Integrating Radiomics Features and CT Semantic Characteristics for Predicting Visceral Pleural Invasion in clinical stage Ia peripheral Lung Adenocarcinoma

doi:10.21203/rs.3.rs-4936484/v1

Download PDF

Article

Integrating Radiomics Features and CT Semantic Characteristics for Predicting Visceral Pleural Invasion in clinical stage Ia peripheral Lung Adenocarcinoma

https://doi.org/10.21203/rs.3.rs-4936484/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Objectives

The aim of this study was to non-invasively predict the visceral pleural invasion (VPI) of peripheral lung adenocarcinoma (LA) highly associated with pleura of clinical stage Ia based on preoperative chest computed tomography (CT) scanning.

Methods

A total of 537 patients diagnosed with clinical stage Ia LA underwent resection and were stratified into training and external validation cohorts at a ratio of 7:3. Radiomics features were extracted using PyRadiomics software following tumor lesion segmentation and were subsequently filtered through spearman correlation analysis, minimum redundancy maximum relevance, and least absolute shrinkage and selection operator regression analysis. Univariate and multivariable logistic regression analyses were conducted to identify independent predictors. A predictive model was established with visual nomogram and external validation, and evaluated in terms of area under the receiver operating characteristic curve (AUC).

Results

The independent predictors of VPI were identified: pleural attachment (p < 0.001), pleural contact angle (p = 0.018) and Rad-score (p < 0.001). The combined model showed good calibration with an AUC of 0.822 (95% confidence intervals (CI): 0.785, 0.869), compared with 0.719 (95% CI: 0.677, 0.760; DeLong’s test p < 0.001) when radiomics was used alone. For validation group, the accuracy of combined prediction model was reasonable with an AUC of 0.785 (95% CI: 0.742, 0.821).

Conclusion

Our predictive model, which integrated radiomics features of primary tumors and peritumoral CT semantic characteristics, offers a non-invasive method for evaluating VPI in patients with clinical stage Ia LA. Additionally, it provides prognostic information and supports surgeons in making more personalized treatment decisions.

Health sciences/Oncology/Cancer

Health sciences/Diseases/Cancer/Lung cancer/Non small cell lung cancer

Computed Tomography

Radiomics

Adenocarcinoma

Non-small Cell Lung Cancer

Visceral Pleural Invasion

1. Peritumoral CT semantic characteristics were significantly associated with VPI in peripheral LA patients with clinical stage Ia, instead of the CT semantic features of primary tumor.

2. Radiomics analysis of the primary tumor has the potential to enhance the predictive accuracy of models, as evidenced by a significant increase in the AUC of the combined model.

Lung cancer represents a significant proportion of cancer-related mortality worldwide, of which approximately 80–85% is non-small cell lung cancer (NSCLC) and lung adenocarcinoma (LA) is the most common histological subtype ^{[1, 2]}. In the past decade, the implementation of early screening for lung cancer and advancements in low-dose thin-layer computed tomography (CT) technology have led to an increase in the incidence of early and small volume lung cancer. The term visceral pleural invasion (VPI) refers to the infiltration of a tumor beyond the elastic layer, with or without visible exposure on the pleural surface, while not extending to neighboring anatomical structures ^[3]. For decades, the presence of VPI has been identified as a negative prognostic indicator in NSCLC with a tumor diameter of less than 3cm ^[4]. Specifically, the addition of VPI elevates the T-stage from T1 to T2 and advances tumor staging from Ia to Ib ^[5]. Consequently, clinical treatment strategies and patient outcomes differ significantly. The recommended surgical approach shifts from segmental resection to lobectomy with mediastinal lymph node (LN) dissection, when T1 stage lesion is accompanied by VPI ^[6].

