Tumor-associated prognostic factors extractable from chest CT scans in patients with lung cancer

Introduction

Chest computed tomography (CT) scans are an essential tool for preoperative evaluations, clinical staging, and response monitoring in patients with lung cancer. The prognostic importance of CT features, including the tumor dimension (1,2), density (2), and radiomics features (3), has been summarized previously. Deep learning-based survival probability estimation and feature extraction are also feasible. This mini-review briefly introduces both qualitative and quantitative tumor-associated prognostic factors that are extractable from chest CT scans.

Tumor dimensions

The tumor dimensions, which can be measured as diameter or volume, are among the most potent and reliable prognostic factors in lung cancer. These measurements can be easily obtained from CT scans. For clinical staging, the single largest dimension is measured on thin-section CT scans, and the tumor is then classified according to the clinical T categorization system (4). For part-solid nodules, the long axis dimension of the largest solid component is used (4). The solid component on CT scans is a surrogate for the invasive component on microscopy, although the CT measurements often overestimate the pathological measurements (5).

Since the publication of the eighth-edition staging system for lung cancer (6), multiple validation studies have been published (7,8). Those studies have shown that the solid component size is a strong prognostic factor in adenocarcinomas manifesting as part-solid nodules. Nevertheless, the total tumor size, defined as the maximum measurement of the ground-glass component, still matters. The prognoses of part-solid nodules with the same solid component size may vary depending on the total tumor size (9). For cases of clinical T1b disease, Kim et al. (9) suggested upstaging part-solid nodules with a total tumor size larger than 3.0 cm to clinical T1c, considering their worse postoperative survival than clinical T1b nodules with a total tumor size ≤3.0 cm [hazard ratio (HR) =3.796; 95% confidence interval (CI): 1.006, 14.317; P=0.049].

In addition, multiple solid components are often observed in part-solid nodules. The staging manual suggests measuring the single largest solid component, but there is little clinical evidence on how many solid portions should be measured and how to summarize those components into a single value. A recent study reported that measuring multiple solid components did not improve predictions of the prognosis of stage IA lung adenocarcinomas (10). Survival discrimination using the single largest solid portion was comparable to that based on measuring up to three solid portions (C-index, 0.82 vs. 0.79; P=0.25). Furthermore, multiplicity of the solid component was not a prognostic factor for recurrence-free survival (P=0.83) (10).

Tumor volume has been proposed to be more accurate than tumor diameter in patients with early and advanced-stage lung cancer (11,12). Volume is more sensitive for detecting dimensional changes than size, and it is more accurate for nodules with irregular shapes. Gross tumor volume, encompassing both the volume of the primary tumor and metastases, is also an important prognostic factor (1). A recent meta-analysis on the prognostic factors for overall survival in patients with stage III non-small cell lung cancer found that gross tumor volume was significant in 7 out of 9 multivariable analyses (1).

However, there are obstacles to employing tumor volume measurements in practice, including measurement variability and the limited accessibility of easy-to-use, well-validated software. It is also crucial to obtain measurements promptly to avoid delaying the clinical workflow. Deep learning algorithms are expected to show superior segmentation performance for lung cancer than rule-based models and may help solve those obstacles (13). Studies on volume-based cutoffs for clinical staging or response evaluation are warranted.

Presence of ground-glass opacity (GGO)

