Unexpectedly Benign Pulmonary Nodules Using Machine Learning

Unexpectedly Benign Pulmonary Nodules Using Machine Learning | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Unexpectedly Benign Pulmonary Nodules Using Machine Learning Yanrong Meng, Fengsheng Wang, Dan Liu, Jing Wang, Yan Wang, Jilai Zhang, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5361749/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background: The widespread application of computed tomography (CT) for pulmonary disease screening has led to an increased detection of pulmonary nodules. However, this has also resulted in a high false-positive rate for suspected malignancies that ultimately prove to be benign. It is crucial to explore the clinical and imaging characteristics of these patients to avoid unnecessary major pulmonary resections. This study aims to evaluate the characteristics of surgically resected benign nodules that were initially presumed to be lung cancers based on lung CT scans. Materials and Methods: This retrospective study analyzed 203 cases of benign lung nodules at the First Medical Centre of the Chinese People’s Liberation Army (PLA) General Hospital from January 2017 to June 2023. Pathological examination following surgical resection confirmed pulmonary granulomatous inflammation in these cases. The study cohort was divided into two groups: 86 patients with benign nodules and 117 with malignant nodules, all diagnosed based on imaging features. Given the overlapping imaging features of benign and malignant nodules, the clinical and imaging characteristics of both groups were compared to reduce the incidence of unnecessary surgeries. Various machine learning models, including Random Forest, SVM linear, SVM nonlinear, Logistic Regression, XGBoost, and k-Nearest Neighbor, were constructed. With the optimal model selected based on performance on a validation set, A framework was developed to identify personalized risk factors using a feature importance ranking algorithm. Results: Analysis of data from 203 patients revealed significant differences in maximum lesion size (pathological diagnosis), nodule density, boundary, calcification, satellite nodules, vascular aggregation sign, vacuole sign, spiculation sign, lobulation sign, pleural indentation sign, and nodule location (p<0.05). The SVMlinear model, which achieved the highest AUC (0.867), was selected for the final predictive model.Using recursive feature elimination method and manual feature selection, the feature of lesion size, lobulation sign, vacuole sing and enlarged lymph nodes have been screened in predicting and distinguishing the nature of nodules through imaging. Conclusion: Benign pulmonary nodules that are unexpectedly resected often present with features typically associated with malignancy, such as larger volumes, lobulation, vacuole signs, and enlarged lymph nodes. This study highlights the importance of accurately distinguishing between benign and malignant nodules to minimize unnecessary surgical interventions. unexpectedly benign pulmonary nodules lobulated sign vacuole sign imaging features machine learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 1. Introduction With the growing emphasis on health consciousness and the widespread use of high-resolution computed tomography (HRCT), early detection and resection of malignant pulmonary nodules hold the potential to improve lung cancer outcomes. Numerous studies support this conclusion, showing that CT screening not only delays lung cancer mortality by more than a decade but also prevents deaths altogether [ 1 – 2 ]. However, there are ongoing concerns that need to be addressed, including the potential drawbacks of overdiagnosis and false-positive lung cancer diagnoses. Surgical resections of benign nodules mistakenly presumed to be malignant are not uncommon, given the overlapping imaging features of benign and malignant nodules. Differentiating between these two types of nodules remains a significant challenge for radiologists and oncologists. Despite the overlapping characteristics, the importance of morphological features should not be underestimated. Numerous studies have indicated that the appearance of nodule density, vascular aggregation, vacuole sign, spiculation, and lobulation are associated with a higher likelihood of malignancy, whereas features like calcification and satellite nodules are more commonly linked to benignity [ 3 – 5 ]. Sometimes, the effects of lung CT scanning on evaluation results of pulmonary nodules are limited. Additional techniques, such as 3D reconstruction CT or 18F-fluorodeoxyglucose (18F-FDG) positron emission tomography (PET-CT), may offer added value. For the latter, the tracer is 18F-FDG, which is an 18F-fluorolabeled glucose analogue. However, it is noteworthy that subcentimeter nodules, carcinoma in situ, or adenocarcinoma precursors can exhibit low or no uptake. Similarly, the glucose metabolism of both tumor and inflammatory graunloma can be elevated, potentially leading to false-positive results in cases of inflammatory granuloma [ 6 ]. Therefore, it is crucial to weigh these factors carefully when distinguishing between malignant and benign nodules in clinical practice. Research on surgically resected benign nodules initially considered malignant is scarce in the existing literature. For example, only 10 patients were investigated in Williams’ study and 19 of 148(12.8%) underwent surgical resection had a benign diagnosis in Archer’s research’s [ 7 – 8 ]. The aim of our study is to assess and compare the characteristics of surgically resected benign nodules detected on CT scans that were initially presumed to be malignant. This research seeks to identify unexpectedly benign pulmonary nodules by comparing imaging-based risk factors for nodules classified as benign versus those deemed malignant. Baseline comparisons were made, followed by an assessment of the significance of the differing risk factors. The optimal model was selected based on the area under the curve (AUC) from various machine learning models. This model aims to help identify the imaging characteristics of unexpectedly benign pulmonary nodules, thereby reducing unnecessary surgeries. 2. Materials and methods 2.1 Study population and design In this study, we reviewed cases of patients who underwent pulmonary nodule resection at the First Medical Centre of the Chinese People’s Liberation Army (PLA) General Hospital. A radiologist specializing in lung imaging and two physicians independently collected demographic data, clinical characteristics, imaging features, and pathological diagnoses of the resected nodules from electronic medical records. Personal information was kept confidential in this retrospective study. The Ethics Committee of the First Medical Centre of the Chinese PLA General Hospital approved the study and the requirement for informed patient consent was waived. 2.2 Inclusion and exclusion criteria Inclusion criteria were applied： 1). All pulmonary nodules were detected from January 2017 to July 2023, 2). All patients >18 years old. 3). Patients surgically resected with a presumed pulmonary granulomatous lesion. Exclusion criteria were as follows： 1). Patients with suspected metastatic disease based on chest CT, 2). Patients with malignant lung nodules diagnosed by histopathological diagnosis. 3). Patients with unconfirmed postoperative pathological tests data incomplete clinical data. 3. Statistical analysis Demographic characteristics were presented as frequencies and percentages for categorical variables and as means with standard deviations for continuous variables (assuming normal distribution). Differences between the two groups’ continuous variables were compared using either a t-test or the Mann–Whitney U test, while categorical data were analyzed using chi-square analysis. P values were calculated using two-sided exact tests, with values less than 0.05 considered statistically significant. All analyses were performed using SPSS (Statistical Program for Social Sciences) software, version 26. Additionally, multiple machine learning models were constructed, including Random Forest, SVM linear, SVM nonlinear, Logistic Regression, XgBoost, and k-Nearest Neighbors, with the optimal model selected based on performance on the validation set. Nodule location was represented using one-hot encoding, and other data were converted to binary values (0 and 1). The entire dataset was divided into a training set (162 samples) and a validation set (41 samples). The model construction details are as follows: 1). Random Forest (RF): The training set was used to identify optimal hyperparameters through grid search. The number of trees in the random forest was set to 500. 2). Support Vector Machine Linear (SVM linear): A linear kernel function was chosen for classification. Optimal hyperparameters were determined using grid search on the training set, with a penalty coefficient of 1. 3). Support Vector Machine Nonlinear (SVM nonlinear): A Gaussian radial basis function kernel was selected for nonlinear classification. Optimal hyperparameters were identified through grid search on the training set, with a penalty coefficient of 0.01 and a gamma coefficient of 0.1. 4). Logistic Regression (Log): Optimal hyperparameters were identified using grid search on the training set. The internal gradient descent algorithm was set to converge with a maximum of 1,000 iterations, and a penalty coefficient of 1. 5). XgBoost (XGB): The training set was used to identify optimal hyperparameters through grid search. The model was set to 100 trees, with a learning rate of 0.1 and a subsample ratio per column of 0.6. 