Development and validation of a CT-based comprehensive nomogram for differentiating benign from malignant subcentimeter solid nodules

Introduction

Lung cancer is the prominent cause for cancer-related death globally (1,2). Subcentimeter pulmonary nodules (SSPNs, <1 cm) represent the smallest classification of nodules in T staging. In contrast to subsolid nodules, which usually grow slowly and have less aggressive characteristics, solid nodules grow faster and are more likely to be invasive adenocarcinomas (IAC) (3). Malignant solid pulmonary nodules have a higher malignancy rate, faster growth, greater risk of lymph node metastasis, and poorer prognosis (4-6). Timely diagnosis and treatment of malignant solid nodules at the subcentimeter stage can significantly improve patient outcomes (7). Numerous studies have shown that early detection and timely diagnostic evaluation of pulmonary nodules can substantially reduce cancer-related mortality and enhance 5-year survival rates (8,9). In addition, delayed diagnosis of malignant solid nodules may worsen patient prognosis (10). Therefore, a more accurate and earlier diagnostic method is essential for patients with SSPNs.

Currently, high-resolution computed tomography (HRCT) is the primary method for observing the lungs or SSPNs. However, small nodules often do not have clear imaging features, which makes it hard to tell the difference between benign and malignant SSPNs (11). With the development of medical imaging technology, there is an urgent need for more advanced non-invasive techniques to overcome these limitations. In recent years, multiplanar volume rendering (MPVR) represents a more advanced three-dimensional (3D) volume reconstruction technique than traditional volume rendering, which enables better differentiation between tissues by customizing the computed tomography (CT) value opacity color curve so as to assign different brightness and colors. MPVR provides detailed visualization of the structural features of pulmonary nodules, thereby enhancing image readability and the ability to identify lesions. Additionally, a recently published article has shown a significant correlation between the size of the solid component measured by MPVR and the invasiveness of lung adenocarcinoma (12).

Radiomics is a method that analyzes lesions by extracting advanced features from CT images, and it has been widely studied for its potential in differential diagnosis and prognostic prediction (13,14). Research shows that radiomic features extracted from CT images can effectively predict whether SSPNs are benign or malignant (15,16). However, previous studies on predicting SSPNs have mainly focused on changes within the tumor itself and have overlooked the potential value of the surrounding area. The tumor microenvironment is enriched with a large number of tumor-infiltrating lymphocytes and tumor-associated macrophages, which are closely associated with tumor invasiveness and patient prognosis (17,18). However, no studies have yet explored the value of the peri-nodular region in differentiating benign from malignant subcentimeter solid nodules.

The aim of this study is to develop, validate, and compare radiomic models based on HRCT images of the tumor itself, the surrounding area, and various combinations of these approaches. The best radiomic model will then be integrated with independent clinical predictors to construct a nomogram. The goal is to enhance the model’s ability to differentiate between benign and malignant SSPNs. We present this article in accordance with the TRIPOD reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-618/rc).

Methods

Patients

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This retrospective study received approval from the Institutional Review Board of The First Affiliated Hospital of Chongqing Medical University (No. K2040-057-01). Due to the study’s retrospective nature, the requirement for informed patient consent was waived. This study retrospectively analyzed data from patients with SSPNs identified in chest CT scans between June 2020 and February 2024. Inclusion criteria were as follows: (I) SSPNs <1 cm in diameter; (II) chest CT scans with a slice thickness (≤1 mm); (III) <1-month time interval between CT examination and surgery; (IV) patients did not receive therapy preoperatively, and (V) SSPNs surgically resected and pathologically confirmed by pathological examination. Exclusion criteria included: (I) multifocal lesions; (II) calcified nodules; (III) patients with a history of malignant tumor; and (IV) chest CT imaging with significant artifacts and/or insufficient thin-layer reconstruction quality for analysis. The workflow of this study is illustrated in Figure 1.

Figure 1 The flowchart of inclusion and exclusion criteria. CT, computed tomography; HRCT, high-resolution computed tomography; SSPN, subcentimeter solid pulmonary nodule.

