Development and validation of a nomogram based on CT texture analysis for discriminating minimally invasive adenocarcinoma from glandular precursor lesions in sub‑centimeter pulmonary ground glass nodules
- Authors:
- Published online on: November 17, 2023 https://doi.org/10.3892/ol.2023.14159
- Article Number: 26
-
Copyright: © Li et al. This is an open access article distributed under the terms of Creative Commons Attribution License.
Abstract
Introduction
The incidence and mortality rates of lung cancer surpass those of all other malignant tumors, accounting for ~20% of cancer-related deaths worldwide (1). Early diagnosis and treatment are pivotal in improving the prognosis and survival rates of lung cancer patients. In the 2021 WHO classification (2), adenocarcinoma in situ (AIS) has been reclassified as a glandular precursor lesion (GPL) along with adenomatous hyperplasia (AAH), while the subcategories of lung cancer now include minimally invasive adenocarcinoma (MIA) and invasive adenocarcinoma (IAC). This classification update necessitates corresponding adjustments to clinical diagnosis and treatment protocols. Currently, management follow-up strategies predominantly apply to AAH and AIS, whereas MIA warrants prompt surgical intervention (3,4). Furthermore, although some studies have indicated comparable long-term efficacy and 5-year survival rates between AAH, AIS and MIA surgeries, others have found that MIA exhibits higher Ki67 levels and EGFR mutation rates compared with AIS (5). Therefore, achieving precise discrimination of MIA from GPL during management follow-up and pre-surgical stages provides important insights for determining optimal clinical intervention timing, implementing surgical protocols and assessing prognosis.
Previous studies have demonstrated that high-resolution low-dose computed tomography (CT) is currently the most effective screening tool, reducing lung cancer mortality by 20% (6). Over 90% of these cases represent early-stage lung cancer, with pulmonary ground glass nodules (GGNs) being the primary manifestation on CT scans. While pathological biopsy serves as the gold standard for diagnosing lung cancer types, frozen section analysis has emerged as a valuable method for rapid intraoperative assessment of nodules, distinguishing between benign and malignant lesions and determining histological subtypes. This information plays a crucial role in guiding surgical strategies for lung nodules (7,8). However, due to the small size and low density of lung nodules, sampling often yields suboptimal results, resulting in a concordance rate of only ~68% between intraoperative frozen section analysis and postoperative paraffin pathology (8). Some researchers have explored the use of traditional CT imaging features (such as nodule size, density and solid component proportion) to differentiate between MIA and GPL (9,10). However, these features exhibit significant overlap across different nodule subtypes, leading to low diagnostic efficacy (11). Moreover, the extraction of these features heavily relies on the subjective interpretation skills and clinical experience of the radiologist (8). In addition, traditional techniques often struggle to identify nodules with smaller volumes and mixed densities. The increasing prevalence of GGNs with diameters <10 mm, detected through low-dose CT screening for early lung cancer, presents new challenges in clinical diagnosis. Some studies have proposed a critical value of 10 mm diameter for distinguishing between glandular precursor lesions and invasive lesions (9,10,12). Nevertheless, clinical practice has revealed numerous GGNs ≤10mm confirmed as MIA or IAC (10), leading to continuing debates about the management strategies for GGNs of this size (13). Consequently, there is an urgent need for innovative and accurate techniques to enable the precise preoperative diagnosis of MIA and GPL.
Radiomics is a powerful technique that enables the extraction and analysis of numerous radiomics features from medical imaging data with high efficiency. This approach holds significant promise in distinguishing between different pathological subcategories of lung nodules, assessing the extent of infiltration and evaluating prognostic outcomes (14). Texture analysis, on the other hand, involves the extraction and quantitative analysis of non-macroscopic and deep-level CT image features that reflect the tumor's heterogeneity to some extent. It has demonstrated utility in various aspects, including tumor differential diagnosis, prognostic evaluation, treatment response prediction and monitoring (15–17). CT texture analysis is widely utilized for identifying benign and malignant solitary lung nodules and evaluating invasiveness, exhibiting exceptional performance (18,19). To date, there have been no reports on the application of texture analysis for the identification of MIA and GPL in sub-centimeter GGNs. Therefore, the present study aimed to develop and validate a nomogram based on CT quantitative parameters and texture features for improving the ability to discriminate MIA from AAH/AIS, thereby providing important guidance for formulating clinical treatment plans and optimizing the timing of surgical interventions.