Currently, chest CT is widely considered the preferred non-invasive preoperative method for detecting lung diseases. The identification of VPI on CT scans presents challenges, even in cases of NSCLC where the tumor is closely located to the pleura or associated with pleural indentation. Radiomics uses machine-learning method to convert images into data, providing quantifiable, objective, and automated analysis and has been extensively employed in the evaluation of NSCLC ^[7]. Previous researches ^{[8, 9]} have indicated that VPI is rarely observed in pure ground glass nodules (pGGN) due to the limited ability of these lesions to penetrate the thick elastic layer.

Consequently, the current research centered on peripheral LA located adjacent to the pleura or exhibiting pleural indentation/tag in clinical stage Ia, and conducting a retrospective, dual-center clinical investigation. The aim was to predict VPI preoperatively and non-invasively by integrating peritumoral CT semantic features and radiomics analysis of the primary tumor, thus aiding surgeons in the selection of suitable treatment approaches and surgical interventions.

2.1. Patients

The study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). The study was approved by the Medical Research Ethics Committee and the Institutional Review Board of () (No.:) and () (No. :), and individual consent for this retrospective analysis was waived. Initially, a total of 741 patients were enrolled with surgery date from January 2016 to September 2023 according to the following inclusion criteria: (a) patients with peripheral LA who underwent surgical resection; (b) available chest thin-layer CT plain images within 1 month before surgery from the picture archiving and communication system; (c) patients identified as clinical stage Ia and available pathological information; and (d) lesions located adjacent to the pleura or exhibiting pleural indentation/tag. We excluded patients who had received preoperative treatment (n = 6), lesions with a pure ground glass density (n = 89), images of poor quality or inability to identify lesions (n = 17), no available VPI information (n = 56), as well as cases of pathologically confirmed multiple primary lung cancer and carcinoma in situ (n = 36) (Fig. 1). Finally, a total of 537 patients were enrolled, of which 385 patients were from ( ) and other 152 patients were from ( ). The cases were allocated into training and external validation groups in a randomized manner, with a distribution ratio of 70% and 30% respectively.

Clinical data, encompassing age, gender, smoking habits, family history, and genetic mutation status, were extracted from the clinical database. LA was classified in accordance with the 2015 World Health Organization classification system ^[10]. The histological subtypes were documented on the form according to their proportion when present in more than 5%. The minimally invasive adenocarcinoma subtype was categorized as low-risk, while the acinar, lepidic, and papillary patterns in invasive LA were classified as moderate-risk subtypes. Conversely, invasive mucinous LA, colloid, fetal and enteric LA, the micropapillary and solid patterns in invasive LA were classified as high-risk subtypes. Tumor lymph node metastasis classification and tumor staging were conducted following the guidelines of the 8th edition of the staging system published by the Union for International Cancer Control and the American Joint Committee on Cancer ^[5].

2.2. Evaluation of clinical stage Ia and VPI

All participants in the study underwent preoperative chest CT scans to assess tumor size and LN status. Further diagnostic tests, including whole abdomen CT, brain magnetic resonance imaging, bone scintigraphy, or positron emission tomography, were performed to confirm the absence of distant metastasis. The criteria for classifying cN0 on CT scans was based on the short-axis diameter of all LNs being less than 10 mm. Elastic van Gieson staining was utilized to evaluate the presence of VPI by a specialized pathologist () who was blinded to the patients' clinical and radiological data. The classification of pleural invasion level (PL) is defined as follows: PL0, no pleural invasion, PL1, tumor invasion of the visceral pleural elastic layer without reaching the surface of the visceral pleura, PL2, tumor invasion of the visceral pleural surface, and PL3, tumor invasion of the parietal pleura or chest wall ^[11]. Patients were categorized into two groups based on the presence of VPI. Pathological findings of PL0 were designated as VPI-negative, while PL1 and PL2 were classified as VPI-positive according to established criteria outlined in a prior investigation ^[3].

2.3. CT scanning protocol

Chest CT examinations were performed using five multidetector CT systems of three types: Lightspeed16, GE Healthcare, Milwaukee, WI, USA; Somatom Sensation 64, Siemens, Erlangen, Germany; Discovery CT750 HD, GE Healthcare. The scanning parameters were: (a) 120 kVp with the automatic regulation of the tube current and 1.5-mm reconstruction thickness and intervals for the 64-detector scanner and (b) 120 kVp, 150–200 mAs, and 1.25-mm reconstruction thickness intervals for the other two types of scanners.