Early-stage lung adenocarcinomas frequently manifest as part-solid nodules on CT scans. As mentioned above, the solid component is a CT surrogate for the invasive component, and the area of GGO indicates the lepidic component (4). In the past decade, researchers have reported that the presence of GGO in lung adenocarcinomas is associated with better postoperative survival. Aokage et al. (14) suggested that the overall survival in patients with part-solid nodules was longer than in patients with solid nodules in clinical stages IA2 (log-rank test: P<0.01) and IA3 (log-rank test: P<0.01). Hattori et al. (15) observed similar results in clinical stage IA lung adenocarcinomas, using the 5-year overall survival rate as the study outcome (stage IA1: 97.8% versus 86.6%, P=0.026; IA2: 89.3% versus 75.2%, P=0.007; IA3: 88.5% versus 62.3%, P=0.003). Those observations were also supported by parallel pathologic data comparing lepidic versus invasive components (stage IA1: 97.9% versus 85.6%, P=0.031; IA2: 86.1% versus 69.4%, P=0.007; IA3: 77.5% versus 55.8%, P=0.001). In this context, the same authors suggested revising the clinical T categorization system according to the presence or absence of GGO in clinical T1 disease (16). GGO was also associated with long-term survival in a cure model analysis of clinical stage IA lung adenocarcinomas [odds ratio (OR), 0.40, 95% CI: 0.18–0.92, P=0.03] (8).

Nevertheless, debate continues regarding the validity of GGO as a prognostic variable. Some studies have argued that its prognostic value was insignificant after including other prognostic factors (17,18). In fact, prognostic studies on the GGO component of early-stage lung adenocarcinomas have been heterogeneous in terms of cancer stage (e.g., stage IA vs. higher); histology (e.g., the inclusion of non-mucinous adenocarcinomas or non-adenocarcinoma histology); the inclusion of pure ground glass nodules or adenocarcinomas in situ, which usually show an extremely good prognosis; study endpoints; survival analytic methods; and covariates. Therefore, a confirmatory study (e.g., a prospective observational cohort study or meta-analysis) would be required to verify the prognostic role of GGO in adenocarcinomas.

Tumor margin characteristics

In cancer development, the stroma becomes a supportive environment for cancer cells. The tumor stroma combines a desmoplastic reaction with a proliferation of fibroblasts and dense deposition of extracellular matrix (19). Cancer-associated fibroblasts (CAFs) promote malignant growth, angiogenesis, invasion, and metastasis through the secretion of growth factors, cytokines, chemokines, and other immune modulators (19). In the setting of tumor growth, CAFs produce transforming growth factor β, platelet-derived growth factor, and fibroblast growth factor 2, which are profibrotic growth factors (19). Of these, fibroblast growth factor 2 plays a fundamental role in tissue fibrosis and desmoplasia and exerts effects on endothelial cells (20). Accordingly, CAFs and fibrotic stroma are associated with worse prognoses in patients with lung cancer (21).

The CT manifestation of fibrotic stroma or desmoplasia is margin spiculation. However, the association of tumor spiculation with survival has not been well established for lung cancer. Park et al. (22) reported that the presence of spiculation was associated with brain metastasis in resectable-stage lung cancers (OR, 3.34; P=0.006), but it is unclear whether tumor spiculation is truly a prognostic factor for overall survival or recurrence-free survival.

Pleural retraction or tag

Pleural retraction or a pleural tag on CT is a predictor of visceral pleural invasion, which is an established prognostic factor and a T2 descriptor in lung cancer. Nevertheless, CT definitions of pleural tag or retraction are highly variable among radiologists and often yield false-positive diagnoses for pathological visceral pleural invasion. Kim et al. (23) investigated combinations of CT findings, including tumor contact with the pleura, pleural retraction, and tags. None of the combinations were associated with recurrence-free survival in clinical T1N0 lung cancers. Therefore, it is uncertain whether these CT features can be reliably identified and whether these features are genuinely indicative of patients’ survival.

Tumor location

Central tumor location is associated with occult nodal metastasis and mediastinal nodal disease in radiologically node-negative, early-stage lung cancers (24). Therefore, staging guidelines recommend invasive diagnostic procedures such as endobronchial ultrasound-guided transbronchial needle aspiration or mediastinoscopy, both for locally advanced disease and for stage I disease involving central lung cancer (25).