Image acquisition

All chest CT examinations were performed using a 128-MDCT scanner (Somatom Definition Flash; Siemens Healthcare, Erlangen, Germany) at the end of inspiration during a single breath-hold. All CT acquisitions were scanned from the lung apex to the base. Imaging parameters were as follows: 110–120 kVp tube voltage, 50–150 mAs tube current with automatic exposure control technology, 0.5 seconds rotation time, 5/5 mm image slice thickness and slice interval, 0.6×64 mm detector collimation, 1.0 pitch; and caudocranial scan direction. All images were reconstructed with 1 mm slice thickness and 512×512 matrix, and a lung algorithm for the lung window image. Independent evaluations of different series of images were performed at a Picture Archiving and Communication System (PACS) workstation (Vue PACS, Carestream, version 12.2.6.3000020) with standard lung window settings [width, 1600 Hounsfield units (HU); level, −600 HU] for all post-processing groups, but the observers were allowed to moderately adjust the window setting to keep with their normal workflow. MPVR is integrated into Carestream Vue PACS. As for MPVR, the CT values threshold of −800 HU was set to differentiate lung nodules and vessels from the lung background on the detection view of MPVR images, and the CT value threshold of −350 HU was set to differentiate the solid component and ground-glass background, which was also adapted from the literature (12).

Clinical and CT characteristics evaluation

The demographic data were obtained from the patients’ electronic medical records and PACS, including gender and age. In addition, two radiologists specialized in the CT diagnosis of lung disease recorded 12 radiological features without knowledge of histopathological results. These features included nodule location, lung window-maximum diameter, MPVR-maximum diameter, volume, shape, border, margin, vascular bundle, vacuolar, lobulation, spiculation, and pleural indentation, 12 imaging features in total. All CT data were extracted from the PACS workstation and subjected to a comprehensive analysis.

The CT radiological features included in this study are defined as follows—(I) nodule location: the anatomical position of the nodule within lung is noted; (II) lung window-maximum diameter: the lesion is observed on the lung window, and the maximum diameter is measured in coronal, axial, and sagittal planes, with the maximum value selected; (III) MPVR-maximum diameter: the maximum diameter of the nodule is measured in coronal, axial, and sagittal planes on the MPVR, selecting the maximum value; (IV) volume: nodule outlines are manually traced layer by layer on thin-section images, with volume measured automatically using the PACS workstation; (V) shape: benign nodules typically display circular, almost circular, or polygonal shapes with flat edges, malignant nodules are more often irregular in shape; (VI) border: the interface between the nodule and surrounding tissue is assessed; (VII) margin: benign nodules usually have smooth, well-defined margins, while malignant nodules have irregular margins; (VIII) vascular bundle: one or more blood vessels converging towards the interior of the lesion along its axis, potentially displaying irregular thickening or distortion; (IX) vacuole: the presence of normal lung tissue containing air—unoccupied by tumor tissue—commonly occurs in early lung adenocarcinoma; (X) lobulation: the tumor grows at different rates in various directions or is impeded by surrounding structures, resulting in multiple arc-shaped protrusions with alternating concave incisions, thereby creating a lobulated appearance; (XI) spiculation: the infiltrative growth of tumors and the resultant exudative or proliferative stromal reaction, indicating active tumor cell growth and obstruction of connective tissue; (XII) pleural indentation: the contraction of the pleura due to fibrosis within the tumor. To provide more accurate diagnostic basis for clinical practice, the same reader should analyze different images of the same CT study with a minimum interval of 2 weeks to avoid memory effects.

Pathologic evaluation

All pathological specimens were obtained after surgical resection. The pathological diagnosis and classification were primarily based on the lung adenocarcinoma classification standards published by the World Health Organization (WHO) in 2021 (19). All samples were fixed in 10% neutral formalin, and the cut sections were embedded in paraffin, sliced using a microtome, and stained with hematoxylin and eosin (H&E), along with the Alcian blue-periodic acid-Schiff technique to assess the production of cytoplasmic mucus. H&E stained sections that were randomly selected were reviewed. If there was uncertainty in the histopathological classification under optical microscopy, immunohistochemical testing was performed for further confirmation. Two pathologists, who were unaware of the clinical outcomes, independently reviewed the results according to the WHO classification, 5th edition (using a Leica DM3000 microscope with a standard eyepiece of 22 mm in diameter) (20).

