicting the invasiveness of histologically confirmed subcentimeter adenocarcinomatous nodules: a two-center study

Introduction

Lung cancer, the leading cause of cancer-related death worldwide, has the highest incidence among cancers in men and the second highest incidence in women after breast cancer (1,2). In 2021, The World Health Organization (WHO) reclassified adenocarcinoma in situ (AIS) and atypical adenomatous hyperplasia (AAH) as precursor glandular lesions rather than retaining them in the adenocarcinoma category. Meanwhile, minimally invasive adenocarcinoma (MIA) and invasive adenocarcinoma (IAC) remained classified as adenocarcinoma (3). Preoperative computed tomography (CT) assessment of lung adenocarcinoma aggressiveness affects clinical decision-making (4). Studies have shown that patients with AIS or MIA have relatively good clinical prognosis, with 5-year survival rates close to 100% (5), while those with IAC have a 5-year survival rate of 74.6% (6). However, these studies (7-9) were primarily based on the old lung tumor classification criteria, which now need to be validated in clinical work according to the new lung tumor classification criteria. The update of lung adenocarcinoma classification has led to more refined, individualized, and clinically oriented assessment of subcentimeter nodules in radiology.

With the improvement of human health awareness, the popularization of lung cancer screening in high-risk groups, and the development of medical imaging technology, the detection rate of small lung nodules has increased significantly (10). In the National Lung Screening Trial involving high-risk participants, subcentimeter pulmonary nodules were defined as those measuring ≤10 mm. Among all pulmonary nodules measuring ≥4 mm, 82.4% were 4–10 mm in diameter (11). In a study employing low-dose CT screening, although most of the detected pulmonary nodules were benign, some of them were pathologically diagnosed as lung cancer (12). Subcentimeter lung nodules are less likely to be definitively diagnosed by positron emission tomography or biopsy than are larger lung nodules because of the limited spatial resolution of positron emission tomography imaging, higher false-negative rates associated with small lesion size, and sampling difficulties during biopsy procedures (13,14). Detection of small nodules can result in overdiagnosis and overtreatment, with associated patient anxiety and unnecessary interventions (15-18). In small lung nodules, especially subcentimeter lung nodules, subtle imaging signs and characteristics are often overlooked due to the limitations in obtaining useful information with the naked eye alone, leading to challenges in the diagnosis of these nodules. Therefore, there is an urgent need to improve the accuracy of preoperative assessment of the invasiveness of subcentimeter pulmonary nodules in clinical practice.

The application of artificial intelligence (AI) in thoracic imaging, particularly in the detection and characterization of pulmonary nodules, has greatly advanced diagnostic accuracy and efficiency (19). AI not only improves the image quality and further reduces the radiation dose to patients but also helps radiologists to improve the diagnostic accuracy and efficiency in various aspects of lesion detection and diagnosis (20,21). Manual quantitative parameter assessment has shortcomings of as manual measurement errors, poor repeatability, and limited parameters measured. In contrast, the application of AI-assisted diagnostic tools renders the measurement and assessment of lesions more accurate. Especially in the measurement of volume and mass, there is a greater accuracy improvement compared with the formula calculation of manual measurement (22). Additionally, AI tools are able to extract quantitative parameters such as entropy, kurtosis, skewness, compactness, and sphericity, which quantify image texture and shape characteristics reflecting underlying tumor heterogeneity and structural complexity. These parameters have been shown to correlate with tumor invasiveness in recent studies (2,23).

Despite advances in imaging and AI analysis, few studies have combined AI-derived quantitative parameters with conventional CT qualitative features to predict invasiveness in subcentimeter pulmonary nodules. Therefore, the aim of this study was to examine the relationship between preoperative quantitative AI parameters, conventional CT imaging features, and the invasiveness of subcentimeter adenocarcinomatous nodules. A quantitative model based on AI parameters, a qualitative model based on CT radiographic features, and a combined model was also developed to predict the invasiveness of subcentimeter adenocarcinomatous nodules. We present this article in accordance with the TRIPOD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-650/rc).

Methods

Patients

This retrospective study was approved by the Ethics Committee of Shanghai Changzheng Hospital (approval No. 2022SL070) and Kunshan Third People’s Hospital (approval No. kssy2021-45). Due to the retrospective nature of the analysis and the anonymity of the data, the requirement for informed patient consent was waived. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

This study consecutively enrolled 281 patients admitted to Shanghai Changzheng Hospital between January 2021 and December 2022 and 56 patients admitted to Kunshan Third People’s Hospital between January 2021 and December 2022. The missing data were imputed via the mean value or multiple imputation method (24). Patients from Shanghai Changzheng Hospital were divided into a training set (225 cases) and a validation set (56 cases) at a ratio of 8:2, with the random seed set at 111. The 56 patients from Kunshan Third People’s Hospital served as an external test set.

