Incremental prognostic value of solid component volume ratio and entropy for pathological stage IA invasive lung adenocarcinoma after surgery

Introduction

Non-small cell lung cancer (NSCLC) is one of the leading causes of cancer-related death worldwide, and lung adenocarcinoma (LUAD) is a major histopathological subtype. With the popularization of lung cancer screening, more and more NSCLC can be detected in the early stage. The patients with early-stage NSCLC, particularly stage IA, have a favorable prognosis after complete surgical resection (1,2). Nevertheless, some stage IA patients still experience recurrence, resulting in significant variability in clinical prognosis (3-7).

Some studies have investigated the risk factors for recurrence in early-stage NSCLC after surgery. Fick et al. (8) identified the high-risk features associated with recurrence in stage I LUAD, including tumor size, lymphovascular invasion, visceral pleural invasion, sublobar resection. Huang et al. (9) found that a higher proportion of solid and micropapillary histologic subtypes could predict recurrence in stage IA invasive LUAD. Fourdrain et al. (10) observed that peripheral tumor location was associated with a higher rate of local recurrence.

In addition, the Japan Clinical Oncology Group 0201 (JCOG0201) study demonstrated a strong association between the consolidation tumor ratio (CTR) on computed tomography (CT) images and both tumor invasiveness and postoperative prognosis (3). The JCOG series also established a surgical selection strategy based on tumor size and CTR (3-6). However, due to the irregular shape and indistinct boundary between the solid and non-solid components, CTR is difficult and challenging to measure accurately, and is prone to significant inter-observer variability (11-14). Therefore, it is essential to find a more reliable and stable measurement method. Several studies have highlighted the advantages of artificial intelligence (AI) in quantifying solid component, providing a promising alternative to manual assessment (15-18). Furthermore, radiomics parameters such as skewness, kurtosis, and entropy have been used to predict the invasiveness of LUAD (19,20), but their predictive value for recurrence remains unclear.

This study retrospectively analyzed the clinical, pathological, and radiological quantitative parameters of patients with completely resected pathological stage IA invasive LUAD. We used competing risk and Cox proportional hazards models to identify independent risk factors associated with recurrence and death, and developed a predictive model for recurrence risk. We hope to identify patients with high-risk of recurrence to guide follow-up strategies, adjuvant therapy decisions, and personalized treatment plans, ultimately improving long-term outcomes. We present this article in accordance with the TRIPOD reporting checklist (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-838/rc).

Methods

Study design and patients

This single-center, retrospective, observational study included patients with invasive LUAD undergoing surgery at West China Hospital, Sichuan University between August 2017 and March 2023. Inclusion criteria included patients aged 18 to 80 years who underwent lobectomy, segmentectomy, or wedge resection with lymph node dissection. Postoperative pathological staging confirmed stage IA invasive LUAD according to the eighth edition of the tumor node metastasis classification for NSCLC (1). All patients underwent a three-dimensional (3D), thin-section, contrast-enhanced chest CT scan within two weeks before surgery. Exclusion criteria included presence of residual tumor, receipt of any induction or adjuvant therapy, history of malignant tumor within 5 years, synchronous lung cancers, and tumor images cannot be recognized by deep learning (DL) software (Figure 1). The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of West China Hospital, Sichuan University (No. 2024-376) and individual consent for this retrospective analysis was waived.

Figure 1 Flowchart of patient screening.

Chest CT scan

All chest CT images were acquired using a 256-slice multidetector CT scanner (Revolution CT, GE Healthcare, Milwaukee, USA). Before scanning, all patients underwent breath-holding training to minimize respiratory motion artifacts. The tube voltage was automatically determined by the scanner based on scout images, with options of 100 and 120 kVp. The tube current was also automatically selected, ranging from 200 to 650 mA. The preset noise index was 25 Hounsfield unit (HU). The slice thickness and interval were both 1.00 mm, and the matrix size was 512×512. The gantry rotation time was 0.28 s per rotation. After scanning, standard reconstruction was performed using a hybrid iterative reconstruction algorithm (ASIR-V, GE Healthcare) to reconstruct images at the optimal phase.