ROI segmentation and peritumoral region generation

To enhance the reproducibility and repeatability of radiomic features, the data underwent a preprocessing phase. Z-score normalization was applied to the data; all images were resampled to a resolution of 1×1×1 mm³. Voxel intensity values were discretized with a fixed width of 25 units to reduce image noise, thereby maintaining a consistent intensity resolution across all tumor images. Manual 3D segmentation was performed using ITK-SNAP software (ITK-SNAP 3.8.0, www.itksnap.org). During contouring, care was taken to avoid artifacts due to necrosis, calcification, cavitation, vascularization, and bronchial structures. The delineated regions of interest (ROIs) were stored in Nifti format for further analysis. Using the OnekeyAI platform’s mask filling toolkit, the manually drawn ROIs were systematically expanded. The specific procedure involved radial intervals of 3 and 5 mm, with manual adjustments made as necessary to systematically evaluate the impact of different peritumoral distances on the model’s predictive accuracy. Figure 2 shows the segmentation of ROI and the generation of peritumoral regions.

Figure 2 A 57-year-old female patient with a malignant SSPN. (A) 1 mm axial thin-slice CT view. (B) MPVR nodule detection view. (C) Vascular view using MPVR, where the solid components within the pulmonary nodule are shown in red, consistent with the color of pulmonary vessels. ROI segmentation of the SSPN and generation of the peritumoral region. (D) Internal region of the nodule (red). (E) Internal nodule region (red) and the peritumoral region within 3 mm (green). (F) Transition from the internal nodule region to the peritumoral region within 3 mm (red). (G-I) Three-dimensional visualization of the ROI generated using ITK-SNAP software. CT, computed tomography; MPVR, multiplanar volume rendering; ROI, region of interest; SSPN, subcentimeter solid pulmonary nodule.

Feature extraction and selection

In this study, researchers utilized Pyradiomics software (3.0.1) to extract 1,835 features for each ROI, which included various indicators categorized into geometric features, intensity features, and texture features.

The feature screening process was carried out in the following steps: first, the study performed interobserver agreement analysis, setting an intraclass correlation coefficient (ICC) greater than 0.85 as the screening threshold to eliminate features with insufficient stability or poor reproducibility. Subsequently, each feature underwent z-score normalization to conform to a standard normal distribution (21). Then, statistical analysis was conducted using Spearman’s rank correlation coefficient to measure the correlation between two variables; if the Spearman correlation coefficient between features exceeded 0.9, only one feature was retained. Next, the study employed the minimum redundancy maximum relevance (mRMR) algorithm for feature selection to reduce overfitting. Finally, the least absolute shrinkage and selection operator (LASSO) regression model was used to determine the final feature set.

Radiomics signatures and clinical model

In this study, multiple radiomic features were extracted from different regions: intra-nodular (Intra), peri-nodular 3 mm (Peri3mm), peri-nodular 5 mm (Peri5mm), intra-nodular to peri-nodular 3 mm (imagefusion3mm), and intra-nodular to peri-nodular 5 mm (imagefusion5mm). Additionally, two radiomic models were constructed by fusing features from the Peri3mm and Peri5mm with intra regions: IntraPeri3mm and IntraPeri5mm. A clinical model was constructed based on independent risk factors identified through univariate and multivariate logistic regression (LR) analysis. To build predictive models, we utilized several common machine learning algorithms, including LR, support vector machines (SVMs), K-nearest neighbors (KNNs), and light gradient boosting machine (LightGBM). The optimal hyperparameters for each model were determined using five-fold cross-validation and grid search algorithms to ensure performance and stability. The workflow of the necessary steps for constructing the radiomic models is illustrated in Figure 3.

Figure 3 Radiomics workflow. AUC, area under the curve; CI, confidence interval; KNN, K-nearest neighbor; LightGBM, light gradient boosting machine; LR, logistic regression; MPVR, multiplanar volume rendering; MSE, mean squared error; ROI, region of interest; SVM, support vector machine.

Development and validation of the nomogram

This clinical model was built using LR machine learning algorithms. By integrating clinical features with the best-performing model from the test set, a nomogram was developed. The diagnostic performance of the models was evaluated using receiver operating characteristic (ROC) curve analysis, with the area under the curve (AUC) as the primary metric. DeLong tests were used to compare the AUC values between different models for significant differences. Sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were calculated across risk thresholds. Sensitivity represented the proportion of correctly identified malignant SSPNs among all malignancies [true positives/(true positives + false negatives)]. Specificity measured the proportion of correctly classified benign lesions among benign cases [true negatives/(true negatives + false positives)]. In addition, calibration curve analysis and decision curve analysis (DCA) were used to evaluate the goodness of fit and clinical relevance.