2.4. CT image interpretation and preprocessing

Two experienced clinical radiologists, one with 5 years and the other with 8 years of experience in CT imaging of thoracic malignancies, independently analyzed the CT images and confirmed the peripheral lesions highly associated with pleura after training. The clinical and pathological information was withheld from the radiologists, who were subsequently guided by a senior radiologist to address any discrepancies in image interpretation and reach a consensus. The detailed CT semantic descriptors and scoring criteria for primary tumor and peritumor are outlined in Table 1. All images were displayed at a lung window width of 1500 HU, window level of -600 HU, and a mediastinal window width of 350 HU at level of 40 HU. The evaluation of CT descriptors was conducted on multi-planar reconstructed images and documented using a standardized scoring sheet.

2.5. Tumor segmentation and features extraction

The CT images were resampled to a thickness of 1 mm using a linear interpolation algorithm for initial image preprocessing, followed by Gaussian filtering for further linear smoothing. Tumor segmentation was carried out independently by two radiologist utilizing ITK-snap 3.6.0, using the manual method of drawing regions of interest (ROI) on the processed CT images at the lung window. The radiologists were informed of the tumors' location but remained blinded to additional pertinent information. Radiomics features were automatically extracted using the PyRadiomics 3.0 program, an open-source software (http://www.radiomics.io/pyradiomics.html) ^[12]. A total of 1316 radiomics features were extracted from the 3D ROI of CT images, including first-order (n = 108), shape (n = 14), gray level cooccurrence matrix (n = 144), gray level dependence matrix (GLDM, n = 84), gray level run-length matrix (n = 96), gray level size zone matrix (GLSZM, n = 96), neighboring gray tone difference matrix (n = 30) and higher-order wavelet features (n = 744).

2.6. Feature selection and establishment of radiomics signature

The radiomics parameters extracted by two radiologists were averaged after standardizing using the Z-score method. The study employed Spearman pairwise correlation analysis to assess the strength of correlations between features, eliminating those with an absolute correlation value exceeding 0.9. Subsequently, the top 100 features selected by the minimum redundancy maximum relevance (mRMR) method were screened utilizing the least absolute shrinkage and selection operator (LASSO) regression analysis to determine the optimal subsets for evaluating invasive pathological characteristics. The identified features were used to develop a logistic regression model to eliminate non-statistically significant variables. The radiomics score (Rad-score) formula was then derived by linearly combining the selected features with their coefficients as weights:

Radscore = b + Ci × Xi,

where b represents a constant term, Xi denotes the value of the selected feature, and Ci represents the regression coefficient associated with the selected feature. By applying this formula, the Rad-score for each patient was calculated to assess the disparity between different groups.

2.7. Statistical analysis

Statistical analyses were conducted using the software programs R 4.3.0 and SPSS 26.0. Agreement between two readers was assessed using the ĸ index and Kendall coefficient of concordance. The non-parametric two-sample Wilcoxon test was utilized for ranked or continuous variables, while chi-square or Fisher’s test was employed for categorical variables in univariate analysis. Subsequently, multivariate logistic regression analysis was conducted to evaluate the ability to identify VPI in various models. The predictive accuracy of the models was assessed using the bootstrap approach. A predictive model was generated using a 90% random sample of the data, which was repeated 1000 times and the results were averaged with 95% bootstrap confidence intervals (CI). The final prediction model was visualized using a nomogram and evaluated by the Hosmer-Lemeshow test and calibration curve. To assess the potential clinical diagnostic impact of the nomogram model, a decision curve analysis (DCA) was conducted to quantify the net benefits of utilizing the model at various threshold probabilities. Area under the receiver operating characteristic (ROC) curve (AUC) was used to assess the predictive efficacy, and Delong test would compare the AUC of different models. P values < 0.05 were regarded as statistically significant.