The inclusion criteria (Figure 1) were as follows: (I) complete and clear chest CT images; (II) at least one pulmonary nodule with a diameter <10 mm; and (III) available postoperative histopathology results. Meanwhile, the exclusion criteria were as follows: (I) difficulty in clearly depicting the border due to collapsed or obstructive changes in the lung parenchyma surrounding the nodule; (II) puncture biopsy, radiotherapy, surgery, or other interventions conducted prior to CT scanning; and (III) presence of distant metastases.

Figure 1 Flowchart of patient enrollment. CT, computed tomography.

Examination methods

At Shanghai Changzheng Hospital, patients were scanned with am Aquilion 16-slice spiral CT scanner (Toshiba, Tokyo, USA), a Light Speed VCT 64-slice spiral CT scanner (GE HealthCare, Chicago, IL, USA), and a Brilliance 64-slice spiral CT scanner (Philips, Amsterdam, the Netherlands); meanwhile, patients at Kunshan Third People’s Hospital were scanned with a Light Speed VCT 64-slice spiral CT scanner. The CT scanner parameters were as follows: a tube voltage of 120 kV, a tube current of 150–250 mAs or automatic tube current adjustment, a scanning layer thickness of 5 mm, a reconstruction layer thickness of 0.625–1 mm, and a scanning matrix of 512×512. All patients were required to undergo breath-holding training before the scan and breath-holding under free breath during the scan. The scan was from the tip of the lung to the base of the lung, covering both armpits and the chest wall. Moreover, prior to feature determination, we performed rigorous standardized preprocessing on all enrolled CT images, including pixel spacing resampling and image normalization. Image reconstruction was implemented with standard or sharp algorithms. The CT images of patients with AIS, MIA, and IAC are shown in Figure 2.

Figure 2 CT images of patients with AIS, MIA, and IAC. (A) A 31-year-old female patient with pGGN AIS. (B) A 40-year-old male patient with mGGN MIA. (C) A 77-year-old male patient with mGGN IAC. AIS, adenocarcinoma in situ; CT, computed tomography; IAC, invasive adenocarcinoma; mGGN, mixed ground-glass nodule; MIA, minimally invasive adenocarcinoma; pGGN, pure ground-glass nodule.

Qualitative feature analysis

CT images from the two centers were retrospectively analyzed by two radiologists who were blinded to the patients’ pathological findings. Radiologist 1, with 15 years of experience in chest CT imaging diagnosis, and radiologist 2, with 10 years of experience, independently reviewed the images with a window width of 1,500 Hounsfield unit (HU) and a window position of –600 HU to assess the qualitative imaging features. Any discrepancies were resolved via discussion. They recorded key qualitative features such as lesion density (pure ground glass, mixed ground glass, or solid), shape (regular or irregular), boundary profile (clear or fuzzy), lobulation (present or absent), burr (present or absent), vacuole (present or absent), bronchiolar sign (present or absent), pleural indentation (present or absent), and vascular sign (present or absent). In cases of disagreement, a senior radiologist, with 25 years of experience in chest imaging diagnosis, served as the arbitrator.

Quantitative feature analysis

For each detected nodule, the system generated an automated region of interest (ROI) based on the segmentation boundary. Quantitative imaging features and radiomic parameters were extracted from these AI-defined ROIs for further analysis.

Digital Imaging and Communications in Medicine (DICOM) images from the two centers were imported into the lung nodule diagnosis software (LungDoc v. 6.21, Shukun Technology, Beijing, China). The LungDoc v. 6.21 Research Portal is an AI platform that integrates deep learning and machine learning algorithms. Its AI-assisted detection software for lung nodules, centered on the vendor’s proprietary deep-learning model, has been granted China National Medical Products Administration (NMPA) class III certification. Quantitative parameters, including mean diameter, longest diameter, vertical diameter, mean CT value, maximum CT value, minimum CT value, median, standard deviation, entropy, kurtosis, skewness, compactness, sphericity, maximum layer area, nodule mass, and total nodule volume, were automatically extracted by the AI analysis platform without manual adjustment. Radiologist 3 recorded the automatically generated results for subsequent statistical analysis while being blinded to both the pathological findings and the qualitative imaging assessments. The qualitative and quantitative analyses were performed independently to ensure methodological consistency.