Quantitative parameters measured by radiologists

The maximum total size on axial images (MTSA) and the maximum solid component size on axial images (MSSA) were evaluated by two experienced chest radiologists (X.C. and Y.Z., with 8 and 6 years of experience, respectively) using a window level of −400 HU and a window width of 1,800 HU. CTR was defined as the ratio of the maximum size of solid component to the maximum size of tumor, calculated as MSSA divided by MTSA. The distance from tumor to visceral pleura was measured as the shortest distance between the tumor edge and the visceral pleura on CT images.

Quantitative parameters measured by AI

DL-based quantitative parameters were automatically obtained using a commercially available chest CT analysis platform (Deepwise, Beijing, China). Density thresholds of 0, −50, −100, −150, −200, −250, −300, −350, and −400 HU were applied to separate solid component, respectively. The maximum tumor size on multiplanar reconstructed images (MTSMPR), total volume (TV), solid component volume (SV), and solid component volume ratio (SV%) at each threshold were evaluated (Figure 2A). Additional radiomics parameters, including variance, skewness, kurtosis, entropy, and sphericity were also extracted from DL software (Figure 2B-2E). Detailed definitions and calculation formulas are provided in Appendix 1.

Figure 2 An example of using deep learning software to assess lung nodule. (A) Three-dimensional CT images of a lung nodule are loaded into the software, and segmentation of solid component at a specific density threshold is employed. Red represents solid component, and gray represents non-solid component. (B-E) Four radiomics parameters (variance, skewness, kurtosis, and entropy). CT, computed tomography.

Outcomes and follow-up

The primary endpoint was the cumulative recurrence rate, including both local and distant recurrences. Local recurrence was defined as recurrence occurring in the ipsilateral thorax, including resection margins of the lung or bronchus, ipsilateral pleural, malignant pleural effusion, and hilar and mediastinal lymph nodes, as well as the emergence of new nodules on the same side. Distant recurrence was defined as recurrence occurring in any locations other than local recurrence, including the contralateral lung, hilar and mediastinal lymph nodes, cervical lymph nodes, and any other tissues or organs. Patients who experienced recurrence were classified as local recurrence, distant recurrence, or both, based on the initial site of recurrence. Simultaneous local and distant recurrences were classified as distant. All recurrences were determined by two experienced thoracic surgeons based on imaging or pathological evidence.

The secondary endpoints were recurrence-free survival (RFS) and overall survival (OS). RFS was defined as the period from surgery to recurrence, death from any cause, or the last follow-up. OS was defined as the period from surgery to death from any cause or the last follow-up. All outcomes were collected from follow-up or hospital electronic medical records. Follow-up was performed every 6 months after surgery.

Statistical analysis

All patients were randomly divided into a training group and a validation group in a ratio of 7:3. Continuous variables were expressed as mean ± standard deviation or median [interquartile range], and compared using Student’s t-test or Mann-Whitney U-test. Categorical variables were expressed as number (percentage), and compared using Pearson’s chi-square test or Fisher’s exact test.

For the primary endpoint, competing risk model was used to estimate the association between variables and the risk of recurrence, death without recurrence as a competing event. For the secondary endpoints, Cox proportional hazards model was employed to evaluate variables associated with RFS and OS. The variables with a P value less than 0.05 in univariate analysis were included in multivariate analysis. The optimal cutoff values for radiological quantitative parameters were determined using “cutoff” package in R. Three nested predictive models were developed based on variables that remained significant in multivariate analysis to predict recurrence. The predictive accuracy was evaluated using Harrell’s concordance index (C-index) and receiver operating characteristic (ROC) curve analysis. The different models were compared by net reclassification improvement index (NRI) and integrated discrimination improvement index (IDI). Calibration curve was used to evaluate the consistency between predicted and actual probabilities. Decision curve analysis (DCA) was performed to assess the clinical utility. Finally, patients were stratified into high-risk and low-risk groups based on total risk scores. Cumulative incidence curves with Fine-Gray test were applied to compare differences between these groups. A two-sided P value of less than 0.05 was considered statistically significant. All statistical analyses were conducted using R version 4.3.3 (R Foundation, Vienna, Austria).