Materials and methods
Patient and nodule selection
The present retrospective study received approval from the ethics committee and written informed consent was waived. Patients with sub-centimeter GGNs who underwent high-resolution CT (HRCT) were enrolled at the First People's Hospital of Foshan between January 2019 and February 2022. The flowchart illustrating patient inclusion and exclusion is shown in Fig. 1, which was in accordance with previous studies (17,19,20). The inclusion criteria were as follows: i) GGNs with a maximum diameter ≤10 mm, evaluated using lung window settings [level: −600 Hounsfield units (HU); width: 1,500 HU]; ii) confirmation of AAH, AIS or MIA through surgical excision and pathology examination referring to the 2021 WHO classification (2); iii) availability of chest HRCT examination conducted within one month prior to surgical treatment; iv) absence of prior history of puncture, chemotherapy, or radiotherapy; v) absence of lung cancer or other malignant tumors. The exclusion criteria consisted of: i) inability to accurately delineate the region of interest due to the presence of artifacts or other lesions; ii) inaccurate identification or segmentation of CT images, including incomplete delineation or failure to avoid blood vessels, bronchi, pleura and other structures. The final selected cases were divided into two groups: 49 cases with AAH/AIS and 62 cases with MIA. The recruited patients were randomly assigned to a training set (78 nodules) and a validation set (33 nodules) at a ratio of 7:3.
CT image acquisition
The CT examinations for all patients included in the study were conducted using Philips CT scanners: the Philips Ingenuity 64-slice CT scanner (Philips Medical Systems, Inc.) and the Philips Brilliance iCT 256-slice CT scanner (Philips Ultrasound, Inc.). The patients were positioned in the supine (lying face-up) posture and the scanning range extended from the apex to the base of the lungs. Scans were performed at the end of a deep inspiration, followed by breath-holding to ensure stability. A tube voltage of 120 kV was used and the tube current employed automatic milliamp-second technology. The pitch value was set to 1.0, the collimation was 0.625×1.25 mm and the field of view (FOV) was set to 350×350 mm, with a pixel size of 512×512. The acquired images were reconstructed using both the standard algorithm and the high-resolution algorithm. The reconstructed slices had a thickness of 1 mm and a spacing of 1 mm.
CT image segmentation and feature extraction
The CT images in DICOM format were imported into the uAI-ChestCare software (version 0130; Shanghai United Imaging Healthcare Co., Ltd.), which was used for image segmentation and extraction of texture features (20,21). This software facilitated the automatic delineation of the complete 3D region of interest (ROI) for the identified lesions by outlining the tumor boundary on consecutive axial lung window images (with a window width of 1,500 HU and a window level of −600 HU). Subsequently, various quantitative and texture features were computed, encompassing maximum diameter (MD), solid volume (SV), solid volume rate (SVR), solid quality (SQ), solid quality rate (SQR), maximum computed tomography attenuation (CTmax), minimum computed tomography attenuation (CTmin), mean computed tomography attenuation (CTmean), median computed tomography attenuation (CTmedian), variance, kurtosis, skewness and entropy. To evaluate the accuracy of the software's automated nodule delineation, two radiologists with over 10 years of experience in thoracic imaging diagnosis independently assessed the results. They excluded cases where the nodule contour was incomplete or where blood vessels, bronchi, pleura and similar structures were not properly avoided. In instances where disagreements arose regarding the exclusion of certain cases, they negotiated to precisely delineate nodule boundaries while avoiding the inclusion of bronchi, large vessels, vacuole and normal tissue beyond the pleura in accordance with previous studies (22,23). A senior physician with 15 years of experience in thoracic imaging diagnosis verified the preceding two radiologists' segmentation results and provided the final confirmation of the results (19). The CT quantitative and texture features extraction of AIS (Fig. 2A) and MIA (Fig. 2B) were taken as examples.