Model construction

Univariate and multivariate logistic regression analyses, with stepwise regression, were employed to evaluate both qualitative and quantitative variables. For quantitative CT features, odds ratios (ORs) were calculated per unit increase of each variable, corresponding to per millimeter (mm) for diameter-related features, per HU for CT attenuation features, and per one-unit increment for texture and morphological parameters. This process identified independent predictors and facilitated the development of separate qualitative and quantitative models. Subsequently, a combined model was constructed by integrating age with the independently identified predictive qualitative and quantitative variables. Finally, this combined model was visualized with a nomogram to enhance interpretability.

Statistical analysis

In this study, R software version 4.3.1 (The R Foundation for Statistical Computing) was used for statistical analysis, with the significance level set at P<0.05. For categorical variables, the chi-squared test was employed to compare differences between groups. For continuous variables, the t-test was used for difference analysis. Additionally, the diagnostic capability of the model was assessed by plotting the receiver operating characteristic (ROC) curve, and the area under the curve (AUC) was calculated, along with performance metrics such as accuracy, precision, recall, and F1 score. Decision curve analysis (DCA) was used to determine the clinical application value of the three models, and the DeLong test was used to compare the diagnostic performance of the three models.

Results

Participant characteristics

This study included 337 patients from Shanghai Changzheng Hospital and Kunshan Third People’s Hospital, and their relevant clinical features are summarized in Table 1. The baseline characteristics of patients from Shanghai Changzheng Hospital and those from Kunshan Third People’s Hospital were comparable across all demographic and clinical variables (all P values >0.05). The results indicated significant differences in age between the non-infiltrating and infiltrating groups (P<0.05). However, there were no statistically significant differences between the two groups regarding sex, smoking status, history of other tumors, or family history of lung cancer.

Univariate and multivariate analyses of qualitative and quantitative variables

Table 2 details the results of the univariate and multivariate analyses of qualitative variables, while Table 3 presents the univariate analysis of quantitative variables. Additionally, the correlation between qualitative and quantitative features is visually represented in the string diagram shown in Figure 3.

Figure 3 Chord diagram showing the correlations between qualitative and quantitative features. (A) Chord diagram of the correlations among the 9 qualitative features. (B) Chord diagram of the correlations among the 16 quantitative features. CT, computed tomography.

In the analysis of qualitative features, we observed that increased lesion density [OR =6.776; 95% confidence interval (CI): 3.958–12.077], irregular morphology (i.e., lesion not round or oval in shape; OR =0.193, 95% CI: 0.069–0.465), and leaflet (OR =5.531; 95% CI: 3.074–10.272) were significantly associated with distinguishing between invasive and noninvasive lung cancer. In the univariate analysis, the presence of burrs (OR =5.109; 95% CI: 2.522–11.11; P<0.001), vacuoles (OR =2.140; 95% CI: 1.159–4.028; P=0.016), pleural depressions (OR =3.181; 95% CI: 1.752–5.926; P<0.001), and vascular signs (OR =4.802; 95% CI: 2.710–8.717; P<0.001) were significantly associated with invasive lesions. However, these features did not remain statistically significant in the multivariate model (all P values >0.05), indicating potential collinearity with stronger predictors such as density, irregular morphology, and leaflet signs. Notably, density, irregular morphology, and leaflet were identified as independent predictors of invasive lung cancer.

In the quantitative feature analysis, significant differences were found between invasive and noninvasive lung cancer in parameters such as mean diameter, longest diameter, vertical diameter, mean CT value, maximum CT value, minimum CT value, median, standard deviation, entropy, kurtosis, skewness, compactness, maximum layer area, nodule mass, and total nodule volume. Multivariate analysis further confirmed maximum diameter (OR =1.490; 95% CI: 1.241–1.827), entropy (OR =3.048; 95% CI: 1.859–5.224), and nodule mass (OR =1.017; 95% CI: 1.012–1.022) as independent predictors of invasive lung cancer.

Evaluation of predictive performance of the model

In this study, logistic regression analysis was employed to construct qualitative, quantitative, and combined models, and in the training set, the AUC for these models was 0.818, 0.849, and 0.901, respectively (Table 4). Calibration plots of the three models on the training set show a good consistency between the predicted probabilities and the observed probabilities, indicating that the models have a good fitting effect. The performance of the combined model was significantly superior to that of the single qualitative and quantitative models (P<0.05), with the corresponding ROC curve displayed in Figure 4A. Similar results were observed in both the validation set and the external test set, as shown in Figure 4B,4C, respectively. As shown in Figure 5A-5C, the combined model achieved the highest net benefit across clinically relevant threshold probabilities (0.1–0.8) in the training, validation, and external testing cohorts, indicating superior potential for guiding individualized clinical decision-making, surgical resection, or imaging follow-up. The visual nomogram for the combined model is depicted in Figure 6A, while its calibration graph in Figure 6B confirms the model’s effective predictive accuracy. The calibration intercept was 1.38×10⁻¹³ (95% CI: −0.389 to 0.360), and the slope was 1.00 (95% CI: 0.804–1.283), indicating no significant bias and good overall agreement between the predicted and observed probabilities.