6). k-Nearest Neighbors (KNN): Utilizing Euclidean distance, optimal hyperparameters were identified through grid search on the training set, with the number of neighbors set to 10. The predictive efficiency of these models was evaluated using AUC curves on the test set. 4. Results 4.1 Basic characteristics A total of 203 patients underwent more than one chest CT screening. Of these, 86 were diagnosed with benign nodules based on diagnostic imaging tests (both CT and PET-CT), with 11 of these confirmed as benign on PET-CT. The remaining 117 patients were diagnosed with malignant nodules through chest CT or PET-CT scans. Among them, 88 were identified as malignant on CT. PET-CT further refined the diagnosis for 67 of the 117 patients, including 19 who were initially diagnosed as benign on CT and 10 whose nodules could not be definitively classified as benign or malignant by CT alone. PET-CT results showed that 57 patients were malignant, 7 were benign, and 3 remained inconclusive. Notably, both CT and PET-CT indicated malignancy in 28 patients. Overall, the pathological diagnosis for all 203 patients with primary pulmonary nodules was pulmonary granulomas. The following table provides details on the patients’ characteristics and the imaging features of the nodules. Significant differences were observed between the two groups in maximum lesion size (pathological diagnosis), nodule density, boundary, calcification, satellite nodules, vascular aggregation sign, vacuole sign, spiculation sign, lobulation sign, pleural indentation sign, and nodule location (p<0.05). However, no significant differences were found in age, gender, smoking status, maximum lesion size (imaging diagnosis), nodule shape, proximity to the interlobar fissure, proximity to the pleura, bronchial cut-off sign, lymph node enlargement, or nodule count between the two groups (p>0.05). Table 1 Clinical and imaging features of patients who underwent resection for benign nodule Clinical Factor Benign nodule diagnosed by imaging (n=86) Malignant nodule diagnosed by imaging (n=117) P-value No. % No. % Median age (range) 54.4±10.0 52.2±9.2 Age >=55 years <55 years 36 50 41.86 58.14 61 56 52.14 47.86 0.107 Gender Male Female 55 31 63.95 36.05 65 52 55.56 44.44 0.229 Smoking status Current or past smoker Nonsmoker 37 49 43.02 56.98 40 77 34.19 65.81 0.200 Maximum lesion size, (cm) (imaging diagnosis) >1 1 <=1 63 23 73.26 26.74 79 38 67.52 32.48 <0.001 Nodule density Solid Subsolid (ground-glass, part-solid, etc) 78 8 90.70 9.30 93 24 79.49 20.51 0.030 Boundary Clear Blur 49 37 56.98 43.02 52 65 44.44 55.56 0.024 Shape of node Round Irregular 26 60 30.23 69.77 26 91 22.22 77.78 0.196 Calcification Yes No 25 61 29.07 70.93 14 103 11.97 88.03 0.002 Satellite nodule(s) Yes No 23 63 26.74 73.26 18 99 15.38 84.62 0.046 Vascular aggregation sign Yes No 8 78 9.30 90.70 25 92 21.37 78.63 0.021 Adjacent to interlobar fissure Yes No 29 57 33.72 66.28 40 77 34.19 65.81 0.945 Adjacent to pleura Yes No 71 15 82,56 17.44 89 28 76.07 23.93 0.263 Bronchial cut-off sign Yes No 4 82 4.65 95.35 11 106 9.40 90.60 0.201 Lymph nodes enlarge Yes No 1 85 1.16 98.94 8 109 6.84 93.16 0.052 Vacuole sign Yes No 2 84 2.33 97.67 15 102 12.82 87.18 0.008 Spiculation sign Yes No 18 68 20.93 79,07 51 66 43.59 56.41 <0.001 Lobulation sign Yes No 15 71 17.44 82.56 61 56 52.14 47.86 <0.001 Nodule count Solitary Multiple 66 20 76.74 23.26 96 21 82.05 17.95 0.352 Pleural indentation sign Yes No 57 29 66.28 33.72 70 47 59.83 40.17 0.023 Nodule location Upper lobe of left lung Lower lobe of left lung Upper lobe of right lung Middle lobe of right lung Lower lobe of right lung 15 5 29 17 20 17.44 5.81 33.72 19.77 23.26 20 22 34 12 29 17.09 18.8 29.06 10.26 24.78 0.043 Fig.1 The AUC curve area of six algorithm models 4.2 Comparison of the performance of multiple machine learning methods in identifying malignant or benign nodules In this study, we retrospectively analyzed 203 patients with pulmonary nodules, comprising 86 benign and 117 malignant cases based on imaging procedures. The patients were divided into a training set (n=162) and a validation set (n=41) at a 4:1 ratio. The key reference characteristics for differentiating between benign and malignant pulmonary nodules included age, gender, smoking status, maximum lesion size (as determined by imaging and pathological diagnosis), nodule density, boundary, shape, calcification, satellite nodules, vascular aggregation sign, proximity to the interlobar fissure and pleura, bronchial cut-off sign, lymph node enlargement, vacuole sign, spiculation sign, lobulation sign, nodule count, pleural indentation sign, and nodule location. The performance of several machine learning methods was evaluated, including Random Forest, SVM linear, SVM nonlinear, logistic regression, XgBoost, and k-Nearest Neighbor, on the validation set. The AUC curves of these models were constructed to assess their predictive efficiency on the training set. Among the models, the SVM linear model demonstrated the highest AUC of 0.867, significantly outperforming the others. To maximize the identification of risk factors between malignant and benign nodules based on imaging, the SVM linear model was selected for further development of the predictive model (Fig. 1). Fig.2 the feature prediction of REF method Fig.3 The AUC curve area of reintroduced into the support vector machine model 4.3 Feature selection for constructing the final classifier The Recursive Feature Elimination (RFE) method was employed to validate the accuracy of feature predictions, revealing that error values were minimized when using 23 features during ten-fold cross-validation (Fig. 2). These 23 selected features were then reintroduced into the support vector machine model for a comparative analysis of predictive performance. This analysis showed an increase in the AUC for predictive efficiency, from 0.867 to 0.875 (Fig. 3). The selected features were ranked according to their importance, resulting in a distribution of feature relevance depicted in Figure 4. The lesion size measured by pathological diagnosis emerged as the most significant predictor in differentiating between benign and malignant pulmonary nodules. The top nine factors influencing this clinical diagnosis included lesion size by pathological diagnosis, bronchial cut-off sign, lobulation sign, vacuole sign, lesion size by CT, nodule type, lymph node enlargement, proximity to the interlobar fissure, and nodule location in the upper lobe of the left lung. Fig.4 The selected features were ranked according to their importance by REF method Finally, we eliminated features with collinearity and retained only those with strong relevance to the target outcome (nodule nature) through manual feature selection. Given that the data consisted of binary categories (0 and 1), the Pearson correlation coefficient was unsuitable for calculating correlations between features. Instead, Cramér’s V coefficient was used, which is specifically designed to measure the degree of association between categorical variables (Fig.5). This coefficient ranges from 0 to 1, where 0 indicates no association and 1 indicates complete correlation. The Cramér’s V was calculated coefficient between features, identifying those with a Cramér’s V greater than 0.3 and a p-value less than 0.05 as collinear and those with a Cramér’s V greater than 0.1 or a p-value less than 0.05 as strongly relevant to the target outcome. Ultimately, the features were retained strongly related to the target outcome and removed those that were collinear (Fig. 6). Fig.5 the degree of association between features Fig.6 Coefficient plot of correlation between features and outcome of benign and malignant nodules Through manual selection, we observed an increase in predictive efficacy, with the AUC rising from 0.867 to 0.885 (Fig. 7). The selected features were ranked by importance, resulting in a distribution of feature relevance as shown in Figure 8. Nine key features were identified ultimately: lesion size by pathological diagnosis, lobulation sign, tumor-lung interface, calcification, satellite nodules, spiculation sign, smoking history, vacuole sign, and lymph node enlargement. Fig.7 The AUC curve area of reintroduced into the support vector machine model after manual selection Fig.8 The selected features were ranked according to their importance by manual selection 5. Discussion The management of pulmonary nodules primarily focuses on distinguishing between benign and malignant nodules, aiming to balance the need for prompt intervention for malignant cases against avoiding unnecessary surgeries for benign ones. Unlike biopsy and surgical resection, CT as a non-invasive procedure is commonly used as a diagnostic tool for pre-surgical evaluation [ 9 ]. Radiologists typically estimate the probability of lung cancer based on the CT characteristics of the nodules. In our study, larger lesion size and the presence of subsolid nodules indicated a higher probability of malignancy. Nodule margin is another critical feature in assessing malignancy risk. Malignant nodules often exhibit signs of spiculation, lobulation, and vascular aggregation, while benign nodules are more likely to display calcification and satellite nodules. Additionally, the presence of a vacuole sign, pleural indentation, and hilar or mediastinal lymph nodes enlarged to more than 1.0 cm in the shortest diameter were more likely to suggest malignancy (Table 1 ). In our study, only 38.5% (78/203) of patients with pulmonary nodules underwent PET-CT before surgery. Notably, pulmonary nodules that were unexpectedly benign often displayed