Nomogram model building
The clinical, CT quantitative and texture features of both the training and validation groups were subjected to analysis using the Mann-Whitney U test to identify the effective distinguishing features. Subsequently, binary logistic regression analysis was employed to develop both a single-parameter model and a combined multi-parameter nomogram model. The diagnostic performance of these prediction models was assessed through the construction of receiver operating characteristic (ROC) curves (24) and the comparison of area under the curve (AUC) of ROC curves was performed using DeLong's test (25). Furthermore, the construction of the nomogram was accomplished using the rms package in the R software (version 4.0, R Foundation; http://www.Rproject.org). To evaluate the diagnostic performance of the nomogram prediction model, a calibration curve was employed.
Statistical analysis
The statistical analysis was conducted using SPSS version 20 software (IBM Corp.). Continuous variables were summarized as mean ± standard deviation (SD) or median with the full range. Categorical variables were presented as frequencies and percentages. Differences in age and sex between the two groups were assessed using the independent sample t-test and the χ2 test, respectively. To identify significant variables as predictive indicators of MIA the Mann-Whitney U test was applied based on the CT quantitative and texture features. Variables with a significance level of P<0.05 were selected as significant predictors for constructing both single-parameter and combined multi-parameter prediction models through binary logistic regression analysis. The diagnostic performance of each model was then evaluated by comparing the AUC, specificity and sensitivity of the ROC curve. AUC values greater than 0.5 were considered predictive. The significance of differences between ROC curves was determined using DeLong's test. Furthermore, the nomogram was constructed using the rms package in the R software. To verify the diagnostic performance of the nomogram prediction model, a calibration curve was employed. P<0.05 was considered to indicate a statistically significant difference.
Results
Patient characteristics
A total of 109 patients with 111 GGNs were ultimately included in the present study. The cohort consisted of 35 males and 74 females, with ages ranging 19–78 years. Among these patients, 49 GGNs in 47 patients were classified as GPL (AAH/AIS), with a mean age of (54.26±12.24) years. Additionally, 62 GGNs in 62 patients were categorized as MIA, with a mean age of (52.84±11.47) years. There were no significant differences observed in terms of sex (P=0.564) or age (P=0.522) between the two groups. Utilizing stratified random sampling, the training set consisted of 78 cases (34 AAH/AIS and 44 MIA), while the validation set included 33 cases (15 AAH/AIS and 18 MIA).
CT quantitative parameters and texture feature extraction
The results of patient characteristics, CT quantitative parameters and texture features were conducted using the Mann-Whitney U test and the results are presented in Table I. Notably, significant statistical differences (P<0.05) were observed between the two groups in seven parameters, including SV, SVR, SQ, SQR, CTmax, CTmean and entropy.
Table I.Clinical characteristics, CT quantitative parameters and texture features between GPL and MIA groups. |
Model construction and diagnostic validation based on CT quantitative parameters and texture features
The diagnostic models were constructed and the corresponding ROC curves are displayed in Fig. 3. Notably, the combined multi-parameter model exhibited superior predictive ability compared with each individual single-parameter model. The performance of the nomogram prediction model is illustrated in Table II for the training set and Table III for the validation set. In the training set, the nomogram achieved an AUC of 0.992 (95% CI: 0.980–1.000), a sensitivity of 0.907, a specificity of 1.000 and an accuracy of 0.948. For the validation set, the AUC was 0.975 (95% CI: 0.935–1.000), with a sensitivity of 0.842, a specificity of 0.941 and an accuracy of 0.912. The results of DeLong's test demonstrated significant statistical differences in AUCs between the nomogram and single-parameter models (P<0.001).
Table II.The AUC values for each prediction model were calculated and evaluated on both the training and validation set. |
Nomogram and calibration curve
A nomogram (Fig. 4A) was constructed utilizing five parameters (SV, SVR, CTmax, CTmean and entropy). The model formula was derived as follows: Total Points=0.576 × SV −1.131 × SVR - 0.162× CTmax - 0.136 × CTmean + 21.533 × entropy. Each parameter in the nomogram corresponds to a specific score on the top ‘Points’ axis. The cumulative sum of these scores corresponds to the values displayed on the bottom ‘Total Points’ axis, representing the diagnostic probability of MIA: Two cases of AIS and MIA (Fig. 4B and C) provided illustrative examples from the collected database. The calibration curve (Fig. 5) demonstrates a favorable agreement between the predicted model and the observed data, as the scatter plot closely aligns with the ideal curve.