Model evaluation and statistical analysis

To compare clinical characteristics between different patient groups, we employed the following statistical methods. For continuous variables, independent samples t-tests were used for comparisons between two groups; if the t-test assumptions were not met, Mann-Whitney U tests were performed for non-parametric analysis. For discrete variables, χ² tests were used for comparisons. For categorical variables, counts and percentages were reported, and comparisons were made using χ² tests or Fisher’s exact tests, depending on the dataset’s suitability. Univariate LR analysis was conducted to identify predictive factors associated with prognostic risk, with a significance threshold set at P<0.05. Subsequently, multivariate LR analysis was performed to determine the independent factors that were significant in the univariate analysis, also at P<0.05. Results were presented as odds ratios (ORs) with their 95% confidence intervals (CIs) and corresponding P values. All statistical analyses were performed using R (version 3.4.0, R Foundation) to ensure reproducibility and reliability of the results. All statistical tests were two-sided, with a significance level of P<0.05.

Results

Basic data

A total of 415 patients were included, with 290 in the training set (140 benign, 150 malignant) and 125 in the test set (60 benign, 65 malignant). Table 1 summarizes the clinical and radiologic characteristics of the patients in the training and test sets. Specifically, age, lung window-maximum diameter, MPVR-maximum diameter, shape, margin, border, vascular bundle, vacuole, and pleural indentation were significant predictors of malignancy (P<0.05). Univariate and multivariate analysis of clinical features were performed in the training set, and the OR and corresponding P values for each feature were calculated (Table 2). Univariate LR analysis revealed significant differences between the malignant and benign groups in terms of nodule margin, shape, pleural indentation, MPVR-maximum diameter, vascular bundle, border, and vacuole (P<0.05). However, age, gender, location, volume, lobulation, and spiculation showed no significant differences. Multivariate analysis further confirmed that margin (OR =0.578; 95% CI: 0.367–0.910) and MPVR-maximum diameter (OR =1.175; 95% CI: 1.103–1.252) were independently associated with the occurrence of malignant tumors.

Feature selection and construction of radiomic models

The LASSO method was employed to select radiomic features with non-zero coefficients, using the optimal λ value. The coefficients of the selected features for each model are illustrated in Figure S1. With 14 radiomic features analyzed, the current sample size meets the rigorous criterion of at least 10 events per predictor, ensuring robust machine learning model development and valid statistical inference. After feature selection, the best machine learning technique for constructing the radiomic model was determined through five-fold cross-validation.

This study incorporated the machine learning algorithm that achieved the highest AUC value in the testing set. A comparison of the ROC curves across different machine learning algorithms can be found in Figure S2. Through comparative analysis, we concluded that the radiomic features Intra, Peri5mm, Imagefusion5mm, IntraPeri3mm, and IntraPeri5mm were best suited for model construction using LightGBM. In contrast, the radiomic features Peri3mm and Imagefusion3mm were more appropriate for model construction using SVM.

Performance and comparison of different models

Margin and MPVR-maximum diameter were used to construct the clinical model. This clinical model demonstrated strong predictive capability in both the training and testing sets, achieving AUC values of 0.892 (95% CI: 0.854–0.930) and 0.875 (95% CI: 0.804–0.945), respectively. These results indicate that MPVR help to differentiate the malignancy of SSPNs. In the testing set, the IntraPeri3mm feature performed best, with an AUC of 0.891 (95% CI: 0.828–0.954). Figure 4 provides a detailed comparison of the ROC curves for clinical and radiomic feature predictive performances. For a thorough analysis of calibration curves and DCA for clinical and radiomic features, please refer to Figures S3,S4. Comparisons of the AUC of different models using the DeLong test revealed no significant differences in model performance among the features (P>0.05; Figure 4).

Figure 4 The ROC curves (A,B) and Delong test heatmaps (C,D) of the different models in the training cohort and testing cohort. AUC, area under the curve; CI, confidence interval; ROC, receiver operating characteristic.