Results

Patient characteristics

According to the inclusion and exclusion criteria, a total of 2,117 patients were finally included and randomly assigned to a training group of 1,481 patients and a validation group of 636 patients in a ratio of 7:3. The baseline characteristics are summarized in Table 1, with no significant differences observed between the two groups.

Follow-up and outcomes

The last follow-up was conducted on February 10, 2025. The median follow-up time was 49.57 [95% confidence interval (CI): 48.57–50.60] months. Recurrence occurred in 139 patients (6.57%), including 42 (30.22%) with local recurrence and 97 (69.78%) with distant recurrence (Table S1). A total of 41 patients (1.94%) died during the follow-up period.

Quantitative parameters

The radiological quantitative parameters measured by radiologists and AI, and their association with postoperative recurrence are summarized in Table 2. Among these parameters, SV% with 0 HU as the solid component threshold (SV_0HU%) showed the strongest correlation with recurrence and demonstrated the highest predictive performance, with a C-index of 0.747 (95% CI: 0.684–0.810), outperforming both manually measured and other thresholds.

Risk factors associated with prognosis

For postoperative recurrence, univariate analysis showed that age (P=0.03), surgical procedure of wedge resection (P=0.001), pathological stage of IA3 (P=0.001), histologic subtype of acinar or papillary predominant (P<0.001), histologic subtype of micropapillary or solid present (P<0.001), vascular or perineural invasion (P<0.001), spread through air space (STAS) (P<0.001), SV_{0 HU}% (P<0.001), skewness (P<0.001), and entropy (P<0.001) were associated with recurrence. In multivariate analysis, pathological stage of IA3 (P=0.001), histologic subtype of acinar or papillary predominant (P=0.02), histologic subtype of micropapillary or solid present (P<0.001), vascular or perineural invasion (P=0.009), STAS (P=0.01), SV_{0 HU}% (P<0.001), and entropy (P=0.003) were identified as independent risk factors for recurrence (Table 3).

For RFS, univariate analysis showed that age (P<0.001), surgical procedure of wedge resection (P=0.008), pathological stage of IA2 (P=0.046), pathological stage of IA3 (P=0.001), histologic subtype of acinar or papillary predominant (P<0.001), histologic subtype of micropapillary or solid present (P<0.001), vascular or perineural invasion (P<0.001), STAS (P<0.001), SV_{0 HU}% (P<0.001), skewness (P<0.001), and entropy (P<0.001) were associated with RFS. In multivariate analysis, age (P=0.005), pathological stage of IA2 (P=0.02), pathological stage of IA3 (P<0.001), vascular or perineural invasion (P=0.009), STAS (P<0.001), SV_{0 HU}% (P<0.001), skewness (P=0.02), and entropy (P=0.009) were identified as independent risk factors for RFS (Table S2).

For OS, univariate analysis showed that age (P<0.001), sex (P=0.03), histologic subtype of micropapillary or solid present (P<0.001), vascular or perineural invasion (P=0.003), STAS (P<0.001), SV_{0 HU}% (P<0.001), skewness (P=0.01), and entropy (P<0.001) were associated with OS. In multivariate analysis, age (P<0.001), STAS (P=0.01), SV_{0 HU}% (P<0.001), and entropy (P=0.003) were identified as independent risk factors for OS (Table S3).

Development and validation of predictive models for recurrence

According to the results of multivariate analysis, three predictive models for recurrence were developed:

• Model 1: clinical and pathological characteristics (pathological stage, histologic subtype, vascular or perineural invasion, and STAS);
• Model 2: Model 1 + SV_{0 HU}%;
• Model 3: Model 2 + entropy.