Figure 4 Comparison of the ROC curves of the three models in the (A) training, (B) validation, and (C) test sets. AUC, area under the curve; ROC, receiver operating characteristic.

Figure 5 Comparison of the decision curves of the three models in the (A) training, (B) validation, and (C) test sets. The x-axis represents the probability threshold, and the y-axis represents net income.

Figure 6 Nomogram and calibration of the combined model. (A) Nomogram of the combined model integrating AI-derived quantitative CT parameters and radiologic features for predicting the invasiveness of subcentimeter adenocarcinomatous nodules. (B) Calibration curve of the combined model showing good agreement between predicted and observed probabilities in the training cohort. AI, artificial intelligence; CT, computed tomography.

Discussion

In this study, by recording and analyzing CT lesion characteristics and measurement parameters in 337 patients with lung adenocarcinoma, we found that the density, shape and lobulation, maximum diameter, entropy, and nodal mass were independent risk factors for invasive lung adenocarcinoma. A qualitative model based on CT radiological features, a quantitative model based on AI measurement parameters, and a combined model were constructed, and the AUC values were 0.818, 0.849, and 0.901 in the internal validation set, respectively (Table 4).

The findings of this study hold significant potential for integration into multiple stages of the clinical management pathway for pulmonary nodules. It may serve as a triage tool during multidisciplinary team discussions, particularly for evaluating subcentimeter nodules of uncertain malignant potential. Furthermore, following initial nodule detection—whether incidentally or via screening—the model could provide rapid and objective risk stratification, helping to identify patients who require intensified follow-up while avoiding unnecessary procedures for those with low-risk lesions. This is especially relevant for subcentimeter ground-glass nodules (GGNs), which often lack typical morphological signs of invasiveness. By reducing unwarranted follow-up CT scans and invasive biopsies, the algorithm could significantly decrease diagnostic costs and patient anxiety.

In addition, the model may support preoperative decision-making by offering quantitative insights into tumor invasiveness, thereby assisting in refining surgical plans. Given these potential applications, it is essential to conduct early health technology assessment and formal cost-effectiveness analyses in real-world settings. Such evaluations will be crucial to objectively quantifying the clinical utility, operational efficiency, and economic impact of AI-based tools before broader implementation (25-27).

Conventional CT imaging features make the diagnosis based on morphological changes, but subcentimeter GGNs are usually in the early stages and typical imaging features tend to be less frequent (28). With the increased chance of detecting subcentimeter lung lesions in recent years, the clinicopathological features and appropriate management of tumors smaller than 1 cm have become a greater concern in thoracic surgery (29-32).

In the quantitative feature analysis, significant differences were found between invasive and noninvasive lung cancer in parameters including mean diameter, longest diameter, vertical diameter, mean CT value, maximum CT value, minimum CT value, median, standard deviation, entropy, kurtosis, skewness, compactness, maximum layer area, nodule mass, and total nodule volume. Multivariate analysis further confirmed maximum diameter, entropy, and nodule mass as independent predictors of invasive lung cancer. These findings are consistent with previous studies. For instance, in a study by Maeyashiki et al., who analyzed 398 patients with clinically staged IA lung cancer, maximum diameter of the solid component within the tumor was an independent risk factor for lung cancer prognosis (33). Solid lung cancers without a ground-glass opacity component exhibit a more malignant behavior and a worse prognosis than part-solid lung cancers (34-36). Morphologically, round or round-like shape, lobular sign, burr sign, pleural depression sign, and vascular collection sign often suggest the possibility of malignancy. Weng et al. also found that regular shape (round or oval) was an independent risk predictor for differentiating between MIA and IA (37). Compared to the predictive performance in the study by Weng et al. (AUC =0.888), which also included patients with adenocarcinoma, that in our study was superior (AUC =0.901). This improvement may stem from our larger, more representative cohort. Beyond the shape and solid component features, our CT quantitative analysis additionally identified lobulation as a predictor of lung adenocarcinoma infiltration depth, yielding superior AUC values.