We combined independent clinical predictors with the optimal IntraPeri3mm features identified in the testing set to develop a nomogram (Figure 5). Figure 6 summarizes the performance of the nomogram. The nomogram showed the highest predictive accuracy in both the training and testing sets, with AUC values of 0.965 (95% CI: 0.946–0.985) and 0.966 (95% CI: 0.938–0.995), respectively. This performance was significantly better than the clinical features or IntraPeri3mm features alone (P<0.05). Table 3 presents the accuracy, sensitivity, specificity, NPV, and PPV for all features. The calibration curve analysis indicated that the probabilities of benign and malignant SSPNs assessed by the nomogram demonstrated the best alignment with actual probabilities (Figure 7). DCA further confirmed that the nomogram provided the best overall net benefit for distinguishing between benign and malignant SSPNs, outperforming other models across a wide range of threshold values (Figure 8).

Figure 5 Comprehensive nomogram for distinguishing benign and malignant in SSPNs (margin: 0 means smooth, well-defined margins, 1 means irregular margins). MPVR, multiplanar volume rendering; SSPN, subcentimeter solid pulmonary nodule.

Figure 6 The ROC curves of the Clinic, IntraPeri3mm and combined models in the training cohort (A) and testing cohort (B). AUC, area under the curve; CI, confidence interval; ROC, receiver operating characteristic.

Figure 7 The calibration curves of the Clinic, IntraPeri3mm and combined models in the training cohort (A) and testing cohort (B).

Figure 8 The DCA of the Clinic, IntraPeri3mm and combined models in the training cohort (A) and testing cohort (B). DCA, decision curve analysis.

Discussion

Distinguishing benign from malignant SSPNs is challenging and significantly impacts clinical decision-making. Increasingly, radiomics has shown substantial potential in achieving this goal (22-25). Accurate prediction of the benign or malignant of SSPNs is crucial for establishing an early prevention, diagnosis, and treatment system (26). This study retrospectively analyzed 415 patients who underwent surgical resection for SSPNs at The First Affiliated Hospital of Chongqing Medical University. We propose a radiomics approach combined with MPVR derived from HRCT to differentiate benign and malignant SSPNs. Independent risk factors were identified through multivariate analysis, and radiomic features were extracted from both the intra-nodular and peri-nodular regions of the CT images. Multiple machine learning models were developed to enhance the differentiation capability of the nodules. The IntraPeri3mm model performed best in the test set, achieving an AUC of 0.891, demonstrating excellent predictive performance. Furthermore, an integrated nomogram was constructed by combining MPVR-maximum diameter and margin. DCA and calibration curve analysis results indicate that the nomogram offers high net benefit and calibration in clinical practice, providing significant non-invasive support for the evaluation of SSPNs.

In contrast to the indolent pathological behavior of subsolid nodules, SSPNs tend to grow more rapidly and are more likely to be IAC (4). Our study found that margin is predictive factor for differentiating between benign and malignant SSPNs. Azour et al. showed that nodule shape, margin, spiculation, and pleural retraction can be used to determine the nature of nodules (27). However, these signs are subjective and often not apparent when nodules are small (28). In addition, the study by de Morais et al. demonstrated that malignant tumors frequently exhibit irregular margins compared to benign SSPNs with regular margins (29). Our findings also demonstrate the importance of margin in differentiating between benign and malignant SSPNs. Therefore, the performance of margin in predicting the malignancy of SSPNs requires further evaluation through additional non-invasive examinations. Additionally, our study used MPVR, which is a 3D visualization technology. A previous study has shown that the size of the solid component measured by MPVR is significantly correlated with the invasiveness of pulmonary adenocarcinoma (12). 3D observation of nodule morphology and density features can enhance the accuracy of imaging-pathological diagnoses and optimize clinical decision-making. This supports our view that MPVR is an independent risk factor. The clinical model incorporating MPVR demonstrated encouraging performance, with AUCs of 0.892 (95% CI: 0.854–0.930) in the training set and 0.875 (95% CI: 0.804–0.945) in the testing set. These results indicate that MPVR provides crucial evidence for differentiating benign and malignant SSPNs, assisting doctors early diagnosis of malignant nodules.