In addition, a central tumor location predicts a less favorable prognosis in pathologically node-negative, early-stage lung cancers. Recent studies revealed that central lung cancer on chest CT, defined either qualitatively or quantitatively, was an independent adverse factor for recurrence-free survival in stage IA lung adenocarcinomas (HR, 2.90; 95% CI: 1.06, 7.96; P=0.04) (26,27). This might be attributable to more crowded vessels and lymphatic vessels in the central lung, leading to a higher chance of undetectable micrometastasis at the time of treatment (26). However, there is no consensus on the radiologic definition of central lung cancer. A subjective assessment of the tumor location is vulnerable to reader variability, and in this context, quantitative measurement using anatomic landmarks has been suggested (24,26). Nevertheless, an automated, objective tool is necessary to exploit the tumor location as a meaningful clinical variable for practice and research.

Deep learning-based prognostic CT features

Radiomics and radiogenomics studies have demonstrated that chest CT scans contain both anatomical information for clinical staging and quantitative prognostic data beyond what human eyes can visualize (28). Deep learning models such as convolutional neural networks have recently enabled an end-to-end analysis of chest CT scans without manual feature engineering. For example, a deep learning model can elicit prognostic factors from a CT tumor patch, either three-dimensional or two-dimensional (29-38). One study proposed a three-dimensional convolutional neural network consisting of dense blocks and transitional layers to estimate the cumulative disease-free survival probability up to 3 years in patients with lung adenocarcinoma undergoing curative resection (29). The model-derived prediction was obtained using a discrete-time survival model and was an independent prognostic factor in an external cohort comprising stage I lung adenocarcinomas (HR, 3.6; 95% CI: 1.6, 8.5; P=0.003) (29). This model was then applied to patients with early-stage lung adenocarcinomas treated with stereotactic ablative radiotherapy and showed good discrimination performance for local recurrence-free survival [area under the curve (AUC), 0.72], disease-free survival (AUC, 0.70), and overall survival (AUC, 0.66) (39). Interestingly, the model showed relatively robust performance regarding the CT acquisition settings considering that the radiotherapy cohort underwent planning CT scans with larger fields of view and lower in-plane resolution than the diagnostic CT scans (39). In addition, multivariable regression analysis showed that the model-derived prognostic factors or CT features were associated with histopathologic risk factors, such as an aggressive adenocarcinoma subtype (cribriform, morular, solid, or micropapillary predominant subtype; OR, 1.03; 95% CI: 1.002, 1.05; P=0.03), venous invasion (OR, 1.03; 95% CI: 1.004, 1.06; P=0.02), and visceral pleural invasion (OR, 1.08; 95% CI: 1.06, 1.10; P<0.001), in patients with resected lung adenocarcinoma (40). Therefore, it can be inferred that the deep learning model could extract CT imaging surrogates for the histopathologic factors of lung adenocarcinomas (40).

Deep learning-based prognostication is also feasible in advanced-stage lung cancers. Deng et al. (35) recently proposed a deep learning model using pre-therapy CT scans, which provided a probability score to identify low-risk or high-risk patients receiving tyrosine kinase inhibitors and immune checkpoint inhibitors. The model was able to identify which patients would receive additional survival benefits beyond the median progression-free survival.

Deep learning is not limited to obtaining the cumulative survival probability; this method can be applied to capture the semantic features of lung cancer more broadly. Ahn et al. (41) reported that automatic measurements of the solid portions of lung cancer were comparable with manual measurements made by radiologists (intraclass correlation coefficient, 0.82–0.89) and showed good agreement with the invasive component size on microscopic examinations (intraclass correlation coefficient, 0.67). Kawaguchi et al. (42) showed that the solid component volume measured using deep learning had higher prognostic discrimination for recurrence or death (AUC, 0.752) than the solid component size (AUC, 0.722) or traditional three-dimensional volumetric analysis (AUC, 0.723). The presence of visceral pleural invasion was also estimated using preoperative CT scans (31). In a study by Choi et al. (31), the diagnostic performance of the deep learning model (AUC, 0.75) for visceral pleural invasion was on par with that of board-certified radiologists (AUC, 0.73–0.79).