Imaging signs of lung cancer play an important role in diagnosis, but the interpretation of imaging signs is subjective and may vary depending on the radiologist’s training, years of experience, and familiarity with subcentimeter GGNs. We aimed to analyze the risk factors of subcentimeter invasive lung adenocarcinoma from more objective indicators through quantitative methods. In the analysis of qualitative features, we observed that increased lesion density, irregular morphology, and lobulation were significant indicators of invasive or noninvasive lung cancer. Notably, density, shape, and lobulation were identified as independent predictors of invasive lung cancer. Lee et al. (5) found that the larger the diameter of GGN is, the greater its invasiveness. In a study by Qiu et al., it was also demonstrated that nodule diameter and GGN mass doubling time could differentially diagnosis infiltrative and non-infiltrative lung cancer (38). A multivariate analysis by Zhang et al. indicated that larger nodule diameters are significantly correlated with IAC (39).

While our study focused on AI technologies applied to CT imaging, the development of AI-enhanced magnetic resonance imaging (MRI) has also shown promising potential in pulmonary imaging. Quantitative imaging biomarkers of pulmonary nodules may be mutually applicable between CT and MRI, and MRI-based AI could provide a radiation-free alternative for clinical evaluation and follow-up. Marka et al. performed chest CT and three-dimensional gradient echo MRI using parallel imaging and compressed-sense AI (CSAI) acceleration on patients with benign and malignant lung nodules. The CSAI gradient echo sequence achieved a detection rate of 96.3% (40). The application of compressed sensing and AI technologies in pulmonary MRI can not only improve the detection rate for lung adenocarcinoma to a certain extent but also significantly reduce the scan time without compromising the detection rate or characteristics of nodules (41).

This study focused on subcentimeter adenocarcinomatous nodules because the evaluation of invasiveness in such small lesions remains one of the most challenging aspects in clinical diagnosis. Larger nodules typically exhibit typical invasive features, whereas subcentimeter nodules often have overlapping imaging characteristics between noninvasive lesions (AAH, AIS, MIA) and IACs. In addition, AAH, AIS, and MIA were classified as the noninvasive group, and IAC as the invasive group, based on the markedly better prognosis of patients with AIS or MIA reported in recent studies (42-44).

Certain limitations to this study should be acknowledged. First, the sample size of the study was small, and thus a larger cohort is needed for future research. Second, as we employed a retrospective design and examined only patients with pathologically confirmed lung adenocarcinomas that underwent surgery or biopsy, selection bias might have been introduced; therefore, prospective or external validation studies should be conducted. Third, all patients in the study already had confirmed lung adenocarcinoma, which limits applicability to mixed-nodule populations. Evaluation in a more heterogeneous cohort, including benign nodules, is necessary to assess the diagnostic specificity and potential screening applications of the models. Finally, this study did not conduct a cost–benefit analysis or workflow integration analysis, and further exploration of the potential applications of the models in clinical practice is warranted.

Conclusions

This study developed and validated predictive models combining AI-derived quantitative CT parameters with conventional radiologic features to assess the invasiveness of subcentimeter adenocarcinomatous nodules. The combined model demonstrated the highest predictive performance among all models, suggesting its potential clinical utility for preoperative risk stratification and individualized surgical decision-making. Further multicenter prospective studies should be conducted to confirm these findings and enhance model generalizability.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-650/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-650/dss

Funding: This study was funded by the National Natural Science Foundation of China (No. 82271994); the Military Commission Health Care Special Project (No. 22BJZ07); the National Health Commission Capacity Building and Continuing Education Center (No. YXFSC2022JJSJ010); the Shanghai Hospital Development Center (No. SHDC22022310-B); Navy Medical University Teaching Achievement Cultivation Project (No. JPY2022B15); and Kunshan City Key Research and Development Plan (Social Development) Project (Nos. KS2337 & KS2442).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-650/coif). All authors report funding from the National Natural Science Foundation of China (No. 82271994); the Military Commission Health Care Special Project (No. 22BJZ07); the National Health Commission Capacity Building and Continuing Education Center (No. YXFSC2022JJSJ010); the Shanghai Hospital Development Center (No. SHDC22022310-B); Navy Medical University Teaching Achievement Cultivation Project (No. JPY2022B15); and Kunshan City Key Research and Development Plan (Social Development) Project (Nos. KS2337 & KS2442). 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. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This retrospective study was approved by the Ethics Committee of Shanghai Changzheng Hospital (approval No. 2022SL070) and Kunshan Third People’s Hospital (approval No. kssy2021-45). Due to the retrospective nature of the analysis and the anonymity of the data, the requirement for informed patient consent was waived.