The small size of SSPNs often means they lack clear morphological features (30,31) making it challenging for observers to reach a consensus in diagnosis (32). Therefore, assessing the invasiveness of SSPNs based solely on visible characteristics is particularly difficult. Radiomics can extract detailed data from CT images, providing a more objective view of tumor heterogeneity. In this study, we applied radiomics to analyze CT images to develop a more accurate predictive model for differentiating benign and malignant SSPNs. Previous studies have investigated the use of CT-based radiomics in predicting the malignancy of SSPNs (33-35). However, these studies primarily focused on extracting intratumoral features, often neglecting the importance of peritumoral tissues. In contrast, our study considered the potential influence of the surrounding tumor region. Wu et al. evaluated the ability of CT radiomic features extracted from 2 and 5 mm peritumoral margins to differentiate IAC from adenocarcinoma in situ (AIS) and minimally invasive adenocarcinoma (MIA). Their findings indicated that both 2 mm and 5 mm peritumoral radiomic models showed good predictive performance, but radiomic features from 5 mm peritumoral lung parenchyma did not significantly enhance the prediction of lung adenocarcinoma invasiveness (31). Xu et al. extracted radiomic features from both intratumoral and peritumoral regions to predict high-grade patterns of IAC. They found that the radiomic model combining intratumoral and 3 mm peritumoral regions performed better compared to models built from intratumoral, intratumoral to 6 mm peritumoral, and intratumoral to 9 mm peritumoral regions (36). These studies highlight the critical role of the tumor microenvironment in predicting tumor invasiveness and classification, and suggest that indiscriminately expanding the peritumoral region does not necessarily enhance model effectiveness. As the distance from the tumor increases, the volume of interest (VOI) contains more normal lung tissue and less tumor tissue, ultimately reducing the model’s predictive performance (15). Previous research has shown that there is a transition zone between lung adenocarcinoma and normal lung tissue, typically ranging from 1 mm to 5 mm, with an average of about 3.5 mm (37,38). Therefore, we focused on the 3 mm and 5 mm peritumoral regions. Our results indicate that radiological features of the peritumoral region have potential predictive power in distinguishing the malignancy of SSPNs. Both Peri3mm and Peri5mm radiomic features performed well. Specifically, the combined model of Peri3mm and intratumoral features (IntraPeri3mm) exhibited the best performance in the testing set, with an AUC of 0.891. Integrating intratumoral and peritumoral radiomics effectively distinguishes the malignancy of SSPNs.

Recently, an increasing number of studies have developed nomograms to improve clinical decision-making, thereby enabling more precise and personalized cancer treatment strategies. Previous research has shown that nomograms are more effective than single prediction models. Chen et al. developed a nomogram based on CT imaging and clinical features to predict the benign or malignant SSPNs, achieving AUCs of 0.930 and 0.905 in the training and testing sets, respectively (33). Similarly, Lin et al. retrospectively analyzed 180 SSPNs and created a clinical-radiomic nomogram, which AUCs of 0.940 and 0.903 in the training and testing sets, respectively (34). In contrast, Liu et al. used CT enhancement images to develop a clinical-radiomic nomogram, achieving AUCs of 0.942 and 0.930 in the training and testing sets, respectively, outperforming previous studies (35). In our study, we integrated the IntraPeri3mm radiomic features with margin and MPVR-maximum diameter to create a nomogram. This nomogram demonstrated superior discrimination performance in both the training set (AUC =0.965) and the testing set (AUC =0.966), exceeding previous studies. Compared to prior research, our study included the largest sample size to date. Additionally, we incorporated the innovative imaging method MPVR and combined intratumoral and peritumoral information to construct radiomics models, thereby capturing the spatial information of SSPNs more comprehensively and enhancing the accuracy and reliability of the diagnostic process.

We recognize several limitations in our study. First, it is a single-center retrospective study, and incomplete patient data led to a relatively small sample size. In future work, the predictive model should be validated in larger, prospective, multicenter studies. Second, the CT images were obtained from a single scanner, and radiomics can show significant differences both within and between scanners; therefore, further validation of the model’s generalizability is needed. Third, the radiomics analysis model employed in this study is solely based on HRCT images. It is anticipated that integrating supplementary imaging modalities, such as contrast-enhanced CT, contrast-enhanced MRI, and functional imaging, will further augment the predictive value of radiomics analyses. Finally, the segmentation of SSPNs was done semi-automatically, with lesion boundaries confirmed manually, which is time-consuming. However, the current gold standard for image segmentation relies on manual delineation by experienced radiologists. More advanced artificial intelligence (AI)-based automatic segmentation algorithms need to be developed and implemented to improve efficiency and reduce subjective inconsistencies. Combining radiomics data with other multi-scale and multimodal inputs is an emerging field that can provide clearer biological interpretations of tumor heterogeneity and treatment outcomes. Our future research will investigate the relationships between radiomics, pathology, and genomics in SSPNs to offer better biological insights for patients.