The C-index for Models 1–3 in the training group were 0.841 (95% CI: 0.796–0.886), 0.862 (95% CI: 0.815–0.909), and 0.865 (95% CI: 0.822–0.908), respectively, and in the validation group were 0.835 (95% CI: 0.770–0.900), 0.861 (95% CI: 0.790–0.932), and 0.862 (95% CI: 0.803–0.921), respectively, indicating strong discriminative performance. The ROC curves showed that the 1-, 3-, and 5-year area under the ROC curve (AUC) of Model 3 in the training group were 0.854, 0.878, and 0.880, respectively, and in the validation group were 0.819, 0.851, and 0.862, respectively, which were higher than the other models (Figure 3). The NRI and IDI for Models 1–3 are shown in Tables S4,S5, respectively, demonstrating the improvement in predictive performance. The calibration curves indicated a high consistency between actual recurrence-free probabilities and those predicted by the models in the training and validation groups, especially for Model 3 (Figure 4). The DCA showed that Model 3 provided a greater clinical net benefits with a wide range of threshold probabilities compared with other models (Figure 5).

Figure 3 The ROC curves for Models 1–3. The 1-, 3-, and 5-year AUC of Model 3 in the training group (A-C) are 0.854, 0.878, and 0.880, respectively, and in the validation group (D-F) are 0.819, 0.851, and 0.862, respectively, which are higher than Model 1 and Model 2. Model 1: clinical and pathological characteristics (pathological stage, histologic subtype, vascular or perineural invasion, and STAS). Model 2: Model 1 + SV_{0 HU}%. Model 3: Model 2 + entropy. AUC, area under the ROC curve; HU, Hounsfield unit; ROC, receiver operating characteristic; STAS, spread through air space; SV_{0 HU}%, solid component volume ratio with 0 HU as the solid component threshold.

Figure 4 The calibration curves for Models 1–3. The 1-, 3-, and 5-year actual and predicted recurrence-free probabilities in the training group (A-C) and validation group (D-F). The calibration curves indicate a high consistency between actual recurrence-free probabilities and those predicted by the models in the training and validation groups, especially for Model 3. Model 1: clinical and pathological characteristics (pathological stage, histologic subtype, vascular or perineural invasion, and STAS). Model 2: Model 1 + SV_{0 HU}%. Model 3: Model 2 + entropy. HU, Hounsfield unit; STAS, spread through air space; SV_{0 HU}%, solid component volume ratio with 0 HU as the solid component threshold.

Figure 5 The decision curve analysis for Models 1–3. The 1-, 3-, and 5-year prediction for recurrence in the training group (A-C) and validation group (D-F). Model 3 can provide greater clinical net benefits with a wide range of threshold probabilities compared with Model 1 and Model 2. Model 1: clinical and pathological characteristics (pathological stage, histologic subtype, vascular or perineural invasion, and STAS). Model 2: Model 1 + SV_{0 HU}%. Model 3: Model 2 + entropy. HU, Hounsfield unit; STAS, spread through air space; SV_{0 HU}%, solid component volume ratio with 0 HU as the solid component threshold.

Risk stratification based on the predictive models

Risk scores derived from the predictive models were used to classify patients into low-risk and high-risk groups in both the training and validation groups. Cumulative recurrence curves revealed significant differences between the two groups, particularly in Model 3, with hazard ratios of 27.32 (95% CI: 17.25–43.27; P<0.001) and 17.53 (95% CI: 8.47–36.34; P<0.001), respectively (Figure 6).

Figure 6 Cumulative recurrence curves according to the risk scores predicted by Models 1–3. The high-risk groups exhibit higher recurrence rate compared with the low-risk groups in the training group (A-C) and validation group (D-F). Model 1: clinical and pathological characteristics (pathological stage, histologic subtype, vascular or perineural invasion, and STAS). Model 2: Model 1 + SV_{0 HU}%. Model 3: Model 2 + entropy. HU, Hounsfield unit; STAS, spread through air space; SV_{0 HU}%, solid component volume ratio with 0 HU as the solid component threshold.