malignant characteristics on lung CT or PET-CT scans before surgical resection. To further analyze the proportion of factors distinguishing imaging-diagnosed benign pulmonary nodules from unexpectedly benign nodules, multiple machine learning models were employed to differentiate between benign and malignant nodules. The SVM linear model performed exceptionally well in predictions. After that, feature selection was performed on distinguishing benign nodules and unexpectedly benign pulmonary Nodules using SVM-RFE and SVM-manual selection to select several optimal feature subsets [ 11 ]. The RFE method selected 23 features, and rebuilding the SVM model with these improved the predictive efficacy from 0.867 to 0.875. Through manual selection, nine key features were identified and used to rebuild the SVM model, resulting in an increase in predictive efficacy from 0.867 to 0.885. Both feature selection methods enhanced the model’s predictive performance [ 10 ]. The nine features identified through manual selection were lesion size by pathological diagnosis, lobulation sign, tumor-lung interface, calcification, satellite nodules, spiculation sign, smoking history, vacuole sign, and enlarged lymph nodes. The RFE method identified the top nine features as lesion size by pathological diagnosis, bronchial cut-off sign, lobulation sign, vacuole sign, lesion size by CT, nodule type, enlarged lymph nodes, interlobar fissure and nodule location in the upper lobe of the left lung. Lesion size by pathological diagnosis emerged as a crucial factor in predicting and distinguishing the nature of nodules through imaging. The nodule size was instrumental in assessing malignant potential, consistent with findings from previous observational studies [ 9 , 12 ]. The likelihood of malignancy positively correlates with nodule diameter. Additionally, the presence of lobulation sign, vacuole sign, and enlarged lymph nodes was more common in unexpectedly benign nodules than in the benign group, indicating that these signs play a key role in the diagnosis of malignant nodules[ 13 ]. Reducing unnecessary surgery for lung nodules that are diagnostically imaged as malignant but ultimately found to be inflammatory granulomas postoperatively is crucial. According to the results of this retrospective clinical study, several strategies can help avoid surgery for unexpectedly benign nodules. One key approach involves making more cautious decisions for younger patients, as the data indicated that patients in this study were generally younger than those typically diagnosed with lung cancer. Additionally, PET-CT showed poor accuracy for small nodules, with 28 cases in the unexpectedly benign group where both CT and PET-CT diagnoses were falsely identified as malignant. For small nodules, extending follow-up rather than immediately opting for surgical removal may be more beneficial. Moreover, if a nodule appears malignant on axial CT scans, additional confidence can be gained by reviewing coronal and sagittal views to better clarify the nature of the nodule. Post-processing with three-dimensional reconstruction technology can further enhance the assessment by providing a direct and comprehensive view of the nodule’s margins, density, size, position, shape, and its relationship with surrounding tissues. This comprehensive visualization aids physicians in better understanding the lesion and its interaction with adjacent structures. Implementing these approaches can significantly reduce unnecessary surgeries for benign nodules. This study has several limitations. First, as a retrospective clinical study, there may be inherent biases in case selection, since we only included pulmonary nodules that were surgically resected, excluding those that were monitored over the long term without surgical intervention. Second, the data used to build our model were sourced from a single center, which may limit the model’s generalizability. 6. Conclusion Unexpectedly benign nodules often display imaging features characteristic of non-solid nodules, including signs typically associated with malignancy such as vascular clustering, enlarged lymph nodes, vacuole signs, spiculation, and lobulation. These are found alongside features like calcification and satellite nodules, which are more commonly linked to benign nodules. SVM linear model as a diagnostic model for pulmonary nodules selected relevant features and clarify the proportion of imaging differences between benign and malignant nodules. Ultimately, benign pulmonary nodules that are mistakenly suspected of malignancy often present in imaging studies with larger volumes and are accompanied by lobulation, vacuole signs, and enlarged lymph nodes. Declarations Authors’ contributions Yanrong Meng and Fengsheng Wang: data review, data analysis and manuscript revising. Dan Liu: patient enrollment and data collection. Jing Wang and Yan Wang: data review and manuscript revising. Jilai Zhang and Junkang Wang : data collection. Shaoquan Huang and Jiayi Xing: manuscript revising. Xindan Kang: data collection and analysis, manuscript supervision and editing. All authors have reviewed and approved the final manuscript. Ethical approval and consent to participate Approval for this study was obtained from the Ethic Committee of the First Medical Centre of Chinese People’s Liberation Army General Hospital. All procedures performed in our studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and principles of the Declaration of Helsinki by the World Medical Association. All patients provided informed consent before inclusion in the study. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Funding not applicable. Availability of data and materials All data generated or analysed during this study are included in this published article. The data of this study are available from the corresponding author on reasonable request. References National Lung Screening Trial Research Team. Lung cancer incidence and mortality with extended follow-up in the National Lung Screening Trial. J Thorac Oncol. 2019;14(10):1732–42. Scott J, Adams E, Stone DR, Baldwin, et al. Lung cancer Screen Lancet. 2023;401(10374):390–408. Snoeckx A, Reyntiens P, Desbuquoit D, et al. Evaluation of the solitary pulmonary nodule: size matters, but do not ignore the power of morphology. Insights Imaging. 2018;9:73–86. Xue W, Kong L, Zhang X, et al. Tumor blood vessel in 3D reconstruction CT imaging as a risk indicator for growth of pulmonary nodule with ground-glass opacity. J Cardiothorac Surg. 2023;18(1):333. Gao F, Sun Y, Zhang G, et al. CT characterization of different pathological types of sub-centimeter pulmonary ground-glass nodular lesions. Br J Radiol. 2019;92(1094):20180204. Ambrosini V, Nicolini S, Caroli P, et al. PET/CT imaging in different types of lung cancer: an overview. Eur J Radiol. 2012;81:988–1001. Brittney M, Williams J, Herb L, Dawson et al. The prevalence of benign pathology following major pulmonary resection for suspected malignancy. J Surg Res.2021;(268): 498–506. Archer JM, Mendoza DP, Hung YP,, et al. JTO Clin Res Rep. 2023;4(12):100605. Wang QZYFY, et al. China national guideline of classification, diagnosis and treatment for lung nodules (2016 Version). Chin J Lung Cancer. 2016;19:793–8. Zihan Zhou W, Guo D, Liu, et al. Multiparameter prediction model of immune checkpoint inhibitors combined with chemotherapy for non-small cell lung cancer based on support vector machine learning. Sci Rep. 2023;13(1):4469. Syaza AN, Samah A, Azurah L, JiTong et al. Support vector machine–Recursive feature elimination for feature selection on multi-omics lung cancer data. Progress Microbes Mol Biology.2023; 6(1). Annemie Snoeck P, Reyntiens D, Desbuquoit, et al. Evaluation of the solitary pulmonary nodule: size matters, but do not ignore the power of morphology. Insights into imaging. 2018;9:73–86. Ma XB, Xu QL, Li N, et al. A decision tree model to distinguish between benign and malignant pulmonary nodules on CT scans. Eur Rev Med Pharmacol Sci. 2023;27(12):5692–9. Blechter B, Wong JYY, Chien LH, et al. Age at lung cancer diagnosis in females versus males who never smoke by race and ethnicity. Br J Cancer. 2024;130(8):1286–94. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5361749","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":375846095,"identity":"de8cf3d4-0e0b-4805-a0c4-9cc885a20838","order_by":0,"name":"Yanrong Meng","email":"","orcid":"","institution":"Chinese People’s Liberation Army General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yanrong","middleName":"","lastName":"Meng","suffix":""},{"id":375846098,"identity":"17c621c7-dcc5-490c-b16e-eb05e8e96987","order_by":1,"name":"Fengsheng Wang","email":"","orcid":"","institution":"Chinese People’s Liberation Army General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Fengsheng","middleName":"","lastName":"Wang","suffix":""},{"id":375846099,"identity":"73ec048d-71bc-4288-9b7c-6df95c413b01","order_by":2,"name":"Dan Liu","email":"","orcid":"","institution":"Yulin Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Dan","middleName":"","lastName":"Liu","suffix":""},{"id":375846100,"identity":"04839772-3057-458a-81c5-45a19a6c68aa","order_by":3,"name":"Jing Wang","email":"","orcid":"","institution":"Chinese People’s Liberation Army General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jing","middleName":"","lastName":"Wang","suffix":""},{"id":375846101,"identity":"fa3f8e88-6c21-4277-b7ca-c752f132cf8e","order_by":4,"name":"Yan Wang","email":"","orcid":"","institution":"Chinese People’s Liberation Army