Discussion
The present study retrospectively collected surgically resected and pathologically confirmed cases of GPL and MIA that presented as sub-centimeter GGNs on CT images, then extracted and analyzed their CT quantitative and texture features. Subsequently, a nomogram model incorporating the five most informative identification indicators (SV, SVR, CTmax, CTmean and entropy) was developed to accurately differentiate MIA from GPL. The ROC curve revealed high predictive accuracy in distinguishing between MIA and GPL, with an AUC values of 0.992 and 0.975 for the training set and validation set, respectively. Early-stage lung cancers predominantly present as solitary GGNs, posing challenges in distinguishing benign nodules from malignant ones due to the diverse manifestations and overlapping features observed on CT images. In line with previous studies by Wu et al (9,10), who extensively analyzed the CT imaging features of sub-centimeter pure GGNs, the present study identified relevant morphological features such as lesion size, vessel changes and tumor-lung interface, which reflect the invasiveness of GGNs. Moreover, previous investigations (26–28) have established a close association between the size of the solid component and the average CT attenuation of GGNs with their invasiveness and pathology. The findings of the present study corroborated these conclusions, demonstrating significant differences in SV, SVR, SQ, SQR, CTmax and CTmean between GPL and MIA. Notably, the values of these variables were consistently higher in the MIA group compared with the GPL group. These results confirm the importance of the solid component in predicting the behavior of GGNs. To summarize, the present study retrospectively analyzed surgically resected GGNs, extracting and analyzing their CT quantitative and texture features and achieved high predictive performance in discriminating between MIA and GPL through the development of a nomogram model incorporating five informative identification indicators. The challenges associated with differentiating benign and malignant nodules in early-stage lung cancers were underscored, emphasizing the significance of considering the presence and characteristics of the solid component in predicting invasiveness and pathological behavior of GGNs.
Texture analysis plays a crucial role in efficiently and accurately extracting biological information that reflects tumor heterogeneity, which may not be discernible by visual examination of images alone. It enables quantitative assessment of subtle changes in image pixel values and their arrangement. As a result, it holds great value in qualitative diagnosis, invasiveness assessment, prognostic prediction of tumors and informed clinical management. Qiu et al (17) demonstrated the independent prognostic significance of mean CT attenuation and entropy in evaluating the invasiveness of 428 cases of clinical stage IA lung adenocarcinoma. Zhu et al (19) employed the ANOVA test and the least absolute shrinkage and selection operator algorithm to identify 18 CT texture features, including entropy. They successfully developed a diagnostic model capable of distinguishing MIA from GPL presenting as pure GGNs, achieving high identification performance with an AUC of 0.884 in the training set and 0.872 in the validation set. In the present study, there was a statistically significant difference in entropy between GPL and MIA, which is consistent with previous findings. Moreover, the combined multi-parameter model exhibited superior predictive performance, with AUC values of 0.992 in the training set and 0.975 in the validation set, surpassing the research conducted by Zhu et al (19). This disparity may be attributed to differences in the composition of enrolled cases. The present study encompassed not only pure vitreous nodules but also mixed-density nodules. Previous studies (26,28) have indicated the diagnostic relevance of the solid component within GGNs. Another study (29) employed machine learning and deep learning methods to evaluate benign and malignant pulmonary nodules, revealing AUC values of 0.763 for the support vector machine (SVM) model and 0.723 for the convolutional neural networks (CNN) model in distinguishing GPL from MIA. Notably, the predictive performance of both SVM and CNN models was lower than that of the nomogram model established in the present study. These findings suggest that CT texture analysis may offer greater assistance and efficiency in achieving this objective. The patients in the present study were sourced from a single center, resulting in a relatively small sample size, but the parameters and quality of these CT images were highly standardized and homogenized. Conversely, in a previous multicenter study (29), although the sample size was larger, challenges related to CT image homogenization may have persisted. This discrepancy could explain the superior diagnostic effectiveness of the constructed model. Consequently, future studies should focus on expanding the sample size and analyzing data from different centers to obtain a more accurate and efficient predictive model.