Discussion

In this study, we compared the associations of various radiological quantitative parameters with the prognosis of pathological stage IA invasive LUAD, integrating clinical and pathological characteristics, identified risk factors for recurrence, and developed predictive models. The results indicated that, (I) DL-based quantitative parameters were superior to manual measurements in predicting the prognosis of pathological stage IA invasive LUAD; (II) SV_{0 HU}% showed the strongest correlation with recurrence and demonstrated the highest predictive performance; (III) pathological stage, histologic subtype, vascular or perineural invasion, STAS, SV_{0 HU}%, and entropy were independent risk factors for postoperative recurrence; (IV) DL-based SV_{0 HU}% and entropy provided incremental prognostic value.

The JCOG0201 study established CTR as a radiological marker predictive of pathological invasiveness in clinical stage I lung cancer, demonstrating a strong correlation with prognosis (3). However, due to the irregular shape and ambiguous boundary between solid and non-solid components, CTR is difficult to measure accurately and shows high inter-observer variability (11,12). In addition, traditional CTR does not fully utilize the 3D information of the tumor (13). Previous studies have shown that 3D volumetric assessment of solid component can provide more accurate prognostic information (14,15,17). DL software can automatically and accurately segment tumor into solid and non-solid components using a specific density threshold. A recent study demonstrated that DL algorithms outperform manual measurements in prognostic prediction for LUAD (18). Consistent with previous studies, our results found that DL-based quantitative parameters, especially SV%, were more effective than CTR in predicting prognosis in patients with pathological stage IA invasive LUAD. Among various thresholds, using 0 HU to delineate the solid component yielded optimal predictive performance. However, the optimal threshold may vary between nodules due to differences in the density of solid components in different nodules. Dynamically tailoring thresholds for individual nodules, potentially with the aid of AI, may further enhance segmentation accuracy.

Our study also incorporated quantifiable radiomics parameters. Previous studies have used such features to assess the invasiveness of LUAD. For example, Yoshiyasu et al. used total tumor volume, solid volume ratio, skewness, and entropy to develop a predictive model for pathological stage IA LUAD, achieving an AUC of 0.900 (19). They also suggested that this approach could help identify patients suitable for sublobar resection. Similarly, Qiu et al. integrated smoking status, mean CT attenuation, and entropy into a model predicting clinical stage IA LUAD invasiveness, reporting AUC of 0.898 and 0.849 in training and validation groups, respectively (20). In our study, entropy was significantly associated with recurrence, RFS, and OS, indicating its great value in predicting prognosis. In the field of medical imaging, entropy is widely applied as a texture feature that reflects the heterogeneity of voxel intensity distributions within a region of interest. For example, left ventricular entropy has been used to predict adverse events in patients with hypertrophic cardiomyopathy, dilated cardiomyopathy, and myocarditis. Entropy quantifies the heterogeneity of the tumor on CT images by capturing the complexity and irregularity of CT attenuation patterns, which are difficult to identify and analyze with naked eye. Higher entropy suggests a greater degree of spatial variation in voxel intensities, which may reflect biological aggressiveness, such as irregular tumor growth, necrosis, or microvascular heterogeneity. Compared with conventional size- or density-based parameters, entropy captures more subtle textural variations and therefore adds incremental predictive value in assessing recurrence risk and prognosis. As a quantifiable metric, entropy also mitigates the “black box” limitation of traditional radiomics, allowing objective identification of high-risk patients who may benefit from more aggressive treatment, such as lobectomy or adjuvant therapy. In addition, these radiomics parameters can be analyzed preoperatively. If preoperative imaging analysis can reliably predict prognosis, this will be highly valuable for surgical decisions such as the extent of resection and lymph node dissection strategy.

Our study identified high-risk factors for postoperative recurrence in pathological stage IA invasive LUAD, including pathological stage, histologic subtype, vascular or perineural invasion, STAS, SV_{0 HU}%, and entropy. In contrast, the distance of the tumor from visceral pleura and the number of dissected lymph nodes were not significant risk factors. Interestingly, it has been reported that an association between the distance from visceral pleura and local recurrence in clinical stage IA patients (10). This may be due to the presence of visceral pleural invasion in patients with clinical stage IA but not in those with pathological stage IA. Moreover, consistent with the findings of Fourdrain et al. (10), the number of dissected lymph nodes did not influence recurrence in this population, supporting a less aggressive lymph node dissection strategy for patients at low risk of lymph node metastases. Accurate preoperative assessment of lymph node status can help reduce surgical risks, operative time, and promote faster recovery. However, the minimum standard for the dissected lymph nodes requires more large prospective studies to explore.