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. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, Bray F. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 2021;71:209-49. [Crossref] [PubMed]
2. Luo X, Zang X, Yang L, Huang J, Liang F, Rodriguez-Canales J, Wistuba II, Gazdar A, Xie Y, Xiao G. Comprehensive Computational Pathological Image Analysis Predicts Lung Cancer Prognosis. J Thorac Oncol 2017;12:501-9. [Crossref] [PubMed]
3. WHO Classification of Tumours Editorial Board. Thoracic tumours. International Agency for Research on Cancer; 2021.
4. Venkadesh KV, Setio AAA, Schreuder A, Scholten ET, Chung K. W Wille MM, Saghir Z, van Ginneken B, Prokop M, Jacobs C. Deep Learning for Malignancy Risk Estimation of Pulmonary Nodules Detected at Low-Dose Screening CT. Radiology 2021;300:438-47. [Crossref] [PubMed]
5. Lee SM, Park CM, Goo JM, Lee HJ, Wi JY, Kang CH. Invasive pulmonary adenocarcinomas versus preinvasive lesions appearing as ground-glass nodules: differentiation by using CT features. Radiology 2013;268:265-73. [Crossref] [PubMed]
6. Yoshizawa A, Motoi N, Riely GJ, Sima CS, Gerald WL, Kris MG, Park BJ, Rusch VW, Travis WD. Impact of proposed IASLC/ATS/ERS classification of lung adenocarcinoma: prognostic subgroups and implications for further revision of staging based on analysis of 514 stage I cases. Mod Pathol 2011;24:653-64. [Crossref] [PubMed]
7. Sun Y, Li C, Jin L, Gao P, Zhao W, Ma W, Tan M, Wu W, Duan S, Shan Y, Li M. Radiomics for lung adenocarcinoma manifesting as pure ground-glass nodules: invasive prediction. Eur Radiol 2020;30:3650-9. [Crossref] [PubMed]
8. Zhao W, Xu Y, Yang Z, Sun Y, Li C, Jin L, Gao P, He W, Wang P, Shi H, Hua Y, Li M. Development and validation of a radiomics nomogram for identifying invasiveness of pulmonary adenocarcinomas appearing as subcentimeter ground-glass opacity nodules. Eur J Radiol 2019;112:161-8. [Crossref] [PubMed]
9. Luo T, Xu K, Zhang Z, Zhang L, Wu S. Radiomic features from computed tomography to differentiate invasive pulmonary adenocarcinomas from non-invasive pulmonary adenocarcinomas appearing as part-solid ground-glass nodules. Chin J Cancer Res 2019;31:329-38. [Crossref] [PubMed]
10. Yankelevitz DF, Yip R, Smith JP, Liang M, Liu Y, Xu DM, Salvatore MM, Wolf AS, Flores RM, Henschke CI. CT Screening for Lung Cancer: Nonsolid Nodules in Baseline and Annual Repeat Rounds. Radiology 2015;277:555-64. [Crossref] [PubMed]
11. Church TR, Black WC, Aberle DR, Berg CD, Clingan KL, Duan F, Fagerstrom RM, Gareen IF, Gierada DS, Jones GC, Mahon I, Marcus PM, Sicks JD, Jain A, Baum S. Results of initial low-dose computed tomographic screening for lung cancer. N Engl J Med 2013;368:1980-91. [Crossref] [PubMed]
12. Gould MK, Tang T, Liu IL, Lee J, Zheng C, Danforth KN, Kosco AE, Di Fiore JL, Suh DE. Recent Trends in the Identification of Incidental Pulmonary Nodules. Am J Respir Crit Care Med 2015;192:1208-14. [Crossref] [PubMed]
13. Madsen PH, Holdgaard PC, Christensen JB, Høilund-Carlsen PF. Clinical utility of F-18 FDG PET-CT in the initial evaluation of lung cancer. Eur J Nucl Med Mol Imaging 2016;43:2084-97. [Crossref] [PubMed]
14. Libby DM, Smith JP, Altorki NK, Pasmantier MW, Yankelevitz D, Henschke CI. Managing the small pulmonary nodule discovered by CT. Chest 2004;125:1522-9. [Crossref] [PubMed]
15. Thalanayar PM, Altintas N, Weissfeld JL, Fuhrman CR, Wilson DO. Indolent, Potentially Inconsequential Lung Cancers in the Pittsburgh Lung Screening Study. Ann Am Thorac Soc 2015;12:1193-6. [Crossref] [PubMed]
16. Gao W, Wen CP, Wu A, Welch HG. Association of Computed Tomographic Screening Promotion With Lung Cancer Overdiagnosis Among Asian Women. JAMA Intern Med 2022;182:283-90. [Crossref] [PubMed]