3.1. Inter-observer consistency analysis and patient demographics

Agreement among the two readers was good (Table S1). The intraclass correlation coefficient for maximum diameter, consolidation diameter, tumor shadow disappear rate, pleural contact length and distance from the lesions to pleura (DLP) was 0.92 (range, 0.90–0.94), 0.86 (range, 0.84–0.89), 0.88 (range, 0.86–0.90), 0.91 (range, 0.87–0.92) and 0.92 (range, 0.89–0.95), respectively. No significant difference of either feature was observed between training and validation group (Table S2).

3.2. Correlation of VPI with clinical and CT semantic features in training group

Thirty-two patients in training group developed VPI as confirmed by postoperative pathology. Accordingly, the patients were categorized into VPI and negative groups. The association between clinical features with VPI was presented in Table 2. Significantly, Patients with epidermal growth factor receptor (EGFR) of wild-type [8/63 (12.7%) vs. 6/132 (4.5%)] developed VPI more frequently than EGFR mutant patients (odds ratio (OR) = 3.06, 95% CI: 1.01, 9.22; p = 0.039). No significant association was noted for other clinical features.

Univariate analysis showed that tumors with smaller DLP (p < 0.001), larger pleural contact length (p < 0.001) and pleural contact angle (OR = 4.57, 95% CI: 1.92, 10.84 for score 1; OR = 3.28, 95% CI: 1.56, 6.87 for score 2; p < 0.001), and pleural attachment (OR = 4.57, 95% CI: 1.92, 10.84; p < 0.001) were more likely to develop VPI (Table 3; Fig. 2). Additionally, the CT semantic features of the primary tumor did not show statistical significance (Table S3).

3.3. Screening and integration of radiomics features

A comprehensive analysis was conducted on a dataset comprising 1316 radiomics features extracted from the three-dimensional ROI in each CT image. After Spearman pairwise correlation analysis and the mRMR method, 17 of the top 100 features were identified using the LASSO algorithm to mitigate the risk of overfitting (Table 4; Fig. 3). Additionally, 12 features were excluded through logistic regression with backward step-wise selection to eliminate non-significant variables. This process ultimately identified five radiomics features, which were utilized to generate the Rad-score formula weighted by their respective coefficients (Table 4). Notably, the Rad-score exhibited a significant difference between the two groups (p < 0.001), the mean value in VPI group was significantly higher (2.47 ± 1.68) than that in negative group (–0.92 ± 1.85).

3.4. Predictive model and external validation test

The multivariable logistic regression analysis revealed that pleural attachment (OR = 3.25, 95% CI: 1.84, 4.92, p < 0.001), larger pleural contact angle (OR = 1.92, 95% CI: 1.33, 2.86, p = 0.018) and Rad-score (OR = 2.71, 95% CI: 2.03, 3.62, p < 0.001) were significant independent predictors (Table 5). A nomogram integrating these predictors was constructed and displayed in Fig. 4. The outcome of the Hosmer-Lemeshow goodness-of-fit test yielded a non-significant result (p = 0.652), suggesting a strong agreement between the anticipated and actual probabilities. The calibration curve and DCA were depicted in Fig. 5. The DCA demonstrated that the utilization of the nomogram could offer a supplementary net advantage in contrast to the "treat-all" or "treat-none" approach within a threshold probability range of 10 to 45%, indicating the clinical utility of the nomogram.

The AUC of combined model increased to 0.822 (95% CI: 0.785, 0.869), compared with 0.719 (95% CI: 0.677, 0.760; DeLong’s test p < 0.001) of radiomics model (Fig. 6A). The combined model exhibited a sensitivity of 86.7%, specificity of 64.2%, and accuracy of 78.3% when the cutoff was determined at the maximum Youden index. In the external validation group, the accuracy of the combined prediction model was reasonable with an AUC of 0.785 (95% CI: 0.742, 0.821) with a sensitivity of 84.6%, specificity of 61.6%, and accuracy of 74.0%, compared with 0.728 (95% CI: 0.694, 0.763; DeLong’s test p < 0.001) when utilizing radiomics features in isolation (Fig. 6B).