We developed and validated three nested predictive models. Compared with Model 1, Model 2 and Model 3 demonstrated improved discrimination and calibration, with significant increases in C-index, AUC, NRI, and IDI in both the training and validation groups. The actual recurrence-free probabilities were more consistent with the predicted probabilities of Model 3. Moreover, Model 3 provided a greater clinical net benefit with a wider range of threshold probabilities and was superior to other models in predicting recurrence. The inclusion of DL-based SV% and entropy provided incremental prognostic value for pathological stage IA invasive LUAD after surgery. Combining preoperative imaging with postoperative pathology could further improve prognostic prediction.

There are some limitations in this study. First, due to the retrospective nature of this study, there is a potential selection bias. Second, since this study included patients with pathological stage IA LUAD, there are fewer patients with recurrence and death during follow-up, which limited the statistical power. Third, DL software may struggle to accurately segment certain specific tumors, such as those with surrounding inflammation or atelectasis. Fourth, the interference of blood vessels and bronchi in the tumors on the measurement cannot be completely avoided. Fifth, this study only included patients with LUAD, and it is also important to evaluate benign lesions and other pathological types of tumors. Lastly, the predictive model has not undergone external validation and requires further confirmation in prospective, multicenter studies.

Conclusions

In conclusion, this study demonstrated that DL-based quantitative parameters are superior to manual measurements in predicting the prognosis of pathological stage IA invasive LUAD. Pathological stage, histologic subtype, vascular or perineural invasion, STAS, SV_{0 HU}%, and entropy are high-risk factors for postoperative recurrence in pathological stage IA invasive LUAD and can accurately predict postoperative recurrence to facilitate personalized treatment strategies.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-838/rc

Data Sharing Statement: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-838/dss