Nomogram models are graphical representations employed to illustrate analytical outcomes derived from multifactorial logistic regression models or Cox proportional hazards models. These models use a set of parallel, non-intersecting lines within a coordinate plane to portray the quantitative analysis diagram, depicting the functional relationship between multiple variables. By employing intuitive symbols, nomograms facilitate the calculation of disease occurrence probability, recurrence risk and prognosis. Consequently, they find extensive application in clinical research pertaining to pulmonary GGNs (30). In the present study, a nomogram model for the identification of precursor lesions and MIA was established, based on five CT quantitative and texture features, namely SV, SVR, CTmax, CTmean and entropy. This model serves as an effective and intuitive reference standard, enabling radiologists to achieve accurate and prompt diagnoses.
In conclusion, the use of CT quantitative and texture features offered significant utility in the differentiation of MIA from GPL. The nomogram model developed demonstrated superior discriminatory capabilities. This model's diagnostic efficacy, combined with its graphical representation, facilitates the precise classification of GGN types by radiologists, thereby aiding clinicians in making informed decisions regarding treatment and follow-up strategies for GGNs. Nonetheless, the present study has certain limitations that require acknowledgment. It was a retrospective, single-center study with a small sample size, which may introduce selection bias and potentially affect the precision and applicability of the model. In the future, efforts should be made to enhance the model's robustness and applicability by increasing the sample size, conducting multi-center studies and incorporating external test sets.
Acknowledgements
Not applicable.
Funding
The present study was supported by Project of Foshan Science and Technology Bureau (grant no. 2220001003972), the Science Innovative Project of Foshan (grant no. FSOAA-KJ218-1301-0021), the Foshan 14th Five-Year Plan Key Discipline Foundation (grant no. FSGSP145036), the Medical Research Subject of Foshan Health Bureau (grant no. 20230027), the Medical Research Foundation of Guangdong Province (grant nos. A2021493 and A2022330) and the Natural Science Foundation of Guangdong Province (grant no. 2021A1515220032).
Availability of data and materials
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.
Authors' contributions
CL and ZX participated in the design of the study. YJ and QD performed data analysis and prepared the figures. YY, RD and JZ participated in the analysis of the figures and data. CL and YJ prepared and revised the manuscript. AP, MG and ZX reviewed the results and revised the manuscript. CL, AP, MG and ZX confirm the authenticity of all the raw data. All authors have read and approved the final version of this manuscript.
Ethics approval and consent to participate
The present study was approved by the Institutional Research Ethics Committee of the First Peoples' Hospital of Foshan (approval no. 2021.02) and was conducted in accordance with the ethical principles of the Declaration of Helsinki and the ethical laws and regulations established in China. Written informed consent was waived.
Patient consent for publication
Not applicable.
Use of artificial intelligence tools
The uAI-ChestCare software is an artificial intelligence (AI) software based on deep learning method, specifically designed for applications related to chest or pulmonary care and have achieved good results in prior research endeavors. In the present study, CT lung window images, configured with a window width of 1,500 Hounsfield Units (HU) and a window level of −600 HU in DICOM format, were input into the uAI-ChestCare software. Subsequently, the software autonomously delineated the entire 3D ROI for identified lesions by outlining the tumor boundaries across consecutive axial slices. Following this, an array of quantitative and textural features were computed, encompassing parameters such as MD, SV, SVR, SQ, SQR, maximum computed tomography attenuation CTmax, CTmin, CTmean, CTmedian), variance, kurtosis, skewness and entropy.
Competing interests
The authors declare that they have no competing interests.