General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yan","middleName":"","lastName":"Wang","suffix":""},{"id":375846102,"identity":"35295157-bf39-4c93-8e11-03c63e6c80ed","order_by":5,"name":"Jilai Zhang","email":"","orcid":"","institution":"Chinese People’s Liberation Army General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jilai","middleName":"","lastName":"Zhang","suffix":""},{"id":375846103,"identity":"cf3c6146-20bd-4b1d-a277-af2df9547a96","order_by":6,"name":"Junkang Wang","email":"","orcid":"","institution":"The 943th Hospital of Joint Logistics Support Force of The Chinese People's Liberation Army","correspondingAuthor":false,"prefix":"","firstName":"Junkang","middleName":"","lastName":"Wang","suffix":""},{"id":375846104,"identity":"37bce4cd-ea03-4aee-b548-a7541168c5f7","order_by":7,"name":"Jiayi Xing","email":"","orcid":"","institution":"Chinese People’s Liberation Army General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jiayi","middleName":"","lastName":"Xing","suffix":""},{"id":375846105,"identity":"1a6efe68-ec58-4ddb-83f0-7f1550d9da09","order_by":8,"name":"Shaoquan Huang","email":"","orcid":"","institution":"State Key Laboratory of NBC Protection for Civilian","correspondingAuthor":false,"prefix":"","firstName":"Shaoquan","middleName":"","lastName":"Huang","suffix":""},{"id":375846106,"identity":"2c03efe7-a8c5-4062-9910-ab93e70b3006","order_by":9,"name":"Xindan Kang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAt0lEQVRIiWNgGAWjYBACCQYGg88MDMxybOzNB4jWYjibIYHZmI/nWAJpWhLnSeQoEKdFckbyxubCH9bpbQw5DAw/KrYR1iLNc6yweUZCem4bw9kDjD1nbhPWIsfeY/6YJ+FwbhtjXwIzYxsxWph5DJuBWtLZmHkMiNMizd4D1pLAxkasFskeoF940tIN23jYEg4S5ReJG8AQ47Gxlpef//jggx8VRGhBAQdIVD8KRsEoGAWjABcAANgfN3vvNRH9AAAAAElFTkSuQmCC","orcid":"","institution":"Chinese People’s Liberation Army General Hospital","correspondingAuthor":true,"prefix":"","firstName":"Xindan","middleName":"","lastName":"Kang","suffix":""}],"badges":[],"createdAt":"2024-10-30 14:38:31","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5361749/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5361749/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":71100469,"identity":"a30d9bc3-9579-49a8-9cb1-2c7a50961666","added_by":"auto","created_at":"2024-12-11 06:45:25","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1039753,"visible":true,"origin":"","legend":"\u003cp\u003eThe AUC curve area of six algorithm models\u003c/p\u003e","description":"","filename":"figure1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5361749/v1/bd8ccc872f84fe744dcc5884.jpg"},{"id":71100464,"identity":"3e5539fd-299e-4d21-a44b-6aa2c06825ca","added_by":"auto","created_at":"2024-12-11 06:45:25","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":732366,"visible":true,"origin":"","legend":"\u003cp\u003ethe feature prediction of REF method\u003c/p\u003e","description":"","filename":"figure2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5361749/v1/bf253a55173486631273ba4f.jpg"},{"id":71100467,"identity":"2072ecd3-6cad-40f0-afc6-e6bbc91902e6","added_by":"auto","created_at":"2024-12-11 06:45:25","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":778474,"visible":true,"origin":"","legend":"\u003cp\u003eThe AUC curve area of reintroduced into the support vector machine model\u003c/p\u003e","description":"","filename":"figure3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5361749/v1/d94b7ae3fff46b42d82c11a0.jpg"},{"id":71100873,"identity":"1b7d2322-694c-47df-981f-67fc88d38b31","added_by":"auto","created_at":"2024-12-11 06:53:25","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1216986,"visible":true,"origin":"","legend":"\u003cp\u003eThe selected features were ranked according to their importance by REF method\u003c/p\u003e","description":"","filename":"figure4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5361749/v1/63238442817451126f87d465.jpg"},{"id":71100465,"identity":"5de668c4-091a-4b27-9510-60508bf807fb","added_by":"auto","created_at":"2024-12-11 06:45:25","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":777735,"visible":true,"origin":"","legend":"\u003cp\u003ethe degree of association between features\u003c/p\u003e","description":"","filename":"figure5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5361749/v1/5b1dccdcae3547525b3f45cc.jpg"},{"id":71100468,"identity":"3d9494ab-4b6a-4437-b559-0f5e7a19d197","added_by":"auto","created_at":"2024-12-11 06:45:25","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":443310,"visible":true,"origin":"","legend":"\u003cp\u003eCoefficient plot of correlation between features and outcome of benign and malignant nodules\u003c/p\u003e","description":"","filename":"figure6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5361749/v1/03e83d02b4abec79cbfb3fad.jpg"},{"id":71100874,"identity":"0a83605b-e5a5-4d52-b115-7afc5ef56a7b","added_by":"auto","created_at":"2024-12-11 06:53:25","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":819614,"visible":true,"origin":"","legend":"\u003cp\u003eThe AUC curve area of reintroduced into the support vector machine model after manual selection\u003c/p\u003e","description":"","filename":"figure7.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5361749/v1/79139fd8e214346855b133a0.jpg"},{"id":71100470,"identity":"008ea3db-9900-47d8-91d7-1185d456ca03","added_by":"auto","created_at":"2024-12-11 06:45:25","extension":"jpg","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":983171,"visible":true,"origin":"","legend":"\u003cp\u003eThe selected features were ranked according to their importance by manual selection\u003c/p\u003e","description":"","filename":"figure8.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5361749/v1/f63e6f8e31a3a4b3cf795450.jpg"},{"id":71103135,"identity":"b372233d-aa58-4bad-942b-08432905c1f4","added_by":"auto","created_at":"2024-12-11 07:09:34","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":7182168,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5361749/v1/61b9e8a3-fa05-4206-b4af-f5b9c5b828b2.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Unexpectedly Benign Pulmonary Nodules Using Machine Learning","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eWith the growing emphasis on health consciousness and the widespread use of high-resolution computed tomography (HRCT), early detection and resection of malignant pulmonary nodules hold the potential to improve lung cancer outcomes. Numerous studies support this conclusion, showing that CT screening not only delays lung cancer mortality by more than a decade but also prevents deaths altogether [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. However, there are ongoing concerns that need to be addressed, including the potential drawbacks of overdiagnosis and false-positive lung cancer diagnoses. Surgical resections of benign nodules mistakenly presumed to be malignant are not uncommon, given the overlapping imaging features of benign and malignant nodules. Differentiating between these two types of nodules remains a significant challenge for radiologists and oncologists.\u003c/p\u003e \u003cp\u003eDespite the overlapping characteristics, the importance of morphological features should not be underestimated. Numerous studies have indicated that the appearance of nodule density, vascular aggregation, vacuole sign, spiculation, and lobulation are associated with a higher likelihood of malignancy, whereas features like calcification and satellite nodules are more commonly linked to benignity [\u003cspan additionalcitationids=\"CR4\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Sometimes, the effects of lung CT scanning on evaluation results of pulmonary nodules are limited. Additional techniques, such as 3D reconstruction CT or 18F-fluorodeoxyglucose (18F-FDG) positron emission tomography (PET-CT), may offer added value. For the latter, the tracer is 18F-FDG, which is an 18F-fluorolabeled glucose analogue. However, it is noteworthy that subcentimeter nodules, carcinoma in situ, or adenocarcinoma precursors can exhibit low or no uptake. Similarly, the glucose metabolism of both tumor and inflammatory graunloma can be elevated, potentially leading to false-positive results in cases of inflammatory granuloma [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Therefore, it is crucial to weigh these factors carefully when distinguishing between malignant and benign nodules in clinical practice.\u003c/p\u003e \u003cp\u003eResearch on surgically resected benign nodules initially considered malignant is scarce in the existing literature. For example, only 10 patients were investigated in Williams\u0026rsquo; study and 19 of 148(12.8%) underwent surgical resection had a benign diagnosis in Archer\u0026rsquo;s research\u0026rsquo;s [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. The aim of our study is to assess and compare the characteristics of surgically resected benign nodules detected on CT scans that were initially presumed to be malignant. This research seeks to identify unexpectedly benign pulmonary nodules by comparing imaging-based risk factors for nodules classified as benign versus those deemed malignant. Baseline comparisons were made, followed by an assessment of the significance of the differing risk factors. The optimal model was selected based on the area under the curve (AUC) from various machine learning models. This model aims to help identify the imaging characteristics of unexpectedly benign pulmonary nodules, thereby reducing unnecessary surgeries.