VPI has been regarded as an adverse prognostic factor in LA because of the potential increased risk of cancer recurrence and reduced overall survival, even in small lung neoplasms no more than 2 cm in diameter or with ground-glass opacity [8, 13]. Kudo et al. established that patients with VPI may exhibit a broader spectrum of LN metastases due to tumors near the pleura spreading quickly through the extensive network of lymphatic vessels in the visceral pleura ^[14]. Assessing the likelihood of pleural infiltration is important for surgical decisions, treatment selection, and prognosis ^[15]. However, traditional intraoperative pathological frozen section diagnosis of VPI is time-consuming and inaccurate, with a reported accuracy rate of 56.5% ^[16]. The current gold standard for diagnosing VPI still involves postoperative staining of elastic fibers. The precise preoperative assessment in cases of adjacent pleural-associated lung cancer using imaging techniques could significantly influence surgical decision-making, a task that remains changeling. Subpleural LA typically manifests as pGGN that, because of their limited invasiveness, do not breach the visceral pleura ^{[17, 18]}. Thus, the study population was limited to patients with peripheral pleural-associated LA at clinical stage Ia and excluding pGGN.

Seok et al. identified several key CT features, including pleural contact, pleural thickening, solid proportion > 50%, and nodule size > 2 cm, as significant indicators of VPI in T1 peripheral LA ^[19]. Yanagawa et al. demonstrated a strong association between subsolid nodules with a consolidation-to-tumor ratio > 63% and VPI ^[20]. Manac'h et al. found that larger tumors were more likely to have VPI, with rates of 10.4% for tumors < 3 cm, 19.6% for tumors between 3 and 5 cm, and 33% for tumors > 5 cm ^[21]. In our investigation, no statistically significant disparities were observed between VPI with the proportions of solid components, maximum tumor diameters, or maximum consolidation diameters. These incongruous outcomes may be attributed to variations in inclusion criteria and potential sample selection bias. Our study exclusively enrolled peripheral LA nodules smaller than 3cm while excluding pGGN and those unrelated to the pleura. Furthermore, no association was identified in our study between CT semantic features of primary tumors and VPI, aligning with prior research findings indicating that spiculation, lobulation, and air bronchogram were not reliable indicators for evaluating VPI ^[22–24]. Air bronchogram is characterized by tumor cells spreading in a convoluted manner along bronchioles and alveoli without disrupting the lung architecture, and is correlated with the less invasive nature of LA ^[25]. Conversely, the presence of lobulation and spiculation signifies irregular tumor growth, with tumor cells infiltrating nearby blood and lymphatic vessels, suggesting a higher degree of invasiveness ^[26]. Nevertheless, the extent of tumor invasiveness alone does not comprehensively capture the true nature of VPI. Thus, this study placed greater emphasis on analyzing the CT semantic features surrounding the tumor.

Previous research ^{[27, 28]} has highlighted the significance of factors such as the DLP, as well as the angle and length of the contact interface, in predicting VPI, findings which align with our results. Hsu and Zhao et al. ^{[29, 30]} have underscored the potential utility of pleural attachment and indentation as valuable adjuncts for enhancing the early diagnostic accuracy of VPI in NSCLC. However, our study found no significant correlation between the pleural indentation and VPI. Pleural indentation is the result of the thickening of interlobular septa, which extends from the tumor surface to the pleural surface and introduces tension to the pleura ^[16].The association between pleural indentation and lesion malignancy remains a topic of debate. Gallagher et al. ^[31] suggested that elastosis, inflammatory invasion, and thick fibroblast proliferation contribute to pleural indentation, which consequently is only a reflection of pleural fiber tension, not VPI. Pleural attachment has been determined to be a significant risk factor for VPI in both univariate and multivariate regression analyses, surpassing DLP in predictive value due to the enhanced conditions for pleural invasion resulting from direct contact. Although many studies ^{[29, 32, 33]} have focused on the morphological characteristics of VPI as depicted on CT images, the accuracy remains limited, with the identified morphological features being subject to the expertise and interpretation of radiologists.