Conclusions

CT scans include rich prognostic information that could be extracted either by manual human interpretation or machine learning algorithms. In addition to measurements of tumor dimensions and T categorization in the staging system, CT images can be used to assess tumor volume, density, morphology, margin characteristics, and tumor-based survival probability. Future research should focus on the automated extraction and integrative modeling of these tumor-related prognostic factors.

Acknowledgments

Funding: This study was supported by the Seoul National University Hospital Research Fund (grant No. 03-2022-2170 to Hyungjin Kim). However, the funders had no role in the study design; in the collection, analysis, and interpretation of the data; in the writing of the report; and in the decision to submit the article for publication.

Footnote

Peer Review File: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-22-904/prf

Conflicts of Interest: Both authors have completed the ICMJE uniform disclosure form (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-22-904/coif). CMP serves as an unpaid editorial board member of Translational Lung Cancer Research from September 2021 to August 2023. He also serves as a board member in Korean Society of Radiology, Korean Society of Thoracic Radiology, and Korean Society of Artificial Intelligence in Medicine and participates as a board member of Big Data Review Board of Seoul National University Hospital. He holds stock in Promedius and stock options in Lunit and Coreline Soft. HK received a research grant from Seoul National University Hospital Research Fund (No. 03-2022-2170). He also received consulting fees from RadiSen and holds stock and stock options in Medical IP. 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.