\u003c/p\u003e"},{"header":"2. Materials and methods","content":"\u003ch2\u003e2.1 Study population and design\u003c/h2\u003e\n\u003cp\u003eIn this study, we reviewed cases of patients who underwent pulmonary nodule resection at the First Medical Centre of the Chinese People\u0026rsquo;s Liberation Army (PLA) General Hospital. A radiologist specializing in lung imaging and two physicians independently collected demographic data, clinical characteristics, imaging features, and pathological diagnoses of the resected nodules from electronic medical records. Personal information was kept confidential in this retrospective study. The Ethics Committee of the First Medical Centre of the Chinese PLA General Hospital approved the study\u0026nbsp;and the requirement for informed patient consent was waived.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003e2.2 Inclusion and exclusion criteria\u003c/h2\u003e\n\u003cp\u003eInclusion criteria were applied：\u003c/p\u003e\n\u003cp\u003e1). All pulmonary nodules were detected from January 2017 to July 2023,\u003c/p\u003e\n\u003cp\u003e2). All patients \u0026gt;18 years old.\u003c/p\u003e\n\u003cp\u003e3). Patients surgically resected with a presumed pulmonary granulomatous lesion.\u003c/p\u003e\n\u003cp\u003eExclusion criteria were as follows：\u003c/p\u003e\n\u003cp\u003e1). Patients with suspected metastatic disease based on chest CT,\u003c/p\u003e\n\u003cp\u003e2). Patients with malignant lung nodules diagnosed by histopathological diagnosis.\u003c/p\u003e\n\u003cp\u003e3). Patients with unconfirmed postoperative pathological tests data incomplete clinical data.\u003c/p\u003e"},{"header":"3. Statistical analysis","content":"\u003cp\u003eDemographic characteristics were presented as frequencies and percentages for categorical variables and as means with standard deviations for continuous variables (assuming normal distribution). Differences between the two groups\u0026rsquo; continuous variables were compared using either a t-test or the Mann\u0026ndash;Whitney U test, while categorical data were analyzed using chi-square analysis. P values were calculated using two-sided exact tests, with values less than 0.05 considered statistically significant. All analyses were performed using SPSS (Statistical Program for Social Sciences) software, version 26. Additionally, multiple machine learning models were constructed, including Random Forest, SVM linear, SVM nonlinear, Logistic Regression, XgBoost, and k-Nearest Neighbors, with the optimal model selected based on performance on the validation set. Nodule location was represented using one-hot encoding, and other data were converted to binary values (0 and 1). The entire dataset was divided into a training set (162 samples) and a validation set (41 samples).\u003c/p\u003e \u003cp\u003eThe model construction details are as follows:\u003c/p\u003e \u003cp\u003e1). Random Forest (RF): The training set was used to identify optimal hyperparameters through grid search. The number of trees in the random forest was set to 500.\u003c/p\u003e \u003cp\u003e2). Support Vector Machine Linear (SVM linear): A linear kernel function was chosen for classification. Optimal hyperparameters were determined using grid search on the training set, with a penalty coefficient of 1.\u003c/p\u003e \u003cp\u003e3). Support Vector Machine Nonlinear (SVM nonlinear): A Gaussian radial basis function kernel was selected for nonlinear classification. Optimal hyperparameters were identified through grid search on the training set, with a penalty coefficient of 0.01 and a gamma coefficient of 0.1.\u003c/p\u003e \u003cp\u003e4). Logistic Regression (Log): Optimal hyperparameters were identified using grid search on the training set. The internal gradient descent algorithm was set to converge with a maximum of 1,000 iterations, and a penalty coefficient of 1.\u003c/p\u003e \u003cp\u003e5). XgBoost (XGB): The training set was used to identify optimal hyperparameters through grid search. The model was set to 100 trees, with a learning rate of 0.1 and a subsample ratio per column of 0.6.\u003c/p\u003e \u003cp\u003e6). k-Nearest Neighbors (KNN): Utilizing Euclidean distance, optimal hyperparameters were identified through grid search on the training set, with the number of neighbors set to 10.\u003c/p\u003e \u003cp\u003eThe predictive efficiency of these models was evaluated using AUC curves on the test set.\u003c/p\u003e"},{"header":"4. Results","content":"\u003ch2\u003e4.1 Basic characteristics\u003c/h2\u003e\n\u003cp\u003eA total of 203 patients underwent more than one chest CT screening. Of these, 86 were diagnosed with benign nodules based on diagnostic imaging tests (both CT and PET-CT), with 11 of these confirmed as benign on PET-CT. The remaining 117 patients were diagnosed with malignant nodules through chest CT or PET-CT scans. Among them, 88 were identified as malignant on CT. PET-CT further refined the diagnosis for 67 of the 117 patients, including 19 who were initially diagnosed as benign on CT and 10 whose nodules could not be definitively classified as benign or malignant by CT alone. PET-CT results showed that 57 patients were malignant, 7 were benign, and 3 remained inconclusive. Notably, both CT and PET-CT indicated malignancy in 28 patients.\u003c/p\u003e\n\u003cp\u003eOverall, the pathological diagnosis for all 203 patients with primary pulmonary nodules was pulmonary granulomas. The following table provides details on the patients\u0026rsquo; characteristics and the imaging features of the nodules. Significant differences were observed between the two groups in maximum lesion size (pathological diagnosis), nodule density, boundary, calcification, satellite nodules, vascular aggregation sign, vacuole sign, spiculation sign, lobulation sign, pleural indentation sign, and nodule location (p\u0026lt;0.05). However, no significant differences were found in age, gender, smoking status, maximum lesion size (imaging diagnosis), nodule shape, proximity to the interlobar fissure, proximity to the pleura, bronchial cut-off sign, lymph node enlargement, or nodule count between the two groups (p\u0026gt;0.05).\u003c/p\u003e\n\u003cp\u003eTable 1 Clinical and imaging features of patients who underwent resection for benign nodule\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"699\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eClinical Factor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 227px;\"\u003e\n \u003cp\u003eBenign nodule diagnosed by imaging\u003c/p\u003e\n \u003cp\u003e(n=86)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 189px;\"\u003e\n \u003cp\u003eMalignant nodule diagnosed by imaging\u003c/p\u003e\n \u003cp\u003e(n=117)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003eP-value\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003eNo.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003eNo.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eMedian age (range)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 227px;\"\u003e\n \u003cp\u003e54.4\u0026plusmn;10.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 189px;\"\u003e\n \u003cp\u003e52.2\u0026plusmn;9.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003cp\u003e\u0026gt;=55 years\u003c/p\u003e\n \u003cp\u003e\u0026lt;55 years\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e36\u003c/p\u003e\n \u003cp\u003e50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e41.86\u003c/p\u003e\n \u003cp\u003e58.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e61\u003c/p\u003e\n \u003cp\u003e56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e52.14\u003c/p\u003e\n \u003cp\u003e47.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.107\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eGender\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;Male\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;Female\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e55\u003c/p\u003e\n \u003cp\u003e31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e63.95\u003c/p\u003e\n \u003cp\u003e36.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e65\u003c/p\u003e\n \u003cp\u003e52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e55.56\u003c/p\u003e\n \u003cp\u003e44.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.229\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eSmoking status\u003c/p\u003e\n \u003cp\u003eCurrent or past smoker\u003c/p\u003e\n \u003cp\u003eNonsmoker\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e37\u003c/p\u003e\n \u003cp\u003e49\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e43.02\u003c/p\u003e\n \u003cp\u003e56.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e40\u003c/p\u003e\n \u003cp\u003e77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e34.19\u003c/p\u003e\n \u003cp\u003e65.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.200\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eMaximum lesion size, (cm)\u003c/p\u003e\n \u003cp\u003e(imaging diagnosis)\u003c/p\u003e\n \u003cp\u003e\u0026gt;1\u003c/p\u003e\n \u003cp\u003e\u0026lt;=1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e68\u003c/p\u003e\n \u003cp\u003e18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e79.07\u003c/p\u003e\n \u003cp\u003e20.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e97\u003c/p\u003e\n \u003cp\u003e20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e82.91\u003c/p\u003e\n \u003cp\u003e17.09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.488\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eMaximum lesion size, (cm)\u003c/p\u003e\n \u003cp\u003e(pathological diagnosis)\u003c/p\u003e\n \u003cp\u003e\u0026gt;1\u003c/p\u003e\n \u003cp\u003e\u0026lt;=1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e63\u003c/p\u003e\n \u003cp\u003e23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e73.26\u003c/p\u003e\n \u003cp\u003e26.