Conclusions

This study confirmed that the MPVR-maximum diameter is an independent predictor of the malignancy of SSPNs. By comparing various radiomics models based on HRCT images, we found that the nomogram combining IntraPeri3mm model and the MPVR-maximum diameter had the best predictive efficacy. These results provide an important scientific basis for the early diagnosis and treatment of SSPNs.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-618/rc

Data Sharing Statement: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-618/dss

Peer Review File: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-618/prf

Funding: This work was supported by the National Key R&D Program of China (Grant No. 2020YFA0714002), Chongqing Health Appropriate Technology Promotion Project (No. 2023jstg044), the Key Project of Technological Innovation and Application Development of Chongqing Science and Technology Bureau (No. CSTC2021jscx-ksbN0030), and the Joint Project of Chongqing Health Commission and Science and Technology Bureau (No. 2022ZDXM006).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-618/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This retrospective study received approval from the Institutional Review Board of The First Affiliated Hospital of Chongqing Medical University (No. K2040-057-01). Due to the study’s retrospective nature, the requirement for informed patient consent was waived.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Bray F, Laversanne M, Sung H, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2024;74:229-63. [Crossref] [PubMed]
2. Huang S, Yang J, Shen N, et al. Artificial intelligence in lung cancer diagnosis and prognosis: Current application and future perspective. Semin Cancer Biol 2023;89:30-7. [Crossref] [PubMed]
3. Yang W, Sun Y, Fang W, et al. High-resolution Computed Tomography Features Distinguishing Benign and Malignant Lesions Manifesting as Persistent Solitary Subsolid Nodules. Clin Lung Cancer 2018;19:e75-83. [Crossref] [PubMed]
4. Ye T, Deng L, Wang S, et al. Lung Adenocarcinomas Manifesting as Radiological Part-Solid Nodules Define a Special Clinical Subtype. J Thorac Oncol 2019;14:617-27. [Crossref] [PubMed]
5. Sun K, You A, Wang B, et al. Clinical T1aN0M0 lung cancer: differences in clinicopathological patterns and oncological outcomes based on the findings on high-resolution computed tomography. Eur Radiol 2021;31:7353-62. [Crossref] [PubMed]
6. Li J, Xia T, Yang X, et al. Malignant solitary pulmonary nodules: assessment of mass growth rate and doubling time at follow-up CT. J Thorac Dis 2018;10:S797-806. [Crossref] [PubMed]
7. Gao J, Qi Q, Li H, et al. Artificial-intelligence-based computed tomography histogram analysis predicting tumor invasiveness of lung adenocarcinomas manifesting as radiological part-solid nodules. Front Oncol 2023;13:1096453. [Crossref] [PubMed]
8. Wolf AMD, Oeffinger KC, Shih TY, et al. Screening for lung cancer: 2023 guideline update from the American Cancer Society. CA Cancer J Clin 2024;74:50-81. [Crossref] [PubMed]
9. de Koning HJ, van der Aalst CM, de Jong PA, et al. Reduced Lung-Cancer Mortality with Volume CT Screening in a Randomized Trial. N Engl J Med 2020;382:503-13. [Crossref] [PubMed]
10. Yankelevitz DF, Yip R, Henschke CI. Impact on Prognosis of Stage I Non-Small Cell Lung Cancer Secondary to Delays in Diagnostic Workup. Radiology 2024;313:e240420. [Crossref] [PubMed]
11. Zhao W, Xiong Z, Jiang Y, et al. Radiomics based on enhanced CT for differentiating between pulmonary tuberculosis and pulmonary adenocarcinoma presenting as solid nodules or masses. J Cancer Res Clin Oncol 2023;149:3395-408. [Crossref] [PubMed]
12. Li WJ, Chu ZG, Li D, et al. Accuracy of solid portion size measured on multiplanar volume rendering images for assessing invasiveness in lung adenocarcinoma manifesting as subsolid nodules. Quant Imaging Med Surg 2024;14:1971-84. [Crossref] [PubMed]