17. Gould MK, Creekmur B, Qi L, Golden SE, Kaplan CP, Walter E, Mularski RA, Vaszar LT, Fennig K, Steiner J, de Bie E, Musigdilok VV, Altman DA, Dyer DS, Kelly K, Miglioretti DL, Wiener RS, Slatore CG, Smith-Bindman R. Emotional Distress, Anxiety, and General Health Status in Patients With Newly Identified Small Pulmonary Nodules: Results From the Watch the Spot Trial. Chest 2023;164:1560-71. [Crossref] [PubMed]
18. Jonas DE, Reuland DS, Reddy SM, Nagle M, Clark SD, Weber RP, Enyioha C, Malo TL, Brenner AT, Armstrong C, Coker-Schwimmer M, Middleton JC, Voisin C, Harris RP. Screening for Lung Cancer With Low-Dose Computed Tomography: Updated Evidence Report and Systematic Review for the US Preventive Services Task Force. JAMA 2021;325:971-87. [Crossref] [PubMed]
19. Kim RY, Oke JL, Pickup LC, Munden RF, Dotson TL, Bellinger CR, Cohen A, Simoff MJ, Massion PP, Filippini C, Gleeson FV, Vachani A. Artificial Intelligence Tool for Assessment of Indeterminate Pulmonary Nodules Detected with CT. Radiology 2022;304:683-91. [Crossref] [PubMed]
20. Zhao ZR, Yu YH, Lin ZC, Ma DH, Lin YB, Hu J, Luo QQ, Li GF, Chen C, Yang YL, Yang JC, Lin YB, Long H. Invasiveness assessment by artificial intelligence against intraoperative frozen section for pulmonary nodules ≤ 3 cm. J Cancer Res Clin Oncol 2023;149:7759-65. [Crossref] [PubMed]
21. Shu J, Wen D, Xu Z, Meng X, Zhang Z, Lin S, Zheng M. Improved interobserver agreement on nodule type and Lung-RADS classification of subsolid nodules using computer-aided solid component measurement. Eur J Radiol 2022;152:110339. [Crossref] [PubMed]
22. Adams SJ, Mondal P, Penz E, Tyan CC, Lim H, Babyn P. Development and Cost Analysis of a Lung Nodule Management Strategy Combining Artificial Intelligence and Lung-RADS for Baseline Lung Cancer Screening. J Am Coll Radiol 2021;18:741-51. [Crossref] [PubMed]
23. Yu KH, Zhang C, Berry GJ, Altman RB, Ré C, Rubin DL, Snyder M. Predicting non-small cell lung cancer prognosis by fully automated microscopic pathology image features. Nat Commun 2016;7:12474. [Crossref] [PubMed]
24. Sterne JA, White IR, Carlin JB, Spratt M, Royston P, Kenward MG, Wood AM, Carpenter JR. Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls. BMJ 2009;338:b2393. [Crossref] [PubMed]
25. Marka AW, Luitjens J, Gassert FT, Steinhelfer L, Burian E, Rübenthaler J, Schwarze V, Froelich MF, Makowski MR, Gassert FG. Artificial intelligence support in MR imaging of incidental renal masses: an early health technology assessment. Eur Radiol 2024;34:5856-65. [Crossref] [PubMed]
26. Kemper EHM, Erenstein H, Boverhof BJ, Redekop K, Andreychenko AE, Dietzel M, Groot Lipman KBW, Huisman M, Klontzas ME, Vos F, IJzerman M, Starmans MPA, Visser JJ. ESR Essentials: how to get to valuable radiology AI: the role of early health technology assessment-practice recommendations by the European Society of Medical Imaging Informatics. Eur Radiol 2025;35:3432-41. [Crossref] [PubMed]
27. Vermeulen RJ, Govers TM, van Leeuwen KG. Early health technology assessment: the value of valuing AI applications. Eur Radiol 2024;34:5854-5. [Crossref] [PubMed]
28. Yang X, Chu XP, Huang S, Xiao Y, Li D, Su X, Qi YF, Qiu ZB, Wang Y, Tang WF, Wu YL, Zhu Q, Liang H, Zhong WZ. A novel image deep learning-based sub-centimeter pulmonary nodule management algorithm to expedite resection of the malignant and avoid over-diagnosis of the benign. Eur Radiol 2024;34:2048-61. [Crossref] [PubMed]
29. Lee PC, Korst RJ, Port JL, Kerem Y, Kansler AL, Altorki NK. Long-term survival and recurrence in patients with resected non-small cell lung cancer 1 cm or less in size. J Thorac Cardiovasc Surg 2006;132:1382-9. [Crossref] [PubMed]