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e79\u003c/p\u003e\n \u003cp\u003e38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e67.52\u003c/p\u003e\n \u003cp\u003e32.48\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eNodule density\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eSolid\u003c/p\u003e\n \u003cp\u003eSubsolid (ground-glass, part-solid, etc)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e78\u003c/p\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e90.70\u003c/p\u003e\n \u003cp\u003e9.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e93\u003c/p\u003e\n \u003cp\u003e24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e79.49\u003c/p\u003e\n \u003cp\u003e20.51\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.030\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eBoundary\u003c/p\u003e\n \u003cp\u003eClear\u003c/p\u003e\n \u003cp\u003eBlur\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e49\u003c/p\u003e\n \u003cp\u003e37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e56.98\u003c/p\u003e\n \u003cp\u003e43.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e52\u003c/p\u003e\n \u003cp\u003e65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e44.44\u003c/p\u003e\n \u003cp\u003e55.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.024\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eShape of node\u003c/p\u003e\n \u003cp\u003eRound\u003c/p\u003e\n \u003cp\u003eIrregular\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e26\u003c/p\u003e\n \u003cp\u003e60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e30.23\u003c/p\u003e\n \u003cp\u003e69.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e26\u003c/p\u003e\n \u003cp\u003e91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e22.22\u003c/p\u003e\n \u003cp\u003e77.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.196\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eCalcification\u003c/p\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e25\u003c/p\u003e\n \u003cp\u003e61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e29.07\u003c/p\u003e\n \u003cp\u003e70.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e14\u003c/p\u003e\n \u003cp\u003e103\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e11.97\u003c/p\u003e\n \u003cp\u003e88.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eSatellite nodule(s)\u003c/p\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e23\u003c/p\u003e\n \u003cp\u003e63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e26.74\u003c/p\u003e\n \u003cp\u003e73.26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e18\u003c/p\u003e\n \u003cp\u003e99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e15.38\u003c/p\u003e\n \u003cp\u003e84.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.046\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eVascular aggregation sign\u003c/p\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003cp\u003e78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e9.30\u003c/p\u003e\n \u003cp\u003e90.70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e25\u003c/p\u003e\n \u003cp\u003e92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e21.37\u003c/p\u003e\n \u003cp\u003e78.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.021\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eAdjacent to interlobar fissure\u003c/p\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e29\u003c/p\u003e\n \u003cp\u003e57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e33.72\u003c/p\u003e\n \u003cp\u003e66.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e40\u003c/p\u003e\n \u003cp\u003e77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e34.19\u003c/p\u003e\n \u003cp\u003e65.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.945\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eAdjacent to pleura\u003c/p\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e71\u003c/p\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e82,56\u003c/p\u003e\n \u003cp\u003e17.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e89\u003c/p\u003e\n \u003cp\u003e28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e76.07\u003c/p\u003e\n \u003cp\u003e23.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.263\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eBronchial cut-off sign\u003c/p\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003cp\u003e82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e4.65\u003c/p\u003e\n \u003cp\u003e95.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003cp\u003e106\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e9.40\u003c/p\u003e\n \u003cp\u003e90.60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.201\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eLymph nodes enlarge\u003c/p\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003cp\u003e85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e1.16\u003c/p\u003e\n \u003cp\u003e98.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003cp\u003e109\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e6.84\u003c/p\u003e\n \u003cp\u003e93.16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.052\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eVacuole sign\u003c/p\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003cp\u003e84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e2.33\u003c/p\u003e\n \u003cp\u003e97.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003cp\u003e102\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e12.82\u003c/p\u003e\n \u003cp\u003e87.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.008\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eSpiculation sign\u003c/p\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e18\u003c/p\u003e\n \u003cp\u003e68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e20.93\u003c/p\u003e\n \u003cp\u003e79,07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e51\u003c/p\u003e\n \u003cp\u003e66\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e43.59\u003c/p\u003e\n \u003cp\u003e56.41\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eLobulation sign\u003c/p\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003cp\u003e71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e17.44\u003c/p\u003e\n \u003cp\u003e82.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e61\u003c/p\u003e\n \u003cp\u003e56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e52.14\u003c/p\u003e\n \u003cp\u003e47.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eNodule count\u003c/p\u003e\n \u003cp\u003eSolitary\u003c/p\u003e\n \u003cp\u003eMultiple\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e66\u003c/p\u003e\n \u003cp\u003e20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e76.74\u003c/p\u003e\n \u003cp\u003e23.26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e96\u003c/p\u003e\n \u003cp\u003e21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e82.05\u003c/p\u003e\n \u003cp\u003e17.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.352\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003ePleural indentation sign\u003c/p\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e57\u003c/p\u003e\n \u003cp\u003e29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e66.28\u003c/p\u003e\n \u003cp\u003e33.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e70\u003c/p\u003e\n \u003cp\u003e47\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e59.83\u003c/p\u003e\n \u003cp\u003e40.17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.023\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003eNodule location\u003c/p\u003e\n \u003cp\u003eUpper lobe of left lung\u003c/p\u003e\n \u003cp\u003eLower lobe of left lung\u003c/p\u003e\n \u003cp\u003eUpper lobe of right lung\u003c/p\u003e\n \u003cp\u003eMiddle lobe of right lung\u003c/p\u003e\n \u003cp\u003eLower lobe of right lung\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003cp\u003e29\u003c/p\u003e\n \u003cp\u003e17\u003c/p\u003e\n \u003cp\u003e20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e17.44\u003c/p\u003e\n \u003cp\u003e5.81\u003c/p\u003e\n \u003cp\u003e33.72\u003c/p\u003e\n \u003cp\u003e19.77\u003c/p\u003e\n \u003cp\u003e23.26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e20\u003c/p\u003e\n \u003cp\u003e22\u003c/p\u003e\n \u003cp\u003e34\u003c/p\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003cp\u003e29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e17.09\u003c/p\u003e\n \u003cp\u003e18.8\u003c/p\u003e\n \u003cp\u003e29.06\u003c/p\u003e\n \u003cp\u003e10.26\u003c/p\u003e\n \u003cp\u003e24.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.043\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eFig.1 The AUC curve area of six algorithm models\u003c/p\u003e\n\u003ch2\u003e4.2 Comparison of the performance of multiple machine learning methods in identifying malignant or benign nodules\u003c/h2\u003e\n\u003cp\u003eIn this study, we retrospectively analyzed 203 patients with pulmonary nodules, comprising 86 benign and 117 malignant cases based on imaging procedures. The patients were divided into a training set (n=162) and a validation set (n=41) at a 4:1 ratio. The key reference characteristics for differentiating between benign and malignant pulmonary nodules included age, gender, smoking status, maximum lesion size (as determined by imaging and pathological diagnosis), nodule density, boundary, shape, calcification, satellite nodules, vascular aggregation sign, proximity to the interlobar fissure and pleura, bronchial cut-off sign, lymph node enlargement, vacuole sign, spiculation sign, lobulation sign, nodule count, pleural indentation sign, and nodule location.