The close proximity of the tumor to the pleura posed difficulties in precisely delineating the ROI encompassing the tumor. As a result, this radiomics study specifically concentrated on the internal characteristics of the tumor. The utilization of Spearman pairwise correlation, mRMR method, and LASSO regression analysis served to enhance the efficacy of the feature selection process and mitigate feature redundancy, ultimately leading to the identification of five optimal quantitative radiomics features, namely total energy, skewness, large area low gray level emphasis (LALGLE), dependence non uniformity normalized (DNUN), and dependence entropy. These features encompass characteristics ranging from first-order to higher-order texture, aligning partially with findings from Wei et al.'s prior investigation ^[34]. Skewness and total energy are classified as first-order parameters, with lower values indicating greater lesion heterogeneity. LALGLE, a parameter of the GLSZM, also reflects lesion heterogeneity, with higher values indicating increased heterogeneity. The dependence entropy and DNUN, calculated from the GLDM, demonstrate the relationship between the gray-level intensity of CT voxels and the invasiveness of GGN. A higher value of these features suggests increased heterogeneity in texture patterns, which indicating malignant trait of tumors, encompasses localized variances in tumor proliferation, metabolic activity, cell apoptosis, and blood supply ^[35]. No shape category feature was identified as the optimal feature, potentially due to the morphological similarities and overlaps between VPI and Negative group, similar with the CT semantic findings. We constructed a nomogram model based on radiomics and peritumoral semantic features to identify VPI in peripheral LA patients with clinical stage Ia. The pairwise Delong test evaluation showed that the AUC value of the combined model was significantly higher than that of the radiomics model (p < 0.05).

However, our study has certain limitations. It is important to note that this study is retrospective in nature, which may introduce selection bias. Additionally, variations in CT scanning devices and acquisition protocols could impact the consistency of radiomics features. Furthermore, the feasibility and reproducibility of volume segmentation in clinical practice may be limited, with potential time constraints. Collaborative multi-center research is necessary to confirm the reliability and generalizability of the predictive model proposed in this study.

In summary, this research has introduced a non-invasive predictive tool utilizing radiomics features of primary tumors and peritumoral CT semantic characteristics, which may be integrated into the treatment decision-making process for predicting VPI in peripheral LA patients with clinical stage Ia. This model can assist clinicians in evaluating the probability of VPI prior to surgery, enabling the identification of patients who may benefit from limited surgical interventions.

Author Contribution

1. guarantor of integrity of the entire study --------- Fengnian Zhao2. study concepts and design --------------------------- Fengnian Zhao3. literature research ------------------------------------- Yunqing Zhao4. clinical studies ----------------------------------------- Qingna Yan, Yunqing Zhao5. experimental studies / data analysis --------------- Fengnian Zhao, Zhaoxiang Ye6. statistical analysis ------------------------------------- Yunqing Zhao7 manuscript preparation ------------------------------ Yunqing Zhao8 manuscript editing ------------------------------------ Haoran Sun

Data Availability

The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.