30. Miller DL, Rowland CM, Deschamps C, Allen MS, Trastek VF, Pairolero PC. Surgical treatment of non-small cell lung cancer 1 cm or less in diameter. Ann Thorac Surg 2002;73:1545-50; discussion 1550-1. [Crossref] [PubMed]
31. Zhou Q, Suzuki K, Anami Y, Oh S, Takamochi K. Clinicopathologic features in resected subcentimeter lung cancer--status of lymph node metastases. Interact Cardiovasc Thorac Surg 2010;10:53-7. [Crossref] [PubMed]
32. Asamura H, Suzuki K, Watanabe S, Matsuno Y, Maeshima A, Tsuchiya R. A clinicopathological study of resected subcentimeter lung cancers: a favorable prognosis for ground glass opacity lesions. Ann Thorac Surg 2003;76:1016-22. [Crossref] [PubMed]
33. Maeyashiki T, Suzuki K, Hattori A, Matsunaga T, Takamochi K, Oh S. The size of consolidation on thin-section computed tomography is a better predictor of survival than the maximum tumour dimension in resectable lung cancer. Eur J Cardiothorac Surg 2013;43:915-8. [Crossref] [PubMed]
34. Hattori A, Suzuki K, Matsunaga T, Takamochi K, Oh S. Visceral pleural invasion is not a significant prognostic factor in patients with a part-solid lung cancer. Ann Thorac Surg 2014;98:433-8.
35. Hattori A, Suzuki K, Matsunaga T, Miyasaka Y, Takamochi K, Oh S. What is the appropriate operative strategy for radiologically solid tumours in subcentimetre lung cancer patients?†. Eur J Cardiothorac Surg 2015;47:244-9. [Crossref] [PubMed]
36. Hattori A, Maeyashiki T, Matsunaga T, Takamochi K, Oh S, Suzuki K. Predictors of pathological non-invasive lung cancer with pure-solid appearance on computed tomography to identify possible candidates for sublobar resection. Surg Today 2016;46:102-9. [Crossref] [PubMed]
37. Weng Q, Zhou L, Wang H, Hui J, Chen M, Pang P, Zheng L, Xu M, Wang Z, Ji J. A radiomics model for determining the invasiveness of solitary pulmonary nodules that manifest as part-solid nodules. Clin Radiol 2019;74:933-43. [Crossref] [PubMed]
38. Qiu L, Zhang X, Mao H, Fang X, Ding W, Zhao L, Chen H. Comparison of Comprehensive Morphological and Radiomics Features of Subsolid Pulmonary Nodules to Distinguish Minimally Invasive Adenocarcinomas and Invasive Adenocarcinomas in CT Scan. Front Oncol 2021;11:691112. [Crossref] [PubMed]
39. Zhang Y, Shen Y, Qiang JW, Ye JD, Zhang J, Zhao RY. HRCT features distinguishing pre-invasive from invasive pulmonary adenocarcinomas appearing as ground-glass nodules. Eur Radiol 2016;26:2921-8. [Crossref] [PubMed]
40. Marka AW, Steinhardt M, Rahn L, Lemke T, Gassert FT, Huber T, Sauter A, Weiss K, Makowski MR, Van AT, Karampinos DC, Graf M, Pfeiffer D, Gawlitza J, Ziegelmayer S. AI-Enhanced 3D Gradient Echo MRI: A Radiation-Free Alternative to CT for Lung Nodule Detection and Lung-RADS Classification. Acad Radiol 2025;32:6250-9. [Crossref] [PubMed]
41. Ziegelmayer S, Marka AW, Strenzke M, Lemke T, Rosenkranz H, Scherer B, Huber T, Weiss K, Makowski MR, Karampinos DC, Graf M, Gawlitza J. Speed and efficiency: evaluating pulmonary nodule detection with AI-enhanced 3D gradient echo imaging. Eur Radiol 2025;35:2237-44. [Crossref] [PubMed]
42. Yotsukura M, Asamura H, Motoi N, Kashima J, Yoshida Y, Nakagawa K, Shiraishi K, Kohno T, Yatabe Y, Watanabe SI. Long-Term Prognosis of Patients With Resected Adenocarcinoma In Situ and Minimally Invasive Adenocarcinoma of the Lung. J Thorac Oncol 2021;16:1312-20. [Crossref] [PubMed]
43. Sharma J, Zhou F, Moreira AL. Pulmonary Adenocarcinoma Updates: Histology, Cytology, and Grading. Arch Pathol Lab Med 2025;149:e82-6. [Crossref] [PubMed]
44. Zhu J, Wang W, Xiong Y, Xu S, Chen J, Wen M, Zhao Y, Lei J, Jiang T. Evolution of lung adenocarcinoma from preneoplasia to invasive adenocarcinoma. Cancer Med 2023;12:5545-57. [Crossref] [PubMed]