\u003c/p\u003e\n\u003cp\u003eThe performance of several machine learning methods was evaluated, including Random Forest, SVM linear, SVM nonlinear, logistic regression, XgBoost, and k-Nearest Neighbor, on the validation set. The AUC curves of these models were constructed to assess their predictive efficiency on the training set. Among the models, the SVM linear model demonstrated the highest AUC of 0.867, significantly outperforming the others. To maximize the identification of risk factors between malignant and benign nodules based on imaging, the SVM linear model was selected for further development of the predictive model (Fig. 1).\u003c/p\u003e\n\u003cp\u003eFig.2 the feature prediction of REF method \u0026nbsp; \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFig.3 The AUC curve area of reintroduced into the support vector machine model\u003c/p\u003e\n\u003ch2\u003e4.3 Feature selection for constructing the final classifier\u003c/h2\u003e\n\u003cp\u003eThe Recursive Feature Elimination (RFE) method was employed to validate the accuracy of feature predictions, revealing that error values were minimized when using 23 features during ten-fold cross-validation (Fig. 2). These 23 selected features were then reintroduced into the support vector machine model for a comparative analysis of predictive performance. This analysis showed an increase in the AUC for predictive efficiency, from 0.867 to 0.875 (Fig. 3). The selected features were ranked according to their importance, resulting in a distribution of feature relevance depicted in Figure 4. The lesion size measured by pathological diagnosis emerged as the most significant predictor in differentiating between benign and malignant pulmonary nodules. The top nine factors influencing this clinical diagnosis included lesion size by pathological diagnosis, bronchial cut-off sign, lobulation sign, vacuole sign, lesion size by CT, nodule type, lymph node enlargement, proximity to the interlobar fissure, and nodule location in the upper lobe of the left lung.\u003c/p\u003e\n\u003cp\u003eFig.4 The selected features were ranked according to their importance by REF method\u003c/p\u003e\n\u003cp\u003eFinally, we eliminated features with collinearity and retained only those with strong relevance to the target outcome (nodule nature) through manual feature selection. Given that the data consisted of binary categories (0 and 1), the Pearson correlation coefficient was unsuitable for calculating correlations between features. Instead, Cram\u0026eacute;r\u0026rsquo;s V coefficient was used, which is specifically designed to measure the degree of association between categorical variables (Fig.5). This coefficient ranges from 0 to 1, where 0 indicates no association and 1 indicates complete correlation. The Cram\u0026eacute;r\u0026rsquo;s V was calculated coefficient between features, identifying those with a Cram\u0026eacute;r\u0026rsquo;s V greater than 0.3 and a p-value less than 0.05 as collinear and those with a Cram\u0026eacute;r\u0026rsquo;s V greater than 0.1 or a p-value less than 0.05 as strongly relevant to the target outcome. Ultimately, the features were retained strongly related to the target outcome and removed those that were collinear (Fig. 6).\u003c/p\u003e\n\u003cp\u003eFig.5 the degree of association between features\u003c/p\u003e\n\u003cp\u003eFig.6 Coefficient plot of correlation between features and outcome of benign and malignant nodules\u003c/p\u003e\n\u003cp\u003eThrough manual selection, we observed an increase in predictive efficacy, with the AUC rising from 0.867 to 0.885 (Fig. 7). The selected features were ranked by importance, resulting in a distribution of feature relevance as shown in Figure 8. Nine key features were identified ultimately: lesion size by pathological diagnosis, lobulation sign, tumor-lung interface, calcification, satellite nodules, spiculation sign, smoking history, vacuole sign, and lymph node enlargement.\u003c/p\u003e\n\u003cp\u003eFig.7 The AUC curve area of reintroduced into the support vector machine model after manual selection\u003c/p\u003e\n\u003cp\u003eFig.8 The selected features were ranked according to their importance by manual selection\u003c/p\u003e"},{"header":"5. Discussion","content":"\u003cp\u003eThe management of pulmonary nodules primarily focuses on distinguishing between benign and malignant nodules, aiming to balance the need for prompt intervention for malignant cases against avoiding unnecessary surgeries for benign ones. Unlike biopsy and surgical resection, CT as a non-invasive procedure is commonly used as a diagnostic tool for pre-surgical evaluation [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Radiologists typically estimate the probability of lung cancer based on the CT characteristics of the nodules. In our study, larger lesion size and the presence of subsolid nodules indicated a higher probability of malignancy. Nodule margin is another critical feature in assessing malignancy risk. Malignant nodules often exhibit signs of spiculation, lobulation, and vascular aggregation, while benign nodules are more likely to display calcification and satellite nodules. Additionally, the presence of a vacuole sign, pleural indentation, and hilar or mediastinal lymph nodes enlarged to more than 1.0 cm in the shortest diameter were more likely to suggest malignancy (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). In our study, only 38.5% (78/203) of patients with pulmonary nodules underwent PET-CT before surgery. Notably, pulmonary nodules that were unexpectedly benign often displayed malignant characteristics on lung CT or PET-CT scans before surgical resection.\u003c/p\u003e \u003cp\u003eTo further analyze the proportion of factors distinguishing imaging-diagnosed benign pulmonary nodules from unexpectedly benign nodules, multiple machine learning models were employed to differentiate between benign and malignant nodules. The SVM linear model performed exceptionally well in predictions. After that, feature selection was performed on distinguishing benign nodules and unexpectedly benign pulmonary Nodules using SVM-RFE and SVM-manual selection to select several optimal feature subsets [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. The RFE method selected 23 features, and rebuilding the SVM model with these improved the predictive efficacy from 0.867 to 0.875. Through manual selection, nine key features were identified and used to rebuild the SVM model, resulting in an increase in predictive efficacy from 0.867 to 0.885. Both feature selection methods enhanced the model\u0026rsquo;s predictive performance [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe nine features identified through manual selection were lesion size by pathological diagnosis, lobulation sign, tumor-lung interface, calcification, satellite nodules, spiculation sign, smoking history, vacuole sign, and enlarged lymph nodes. The RFE method identified the top nine features as lesion size by pathological diagnosis, bronchial cut-off sign, lobulation sign, vacuole sign, lesion size by CT, nodule type, enlarged lymph nodes, interlobar fissure and nodule location in the upper lobe of the left lung. Lesion size by pathological diagnosis emerged as a crucial factor in predicting and distinguishing the nature of nodules through imaging. The nodule size was instrumental in assessing malignant potential, consistent with findings from previous observational studies [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. The likelihood of malignancy positively correlates with nodule diameter. Additionally, the presence of lobulation sign, vacuole sign, and enlarged lymph nodes was more common in unexpectedly benign nodules than in the benign group, indicating that these signs play a key role in the diagnosis of malignant nodules[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eReducing unnecessary surgery for lung nodules that are diagnostically imaged as malignant but ultimately found to be inflammatory granulomas postoperatively is crucial. According to the results of this retrospective clinical study, several strategies can help avoid surgery for unexpectedly benign nodules. One key approach involves making more cautious decisions for younger patients, as the data indicated that patients in this study were generally younger than those typically diagnosed with lung cancer. Additionally, PET-CT showed poor accuracy for small nodules, with 28 cases in the unexpectedly benign group where both CT and PET-CT diagnoses were falsely identified as malignant. For small nodules, extending follow-up rather than immediately opting for surgical removal may be more beneficial. Moreover, if a nodule appears malignant on axial CT scans, additional confidence can be gained by reviewing coronal and sagittal views to better clarify the nature of the nodule. Post-processing with three-dimensional reconstruction technology can further enhance the assessment by providing a direct and comprehensive view of the nodule\u0026rsquo;s margins, density, size, position, shape, and its relationship with surrounding tissues. This comprehensive visualization aids physicians in better understanding the lesion and its interaction with adjacent structures. Implementing these approaches can significantly reduce unnecessary surgeries for benign nodules.\u003c/p\u003e \u003cp\u003eThis study has several limitations. First, as a retrospective clinical study, there may be inherent biases in case selection, since we only included pulmonary nodules that were surgically resected, excluding those that were monitored over the long term without surgical intervention. Second, the data used to build our model were sourced from a single center, which may limit the model\u0026rsquo;s generalizability.