Siegel, R. L., Miller, K. D., Fuchs, H. E. & Jemal, A. Cancer statistics, 2022. CA Cancer J. Clin. 72 (1), 7–33 (2022).
Miller, K. D. et al. Cancer treatment and survivorship statistics, 2022. CA Cancer J. Clin. 72 (5), 409–436 (2022).
Travis, W. D. et al. International Staging Committee. International staging C: visceral pleural invasion: pathologic criteria and use of elastic stains: proposal for the 7th edition of the TNM classification for lung cancer. J. Thorac. Oncol. 3, 1384–1390 (2008).
Butnor, K. J. & Travis, W. D. Recent advances in our understanding of lung cancer visceral pleural invasion and other forms of minimal invasion: implications for the next TNM classification. Eur. J. Cardiothorac. Surg. 43, 309–311 (2013).
Rami-Porta, R. et al. The IASLC Lung Cancer Staging Project Proposals for the Revisions of the T Descriptors in the Forthcoming Eighth Edition of the TNM Classification for Lung Cancer. J. Thorac. Oncol. 10 (7), 990–1003 (2015).
Zhang, T. et al. Visceral pleural invasion in T1 tumors (=3cm), particularly T1a, in the eighth tumor node-metastasis classification system for non-small cell lung cancer: a population-based study</at. J. Thorac. Dis. 11, 2754–2762 (2019).
Huang, Y. et al. Radiomics signature: A potential biomarker for the prediction of disease-free survival in early-stage (I or II) non-small cell lung cancer. Radiology. 281, 947–957 (2016).
Heidinger, B. H. et al. Visceral pleural invasion in pulmonary adenocarcinoma: differences in CT patterns between solid and subsolid cancers. Radiol. Cardiothorac. Imaging. 1, e190071 (2019).
Fan, L. et al. Radiomics signature: a biomarker for the preoperative discrimination of lung invasive adenocarcinoma manifesting as a ground-glass nodule. Eur. Radiol. 29, 889–897 (2019).
Travis, W. D. et al. The 2015 World Health Organization Classification of Lung Tumors Impact of Genetic, Clinical and Radiologic Advances Since the 2004 Classification. J. Thorac. Oncol. 10 (9), 1243–1260 (2015).
Oyama, M. et al. Prognostic impact of pleural invasion in 1488 patients with surgically resected non-small cell lung carcinoma. Jpn J. Clin. Oncol. 43 (5), 540–546 (2013).
Van Griethuysen, J. J. M. et al. Computational Radiomics System to Decode the Radiographic Phenotype. Cancer Res. 77 (21), E104–E7 (2017).
Zhang, X. et al. Prognostic value of visceral pleural invasion in the stage pT1-2N2M0 non-small cell lung cancer: a study based on the SEER registry. Curr. Probl. Cancer. 45, 100640 (2021).
Kudo, Y. et al. Impact of visceral pleural invasion on the survival of patients with non-small cell lung cancer. Lung Cancer. 78, 153–160 (2012).
Zhao, L. L. et al. Visceral pleural invasion in lung adenocarcinoma =3cm with ground-glass opacity: a clinical, pathological and radiological study</at. J. Thorac. Dis. 8, 1788–1797 (2016).
Goldstraw, P. et al. The IASLC lung cancer staging project: proposals or revision of the TNM stage groupings in the forthcoming (Eighth) edition of the TNM classification for lung cancer. J. Thorac. Oncol. 11 (1), 39–51 (2016).
Zhao, Q., Wang, J. W., Yang, L., Xue, L. Y. & Lu, W. W. CT diagnosis of pleural and stromal invasion in malignant subpleural pure ground-glass nodules: an exploratory study. Eur. Radiol. 29 (1), 279–286 (2019).
Shi, J. et al. The combination of computed tomography features and circulating tumor cells increases the surgical prediction of visceral pleural invasion in clinical T1N0M0 lung adenocarcinoma. Transl Lung Cancer Res. 10 (11), 4266–4280 (2021).
Seok, Y. & Lee, E. Visceral pleural invasion is a significant prognostic factor in patients with partly solid lung adenocarcinoma sized 30 mm or smaller. Thorac. Cardiovasc. Surg. 66 (2), 150–155 (2018).
Yanagawa, M. et al. Prognostic importance of volumetric measurements in stage I lung adenocarcinoma. Radiology. 272 (2), 557–567 (2014).