Peer Review File: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-838/prf

Funding: This work was supported by the National Science and Technology Major Project (No. 2023ZD0501802 to J.M.).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-838/coif). Y.Y. is a current employee of Hangzhou Deepwise & League of PHD Technology Co., Ltd. The other authors have no 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 study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of West China Hospital, Sichuan University (No. 2024-376) and individual consent for this retrospective analysis 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. Goldstraw P, Chansky K, Crowley J, et al. The IASLC Lung Cancer Staging Project: Proposals for Revision of the TNM Stage Groupings in the Forthcoming (Eighth) Edition of the TNM Classification for Lung Cancer. J Thorac Oncol 2016;11:39-51. [Crossref] [PubMed]
2. Rami-Porta R, Nishimura KK, Giroux DJ, et al. The International Association for the Study of Lung Cancer Lung Cancer Staging Project: Proposals for Revision of the TNM Stage Groups in the Forthcoming (Ninth) Edition of the TNM Classification for Lung Cancer. J Thorac Oncol 2024;19:1007-27. [Crossref] [PubMed]
3. Ito H, Suzuki K, Mizutani T, et al. Long-term survival outcome after lobectomy in patients with clinical T1 N0 lung cancer. J Thorac Cardiovasc Surg 2021;161:281-90. [Crossref] [PubMed]
4. Saji H, Okada M, Tsuboi M, et al. Segmentectomy versus lobectomy in small-sized peripheral non-small-cell lung cancer (JCOG0802/WJOG4607L): a multicentre, open-label, phase 3, randomised, controlled, non-inferiority trial. Lancet 2022;399:1607-17. [Crossref] [PubMed]
5. Suzuki K, Watanabe SI, Wakabayashi M, et al. A single-arm study of sublobar resection for ground-glass opacity dominant peripheral lung cancer. J Thorac Cardiovasc Surg 2022;163:289-301.e2. [Crossref] [PubMed]
6. Aokage K, Suzuki K, Saji H, et al. Segmentectomy for ground-glass-dominant lung cancer with a tumour diameter of 3 cm or less including ground-glass opacity (JCOG1211): a multicentre, single-arm, confirmatory, phase 3 trial. Lancet Respir Med 2023;11:540-9. [Crossref] [PubMed]
7. Altorki N, Wang X, Kozono D, et al. Lobar or Sublobar Resection for Peripheral Stage IA Non-Small-Cell Lung Cancer. N Engl J Med 2023;388:489-98. [Crossref] [PubMed]
8. Fick CN, Dunne EG, Vanstraelen S, et al. High-risk features associated with recurrence in stage I lung adenocarcinoma. J Thorac Cardiovasc Surg 2025;169:436-444.e6. [Crossref] [PubMed]
9. Huang W, Zhang H, Zhang Z, et al. A prognostic nomogram based on a new classification of combined micropapillary and solid components for stage IA invasive lung adenocarcinoma. J Surg Oncol 2022;125:796-808. [Crossref] [PubMed]
10. Fourdrain A, Anastay V, Pauly V, et al. Peripheral location of lung cancer is associated with higher local disease recurrence. Eur J Cardiothorac Surg 2023;63:ezad130. [Crossref] [PubMed]
11. Matsuguma H, Nakahara R, Anraku M, et al. Objective definition and measurement method of ground-glass opacity for planning limited resection in patients with clinical stage IA adenocarcinoma of the lung. Eur J Cardiothorac Surg 2004;25:1102-6. [Crossref] [PubMed]
12. Ridge CA, Yildirim A, Boiselle PM, et al. Differentiating between Subsolid and Solid Pulmonary Nodules at CT: Inter- and Intraobserver Agreement between Experienced Thoracic Radiologists. Radiology 2016;278:888-96. [Crossref] [PubMed]
13. Heuvelmans MA, Walter JE, Vliegenthart R, et al. Disagreement of diameter and volume measurements for pulmonary nodule size estimation in CT lung cancer screening. Thorax 2018;73:779-81. [Crossref] [PubMed]
14. Kawaguchi Y, Nakao M, Omura K, et al. The utility of three-dimensional computed tomography for prediction of tumor invasiveness in clinical stage IA lung adenocarcinoma. J Thorac Dis 2020;12:7218-26. [Crossref] [PubMed]
15. Kamiya S, Iwano S, Umakoshi H, et al. Computer-aided Volumetry of Part-Solid Lung Cancers by Using CT: Solid Component Size Predicts Prognosis. Radiology 2018;287:1030-40. [Crossref] [PubMed]
16. Ahn Y, Lee SM, Noh HN, et al. Use of a Commercially Available Deep Learning Algorithm to Measure the Solid Portions of Lung Cancer Manifesting as Subsolid Lesions at CT: Comparisons with Radiologists and Invasive Component Size at Pathologic Examination. Radiology 2021;299:202-10. [Crossref] [PubMed]
17. Kawaguchi Y, Shimada Y, Murakami K, et al. Prognostic impact of artificial intelligence-based volumetric quantification of the solid part of the tumor in clinical stage 0-I adenocarcinoma. Lung Cancer 2022;170:85-90. [Crossref] [PubMed]
18. Zhu Y, Chen LL, Luo YW, et al. Prognostic impact of deep learning-based quantification in clinical stage 0-I lung adenocarcinoma. Eur Radiol 2023;33:8542-53. [Crossref] [PubMed]
19. Yoshiyasu N, Kojima F, Hayashi K, et al. Radiomics technology for identifying early-stage lung adenocarcinomas suitable for sublobar resection. J Thorac Cardiovasc Surg 2021;162:477-485.e1. [Crossref] [PubMed]
20. Qiu ZB, Zhang C, Chu XP, et al. Quantifying invasiveness of clinical stage IA lung adenocarcinoma with computed tomography texture features. J Thorac Cardiovasc Surg 2022;163:805-815.e3. [Crossref] [PubMed]