\u003c/p\u003e"},{"header":"6. Conclusion","content":"\u003cp\u003eUnexpectedly benign nodules often display imaging features characteristic of non-solid nodules, including signs typically associated with malignancy such as vascular clustering, enlarged lymph nodes, vacuole signs, spiculation, and lobulation. These are found alongside features like calcification and satellite nodules, which are more commonly linked to benign nodules. SVM linear model as a diagnostic model for pulmonary nodules selected relevant features and clarify the proportion of imaging differences between benign and malignant nodules. Ultimately, benign pulmonary nodules that are mistakenly suspected of malignancy often present in imaging studies with larger volumes and are accompanied by lobulation, vacuole signs, and enlarged lymph nodes.\u003c/p\u003e "},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAuthors\u0026rsquo; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eYanrong Meng and\u0026nbsp;Fengsheng Wang: data review, data analysis and manuscript revising.\u003c/p\u003e\n\u003cp\u003eDan Liu: patient enrollment and data collection.\u003c/p\u003e\n\u003cp\u003eJing Wang and\u0026nbsp;Yan Wang: data review and manuscript revising.\u003c/p\u003e\n\u003cp\u003eJilai Zhang and Junkang Wang : data collection.\u003c/p\u003e\n\u003cp\u003eShaoquan Huang and Jiayi Xing: manuscript revising.\u003c/p\u003e\n\u003cp\u003eXindan Kang: data collection and analysis, manuscript supervision and editing.\u003c/p\u003e\n\u003cp\u003eAll authors have reviewed and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthical approval and consent to participate \u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eApproval for this study was obtained from the Ethic Committee of the First Medical Centre of Chinese People\u0026rsquo;s Liberation Army General Hospital. All procedures performed in our studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and principles of the Declaration of Helsinki by the World Medical Association. All patients provided informed consent before inclusion in the study.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication \u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests \u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding \u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003enot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll data generated or analysed during this study are included in this published article. The data of this study are available from the corresponding author on reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eNational Lung Screening Trial Research Team. Lung cancer incidence and mortality with extended follow-up in the National Lung Screening Trial. J Thorac Oncol. 2019;14(10):1732\u0026ndash;42.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eScott J, Adams E, Stone DR, Baldwin, et al. Lung cancer Screen Lancet. 2023;401(10374):390\u0026ndash;408.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSnoeckx A, Reyntiens P, Desbuquoit D, et al. Evaluation of the solitary pulmonary nodule: size matters, but do not ignore the power of morphology. Insights Imaging. 2018;9:73\u0026ndash;86.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXue W, Kong L, Zhang X, et al. Tumor blood vessel in 3D reconstruction CT imaging as a risk indicator for growth of pulmonary nodule with ground-glass opacity. J Cardiothorac Surg. 2023;18(1):333.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGao F, Sun Y, Zhang G, et al. CT characterization of different pathological types of sub-centimeter pulmonary ground-glass nodular lesions. Br J Radiol. 2019;92(1094):20180204.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAmbrosini V, Nicolini S, Caroli P, et al. PET/CT imaging in different types of lung cancer: an overview. Eur J Radiol. 2012;81:988\u0026ndash;1001.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrittney M, Williams J, Herb L, Dawson et al. The prevalence of benign pathology following major pulmonary resection for suspected malignancy. J Surg Res.2021;(268): 498\u0026ndash;506.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArcher JM, Mendoza DP, Hung YP,, et al. JTO Clin Res Rep. 2023;4(12):100605.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang QZYFY, et al. China national guideline of classification, diagnosis and treatment for lung nodules (2016 Version). Chin J Lung Cancer. 2016;19:793\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZihan Zhou W, Guo D, Liu, et al. Multiparameter prediction model of immune checkpoint inhibitors combined with chemotherapy for non-small cell lung cancer based on support vector machine learning. Sci Rep. 2023;13(1):4469.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSyaza AN, Samah A, Azurah L, JiTong et al. Support vector machine\u0026ndash;Recursive feature elimination for feature selection on multi-omics lung cancer data. Progress Microbes Mol Biology.2023; 6(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAnnemie Snoeck P, Reyntiens D, Desbuquoit, et al. Evaluation of the solitary pulmonary nodule: size matters, but do not ignore the power of morphology. Insights into imaging. 2018;9:73\u0026ndash;86.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMa XB, Xu QL, Li N, et al. A decision tree model to distinguish between benign and malignant pulmonary nodules on CT scans. Eur Rev Med Pharmacol Sci. 2023;27(12):5692\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBlechter B, Wong JYY, Chien LH, et al. Age at lung cancer diagnosis in females versus males who never smoke by race and ethnicity. Br J Cancer. 2024;130(8):1286\u0026ndash;94.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"unexpectedly benign pulmonary nodules, lobulated sign, vacuole sign, imaging features, machine learning","lastPublishedDoi":"10.21203/rs.3.rs-5361749/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5361749/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground:\u003c/strong\u003e The widespread application of computed tomography (CT) for pulmonary disease screening has led to an increased detection of pulmonary nodules. However, this has also resulted in a high false-positive rate for suspected malignancies that ultimately prove to be benign. It is crucial to explore the clinical and imaging characteristics of these patients to avoid unnecessary major pulmonary resections. This study aims to evaluate the characteristics of surgically resected benign nodules that were initially presumed to be lung cancers based on lung CT scans.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMaterials and Methods:\u003c/strong\u003e This retrospective study analyzed 203 cases of benign lung nodules at the First Medical Centre of the Chinese People’s Liberation Army (PLA) General Hospital from January 2017 to June 2023. Pathological examination following surgical resection confirmed pulmonary granulomatous inflammation in these cases. The study cohort was divided into two groups: 86 patients with benign nodules and 117 with malignant nodules, all diagnosed based on imaging features. Given the overlapping imaging features of benign and malignant nodules, the clinical and imaging characteristics of both groups were compared to reduce the incidence of unnecessary surgeries. Various machine learning models, including Random Forest, SVM linear, SVM nonlinear, Logistic Regression, XGBoost, and k-Nearest Neighbor, were constructed. With the optimal model selected based on performance on a validation set, A framework was developed to identify personalized risk factors using a feature importance ranking algorithm.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults:\u003c/strong\u003e Analysis of data from 203 patients revealed significant differences in maximum lesion size (pathological diagnosis), nodule density, boundary, calcification, satellite nodules, vascular aggregation sign, vacuole sign, spiculation sign, lobulation sign, pleural indentation sign, and nodule location (p\u0026lt;0.05). The SVMlinear model, which achieved the highest AUC (0.867), was selected for the final predictive model.Using recursive feature elimination method and manual feature selection, the feature of lesion size, lobulation sign, vacuole sing and enlarged lymph nodes have been screened in predicting and distinguishing the nature of nodules through imaging.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion:\u003c/strong\u003eBenign pulmonary nodules that are unexpectedly resected often present with features typically associated with malignancy, such as larger volumes, lobulation, vacuole signs, and enlarged lymph nodes. This study highlights the importance of accurately distinguishing between benign and malignant nodules to minimize unnecessary surgical interventions.\u003c/p\u003e","manuscriptTitle":"Unexpectedly Benign Pulmonary Nodules Using Machine Learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-11 06:45:19","doi":"10.21203/rs.3.rs-5361749/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"c345a666-e087-483f-90b4-e2d4599a7bd4","owner":[],"postedDate":"December 11th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-12-11T06:45:23+00:00","versionOfRecord":[],"versionCreatedAt":"2024-12-11 06:45:19","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5361749","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5361749","identity":"rs-5361749","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}