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. van Laar M, van Amsterdam WAC, van Lindert ASR, et al. Prognostic factors for overall survival of stage III non-small cell lung cancer patients on computed tomography: A systematic review and meta-analysis. Radiother Oncol 2020;151:152-75. [Crossref] [PubMed]
2. Nakada T, Kuroda H. Narrative review of optimal prognostic radiological tools using computed tomography for T1N0-staged non-small cell lung cancer. J Thorac Dis 2021;13:3171-81. [Crossref] [PubMed]
3. Wu L, Lou X, Kong N, et al. Can quantitative peritumoral CT radiomics features predict the prognosis of patients with non-small cell lung cancer? A systematic review. Eur Radiol 2023;33:2105-17. [Crossref] [PubMed]
4. Travis WD, Asamura H, Bankier AA, et al. The IASLC Lung Cancer Staging Project: Proposals for Coding T Categories for Subsolid Nodules and Assessment of Tumor Size in Part-Solid Tumors in the Forthcoming Eighth Edition of the TNM Classification of Lung Cancer. J Thorac Oncol 2016;11:1204-23.
5. Lee KH, Goo JM, Park SJ, et al. Correlation between the size of the solid component on thin-section CT and the invasive component on pathology in small lung adenocarcinomas manifesting as ground-glass nodules. J Thorac Oncol 2014;9:74-82. [Crossref] [PubMed]
6. 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]
7. Hwang JK, Page BJ, Flynn D, et al. Validation of the Eighth Edition TNM Lung Cancer Staging System. J Thorac Oncol 2020;15:649-54.
8. Kim H, Goo JM, Kim YT, et al. Validation of the Eighth Edition Clinical T Categorization System for Clinical Stage IA, Resected Lung Adenocarcinomas: Prognostic Implications of the Ground-Glass Opacity Component. J Thorac Oncol 2020;15:580-8.
9. Kim H, Goo JM, Suh YJ, et al. Implication of total tumor size on the prognosis of patients with clinical stage IA lung adenocarcinomas appearing as part-solid nodules: Does only the solid portion size matter? Eur Radiol 2019;29:1586-94. [Crossref] [PubMed]
10. Kim H, Goo JM, Suh YJ, et al. Measurement of Multiple Solid Portions in Part-Solid Nodules for T Categorization: Evaluation of Prognostic Implication. J Thorac Oncol 2018;13:1864-72. [Crossref] [PubMed]
11. Takenaka T, Yamazaki K, Miura N, et al. The Prognostic Impact of Tumor Volume in Patients with Clinical Stage IA Non-Small Cell Lung Cancer. J Thorac Oncol 2016;11:1074-80. [Crossref] [PubMed]
12. Nishino M, Dahlberg SE, Cardarella S, et al. Tumor volume decrease at 8 weeks is associated with longer survival in EGFR-mutant advanced non-small-cell lung cancer patients treated with EGFR TKI. J Thorac Oncol 2013;8:1059-68. [Crossref] [PubMed]
13. Lee YS, Hong N, Witanto JN, et al. Deep neural network for automatic volumetric segmentation of whole-body CT images for body composition assessment. Clin Nutr 2021;40:5038-46. [Crossref] [PubMed]
14. Aokage K, Miyoshi T, Ishii G, et al. Influence of Ground Glass Opacity and the Corresponding Pathological Findings on Survival in Patients with Clinical Stage I Non-Small Cell Lung Cancer. J Thorac Oncol 2018;13:533-42. [Crossref] [PubMed]
15. Hattori A, Hirayama S, Matsunaga T, et al. Distinct Clinicopathologic Characteristics and Prognosis Based on the Presence of Ground Glass Opacity Component in Clinical Stage IA Lung Adenocarcinoma. J Thorac Oncol 2019;14:265-75. [Crossref] [PubMed]
16. Hattori A, Takamochi K, Oh S, et al. New revisions and current issues in the eighth edition of the TNM classification for non-small cell lung cancer. Jpn J Clin Oncol 2019;49:3-11.
17. Okubo Y, Kashima J, Teishikata T, et al. Prognostic Impact of the Histologic Lepidic Component in Pathologic Stage IA Adenocarcinoma. J Thorac Oncol 2022;17:67-75. [Crossref] [PubMed]
18. Phillips WW, Gill RR, Mazzola E, et al. Impact of Nodule Density in Women With Sublobar Resection for Stage IA Adenocarcinoma. Ann Thorac Surg 2021;112:1067-75. [Crossref] [PubMed]
19. Bremnes RM, Dønnem T, Al-Saad S, et al. The role of tumor stroma in cancer progression and prognosis: emphasis on carcinoma-associated fibroblasts and non-small cell lung cancer. J Thorac Oncol 2011;6:209-17. [Crossref] [PubMed]
20. Donnem T, Al-Shibli K, Al-Saad S, et al. Prognostic impact of fibroblast growth factor 2 in non-small cell lung cancer: coexpression with VEGFR-3 and PDGF-B predicts poor survival. J Thorac Oncol 2009;4:578-85. [Crossref] [PubMed]