References
Hyldgaard C, Trolle C, Harders SMW, Engberg H, Rasmussen TR and Møller H: Increased use of diagnostic CT imaging increases the detection of stage IA lung cancer: Pathways and patient characteristics. BMC Cancer. 22:4642022. View Article : Google Scholar : PubMed/NCBI | |
Walter JE, Heuvelmans MA, Bock GH, Yousaf-Khan U, Groen HJM, Aalst CMV, Nackaerts K, Ooijen PMAV, Koning HJ, Vliegenthart R and Oudkerk M: Characteristics of new solid nodules detected in incidence screening rounds of low-dose CT lung cancer screening: The NELSON study. Thorax. 73:741–747. 2018. View Article : Google Scholar : PubMed/NCBI | |
Tammemagi MC and Lam S: Screening for lung cancer using low dose computed tomography. BMJ. 348:g22532014. View Article : Google Scholar : PubMed/NCBI | |
Nicholson AG, Tsao MS, Beasley MB, Borczuk AC, Brambilla E, Cooper WA, Dacic S, Jain D, Kerr KM, Lantuejoul S, et al: The 2021 WHO classification of lung tumors: Impact of advances since 2015. J Thorac Oncol. 17:362–387. 2022. View Article : Google Scholar : PubMed/NCBI | |
Lantuejoul S, Rouquette I, Brambilla E and Travis WD: New WHO classification of lung adenocarcinoma and preneoplasia. Ann Pathol. 36:5–14. 2016.(In French). View Article : Google Scholar : PubMed/NCBI | |
Yankelevitz DF, Yip R, Smith JP, Liang M, Liu Y, Xu DM, Salvatore MM, Wolf AS, Flores RM and Henschke CI; International Early Lung Cancer Action Program Investigators Group, : CT screening for lung cancer: Nonsolid nodules in baseline and annual repeat rounds. Radiology. 277:555–564. 2015. View Article : Google Scholar : PubMed/NCBI | |
Nie X, Li L, Huang J, Zhang P, Shi H, Cheng G and Zhang YQ: From focal pulmonary pure ground-glass opacity nodule detected by low-dose computed tomography into invasive lung adenocarcinoma: A growth pattern analysis in the elderly. Thorac Cancer. 9:1361–1365. 2018. View Article : Google Scholar : PubMed/NCBI | |
Yamagami T, Yoshimatsu R, Miura H, Yamada K, Takahata A, Matsumoto T and Hasebe T: Diagnostic performance of percutaneous lung biopsy using automated biopsy needles under CT-fluoroscopic guidance for ground-glass opacity lesions. Br J Radiol. 86:201204472013. View Article : Google Scholar : PubMed/NCBI | |
Wu F, Tian SP, Jin X, Jing R, Yang YQ, Jin M and Zhao SH: CT and histopathologic characteristics of lung adenocarcinoma with pure ground-glass nodules 10 mm or less in diameter. Eur Radiol. 27:4037–4043. 2017. View Article : Google Scholar : PubMed/NCBI | |
Wu F, Cai ZL, Tian SP, Jin X, Jing R, Yang YQ, Li Y and Zhao SH: Analysis of histopathologic subtypes and CT characteristics of lung adenocarcinomas presenting as pure ground-glass nodule of 1 cm or less in maximal diameter. Chin J Radiol. 50:260–264. 2016. | |
Jia M, Yu S, Cao L, Sun PL and Gao H: Clinicopathologic Features and genetic alterations in adenocarcinoma in situ and minimally invasive adenocarcinoma of the lung: Long-term follow-up study of 121 Asian patients. Ann Surg Oncol. 27:3052–3063. 2020. View Article : Google Scholar : PubMed/NCBI | |
Zhao W, Xu Y, Yang Z, Sun Y, Li C, Jin L, Gao P, He W, Wang P, Shi H, et al: Development and validation of a radiomics nomogram for identifying invasiveness of pulmonary adenocarcinomas appearing as subcentimeter ground-glass opacity nodules. Eur J Radiol. 112:161–168. 2019. View Article : Google Scholar : PubMed/NCBI | |
Chae HD, Park CM, Park SJ, Lee SM, Kim KG and Goo JM: Computerized texture analysis of persistent part-solid ground-glass nodules: Differentiation of preinvasive lesions from invasive pulmonary adenocarcinomas. Radiology. 273:285–293. 2014. View Article : Google Scholar : PubMed/NCBI | |
Raman SP, Schroeder JL, Huang P, Chen Y, Coquia SF, Kawamoto S and Fishman EK: Preliminary data using computed tomography texture analysis for the classification of hypervascular liver lesions: Generation of a predictive model on the basis of quantitative spatial frequency measurements-a work in progress. J Comput Assist Tomogr. 39:383–395. 2015.PubMed/NCBI | |