Manac'h, D. et al. Visceral pleura invasion by nonsmall cell lung cancer: an underrated bad prognostic factor. Ann. Thorac. Surg. 71 (4), 1088–1093 (2001).
Ahn, S. Y. et al. Predictive CT features of visceral pleural invasion by T1-sized peripheral pulmonary adenocarcinomas manifesting as subsolid nodules. AJR Am. J. Roentgenol. 209, 561–566 (2017).
Deng, H. Y. et al. Novel biologic factors correlated to visceral pleural invasion in early-stage non-small cell lung cancer less than 3 cm. J. Thorac. Dis. 10 (4), 2357–2364 (2018).
Wang, F. et al. Predicting visceral pleural invasion in lung adenocarcinoma presenting as part-solid density utilizing a nomogram model combined with radiomics and clinical features. Thorac. Cancer. 15 (1), 23–34 (2024).
Lederlin, M. et al. Correlation of Radio- and Histomorphological Pattern of Pulmonary Adenocarcinoma. Eur. Respir J. 41, 943–951 (2013).
Wang, Y. et al. Multivariate analysis based on the maximum standard unit value of ¹⁸F-fluorodeoxyglucose positron emission tomography/computed tomography and computed tomography features for preoperative predicting of visceral pleural invasion in patients with subpleural clinical stage IA peripheral lung adenocarcinoma. Diagn. Interv Radiol. 29 (2), 379–389 (2023).
Glazer, H. S. et al. Pleural and chest wall invasion in bronchogenic carcinoma: CT evaluation. Radiology. 157 (1), 191–194 (1985).
Chen, Z., Jiang, S., Li, Z., Rao, L. & Zhang, X. Clinical value of ¹⁸F-FDG PET/CT in prediction of visceral pleural invasion of subsolid nodule stage I lung adenocarcinoma. Acad. Radiol. 27 (12), 1691–1699 (2020).
Hsu, J. S. et al. Pleural Tags on CT scans to Predict Visceral Pleural invasion of non–small cell lung cancer That Does not abut the Pleura. Radiology. 279 (2), 590–596 (2016).
Gruden, J. F. What is the significance of pleural tags? AJR Am. J. Roentgenol. 164 (2), 503–504 (1995).
Gallagher, B. & Urbanski, S. J. The significance of pleural elastica invasion by lung carcinomas. Hum. Pathol. 21 (5), 512–517 (1990).
Ebara, K. et al. Pleural Invasion by Peripheral Lung Cancer: Prediction With Three Dimensional CT. Acad. Radiol. 22, 310–319 (2015).
Gillies, R. J., Kinahan, P. E. & Hricak, H. Radiomics: Images Are More Than Pictures, They Are Data. Radiology. 278, 563–577 (2016).
Wei, S. H. et al. The value of CT radiomics features to predict visceral pleural invasion in ≤ 3 cm peripheral type early non-small cell lung cancer. J. Xray Sci. Technol. 30 (6), 1115–1126 (2022).
Nelson, D. A. et al. Hypoxia and defective apoptosis drive genomic instability and tumorigenesis. Genes Dev. 18, 2095–2107 (2004).

Tables 1 to 5 are available in the Supplementary Files section.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Integrating Radiomics Features and CT Semantic Characteristics for Predicting Visceral Pleural Invasion in clinical stage Ia peripheral Lung Adenocarcinoma

Status:

Version 1

Abstract

Objectives

Methods

Results

Conclusion

Figures

Highlights

1. Introduction

2. Materials and methods

2.1. Patients

2.2. Evaluation of clinical stage Ia and VPI

2.3. CT scanning protocol

2.4. CT image interpretation and preprocessing

2.5. Tumor segmentation and features extraction

2.6. Feature selection and establishment of radiomics signature

2.7. Statistical analysis

3. Results

3.1. Inter-observer consistency analysis and patient demographics

3.2. Correlation of VPI with clinical and CT semantic features in training group

3.3. Screening and integration of radiomics features

3.4. Predictive model and external validation test

4. Discussion

Conclusion

Declarations

Author Contribution

Data Availability

References

Tables

Additional Declarations

Supplementary Files

Status:

Version 1