21. Min KW, Kim DH, Noh YK, et al. Cancer-associated fibroblasts are associated with poor prognosis in solid type of lung adenocarcinoma in a machine learning analysis. Sci Rep 2021;11:16779. [Crossref] [PubMed]
22. Park S, Lee SM, Ahn Y, et al. Identification of predictors for brain metastasis in newly diagnosed non-small cell lung cancer: a single-center cohort study. Eur Radiol 2022;32:990-1001. [Crossref] [PubMed]
23. Kim H, Goo JM, Kim YT, et al. CT-defined Visceral Pleural Invasion in T1 Lung Adenocarcinoma: Lack of Relationship to Disease-Free Survival. Radiology 2019;292:741-9. [Crossref] [PubMed]
24. Kim H, Choi H, Lee KH, et al. Definitions of Central Tumors in Radiologically Node-Negative, Early-Stage Lung Cancer for Preoperative Mediastinal Lymph Node Staging: A Dual-Institution, Multireader Study. Chest 2022;161:1393-406. [Crossref] [PubMed]
25. NCCN Clinical Practice Guidelines in Oncology Version 6.2020 - June 15, 2020: National Comprehensive Cancer Network, 2020. Available online: https://www.nccn.org/professionals/physician_gls/pdf/nscl.pdf
26. Choi H, Kim H, Park CM, et al. Central Tumor Location at Chest CT Is an Adverse Prognostic Factor for Disease-Free Survival of Node-Negative Early-Stage Lung Adenocarcinomas. Radiology 2021;299:438-47. [Crossref] [PubMed]
27. Kang SW, Jeong WG, Lee JE, et al. Prognostic significance of location index in resected T1-sized early-stage non-small cell lung cancer. Acta Radiol 2023;64:1028-37. [Crossref] [PubMed]
28. Wang T, She Y, Yang Y, et al. Radiomics for Survival Risk Stratification of Clinical and Pathologic Stage IA Pure-Solid Non-Small Cell Lung Cancer. Radiology 2022;302:425-34. [Crossref] [PubMed]
29. Kim H, Goo JM, Lee KH, et al. Preoperative CT-based Deep Learning Model for Predicting Disease-Free Survival in Patients with Lung Adenocarcinomas. Radiology 2020;296:216-24. [Crossref] [PubMed]
30. Kim H, Lee D, Cho WS, et al. CT-based deep learning model to differentiate invasive pulmonary adenocarcinomas appearing as subsolid nodules among surgical candidates: comparison of the diagnostic performance with a size-based logistic model and radiologists. Eur Radiol 2020;30:3295-305. [Crossref] [PubMed]
31. Choi H, Kim H, Hong W, et al. Prediction of visceral pleural invasion in lung cancer on CT: deep learning model achieves a radiologist-level performance with adaptive sensitivity and specificity to clinical needs. Eur Radiol 2021;31:2866-76. [Crossref] [PubMed]
32. She Y, Jin Z, Wu J, et al. Development and Validation of a Deep Learning Model for Non-Small Cell Lung Cancer Survival. JAMA Netw Open 2020;3:e205842. [Crossref] [PubMed]
33. Hosny A, Parmar C, Coroller TP, et al. Deep learning for lung cancer prognostication: A retrospective multi-cohort radiomics study. PLoS Med 2018;15:e1002711. [Crossref] [PubMed]
34. Chen W, Hou X, Hu Y, et al. A deep learning- and CT image-based prognostic model for the prediction of survival in non-small cell lung cancer. Med Phys 2021;48:7946-58. [Crossref] [PubMed]
35. Deng K, Wang L, Liu Y, et al. A deep learning-based system for survival benefit prediction of tyrosine kinase inhibitors and immune checkpoint inhibitors in stage IV non-small cell lung cancer patients: A multicenter, prognostic study. EClinicalMedicine 2022;51:101541. [Crossref] [PubMed]
36. Lian J, Long Y, Huang F, et al. Imaging-Based Deep Graph Neural Networks for Survival Analysis in Early Stage Lung Cancer Using CT: A Multicenter Study. Front Oncol 2022;12:868186. [Crossref] [PubMed]
37. Zheng S, Guo J, Langendijk JA, et al. Survival prediction for stage I-IIIA non-small cell lung cancer using deep learning. Radiother Oncol 2023;180:109483. [Crossref] [PubMed]
38. Hou KY, Chen JR, Wang YC, et al. Radiomics-Based Deep Learning Prediction of Overall Survival in Non-Small-Cell Lung Cancer Using Contrast-Enhanced Computed Tomography. Cancers (Basel) 2022;14:3798. [Crossref] [PubMed]
39. Kim H, Lee JH, Kim HJ, et al. Extended application of a CT-based artificial intelligence prognostication model in patients with primary lung cancer undergoing stereotactic ablative radiotherapy. Radiother Oncol 2021;165:166-73. [Crossref] [PubMed]
40. Nam JG, Park S, Park CM, et al. Histopathologic Basis for a Chest CT Deep Learning Survival Prediction Model in Patients with Lung Adenocarcinoma. Radiology 2022;305:441-51. [Crossref] [PubMed]
41. 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]
42. 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]