Ng F, Ganeshan B, Kozarski R, Miles KA and Goh V: Assessment of primary colorectal cancer heterogeneity by using whole-tumor texture analysis: Contrast-enhanced CT texture as a biomarker of 5-year survival. Radiology. 266:177–184. 2013. View Article : Google Scholar : PubMed/NCBI | |
Jiang Y, Che S, Ma S, Liu X, Guo Y, Liu A, Li G and Li Z: Radiomic signature based on CT imaging to distinguish invasive adenocarcinoma from minimally invasive adenocarcinoma in pure ground-glass nodules with pleural contact. Cancer Imaging. 21:12021. View Article : Google Scholar : PubMed/NCBI | |
Qiu ZB, Zhang C, Chu XP, Cai FY, Yang XN, Wu YL and Zhong WZ: Quantifying invasiveness of clinical stage IA lung adenocarcinoma with computed tomography texture features. J Thorac Cardiovasc Surg. 163:805–815. 2022. View Article : Google Scholar : PubMed/NCBI | |
Moon Y, Sung SW, Lee KY, Sim SB and Park JK: Pure ground-glass opacity on chest computed tomography: Predictive factors for invasive adenocarcinoma. J Thorac Dis. 8:1561–1570. 2016. View Article : Google Scholar : PubMed/NCBI | |
Zhu YQ, Liu C, Mo Y, Dong H, Huang C, Duan YN, Tang LL, Chu YY and Qin J: Radiomics for differentiating minimally invasive adenocarcinoma from precursor lesions in pure ground-glass opacities on chest computed tomography. Br J Radiol. 95:202107682022. View Article : Google Scholar : PubMed/NCBI | |
Liu J, Yang X, Li Y, Xu H, He C, Qing H, Ren J and Zhou P: Development and validation of qualitative and quantitative models to predict invasiveness of lung adenocarcinomas manifesting as pure ground-glass nodules based on low-dose computed tomography during lung cancer screening. Quant Imaging Med Surg. 12:2917–2931. 2022. View Article : Google Scholar : PubMed/NCBI | |
Xie Y, Zhang J and Xia Y: Semi-supervised adversarial model for benign-malignant lung nodule classification on chest CT. Med Image Anal. 57:237–248. 2019. View Article : Google Scholar : PubMed/NCBI | |
Wu L, Gao C, Xiang P, Zheng S, Pang P and Xu M: CT-imaging based analysis of invasive lung adenocarcinoma presenting as ground glass nodules using peri- and intra-nodular radiomic features. Front Oncol. 10:8382020. View Article : Google Scholar : PubMed/NCBI | |
Xu F, Zhu W, Shen Y, Wang J, Xu R, Outesh C, Song L, Gan Y, Pu C and Hu H: Radiomic-based quantitative CT analysis of pure ground-glass nodules to predict the invasiveness of lung adenocarcinoma. Front Oncol. 10:8722020. View Article : Google Scholar : PubMed/NCBI | |
Mishra P: Practical explainable AI using python: Artificial intelligence model explanations using pythonbased libraries, extensions, and frameworks. Apress Media; New York, NY: 2021 | |
DeLong ER, DeLong DM and Clarke-Pearson DL: Comparing the areas under two or more correlated receiver operating characteristic curves: A nonparametric approach. Biometrics. 44:837–845. 1988. View Article : Google Scholar : PubMed/NCBI | |
Lee KH, Goo JM, Park SJ, Wi JY, Chung DH, Go H, Park HS, Park CM and Lee SM: 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. 9:74–82. 2014. View Article : Google Scholar : PubMed/NCBI | |
Fang W, Zhang G, Yu Y, Chen H and Liu H: Identification of pathological subtypes of early lung adenocarcinoma based on artificial intelligence parameters and CT signs. Biosci Rep. 42:BSR202124162022. View Article : Google Scholar : PubMed/NCBI | |
Wang X, Wang L, Zhang W, Zhao H and Li F: Can we differentiate minimally invasive adenocarcinoma and non-invasive neoplasms based on high-resolution computed tomography features of pure ground glass nodules? PLoS One. 12:e1805022017. | |
Ashraf SF, Yin K, Meng CX, Wang Q, Wang Q, Pu J and Dhupar R: Predicting benign, preinvasive, and invasive lung nodules on computed tomography scans using machine learning. J Thorac Cardiovasc Surg. 163:1496–1505.e10. 2022. View Article : Google Scholar : PubMed/NCBI | |
Meng F, Guo Y, Li M, Lu X, Wang S, Zhang L and Zhang H: Radiomics nomogram: A noninvasive tool for preoperative evaluation of the invasiveness of pulmonary adenocarcinomas manifesting as ground-glass nodules. Transl Oncol. 14:1009362021. View Article : Google Scholar : PubMed/NCBI |