13. Tang FH, Xue C, Law MY, et al. Prognostic Prediction of Cancer Based on Radiomics Features of Diagnostic Imaging: The Performance of Machine Learning Strategies. J Digit Imaging 2023;36:1081-90. [Crossref] [PubMed]
14. Kashyap M, Wang X, Panjwani N, et al. Automated Deep Learning-Based Detection and Segmentation of Lung Tumors at CT. Radiology 2025;314:e233029. [Crossref] [PubMed]
15. Tu SJ, Wang CW, Pan KT, et al. Localized thin-section CT with radiomics feature extraction and machine learning to classify early-detected pulmonary nodules from lung cancer screening. Phys Med Biol 2018;63:065005. [Crossref] [PubMed]
16. Pan F, Feng L, Liu B, et al. Application of radiomics in diagnosis and treatment of lung cancer. Front Pharmacol 2023;14:1295511. [Crossref] [PubMed]
17. Noritake O, Aokage K, Suzuki A, et al. Prognostic impact of the number of peri-tumoral alveolar macrophages in patients with stage I lung adenocarcinoma. J Cancer Res Clin Oncol 2022;148:3437-47. [Crossref] [PubMed]
18. Wang S, Yu H, Gan Y, et al. Mining whole-lung information by artificial intelligence for predicting EGFR genotype and targeted therapy response in lung cancer: a multicohort study. Lancet Digit Health 2022;4:e309-19. [Crossref] [PubMed]
19. Nicholson AG, Tsao MS, Beasley MB, et al. The 2021 WHO Classification of Lung Tumors: Impact of Advances Since 2015. J Thorac Oncol 2022;17:362-87. [Crossref] [PubMed]
20. Eguchi T, Adusumilli PS. Risk stratification for lung nodules: Size isn't everything. J Thorac Cardiovasc Surg 2017;153:1557-62. [Crossref] [PubMed]
21. Beaumont H, Iannessi A, Bertrand AS, et al. Harmonization of radiomic feature distributions: impact on classification of hepatic tissue in CT imaging. Eur Radiol 2021;31:6059-68. [Crossref] [PubMed]
22. Li H, Luo Q, Zheng Y, et al. A nomogram for predicting invasiveness of lung adenocarcinoma manifesting as ground-glass nodules based on follow-up CT imaging. Transl Lung Cancer Res 2024;13:2617-35. [Crossref] [PubMed]
23. Zhang J, Hao L, Qi M, et al. Radiomics nomogram for preoperative differentiation of pulmonary mucinous adenocarcinoma from tuberculoma in solitary pulmonary solid nodules. BMC Cancer 2023;23:261. [Crossref] [PubMed]
24. Wang C, Shao J, He Y, et al. Data-driven risk stratification and precision management of pulmonary nodules detected on chest computed tomography. Nat Med 2024;30:3184-95. [Crossref] [PubMed]
25. Yi L, Peng Z, Chen Z, et al. Identification of pulmonary adenocarcinoma and benign lesions in isolated solid lung nodules based on a nomogram of intranodal and perinodal CT radiomic features. Front Oncol 2022;12:924055. [Crossref] [PubMed]
26. Zhang CR, Wang Q, Feng H, et al. Computed-tomography-based radiomic nomogram for predicting the risk of indeterminate small (5-20 mm) solid pulmonary nodules. Diagn Interv Radiol 2023;29:283-90. [Crossref] [PubMed]
27. Azour L, Ko JP, Washer SL, et al. Incidental Lung Nodules on Cross-sectional Imaging: Current Reporting and Management. Radiol Clin North Am 2021;59:535-49. [Crossref] [PubMed]
28. Tang J, Liu C, Wang P, et al. Multivariate Analysis of Solid Pulmonary Nodules Smaller than 1 cm in Distinguishing Lung Cancer from Intrapulmonary Lymph Nodes. Zhongguo Fei Ai Za Zhi 2021;24:94-8. [Crossref] [PubMed]
29. de Morais AQ, da Silva TPF, Braga JCD, et al. Factors associated with subcentimeter pulmonary nodule outcomes followed with computed tomography imaging in oncology patients. Eur J Radiol Open 2020;7:100266. [Crossref] [PubMed]
30. Chen W, Li M, Mao D, et al. Radiomics signature on CECT as a predictive factor for invasiveness of lung adenocarcinoma manifesting as subcentimeter ground glass nodules. Sci Rep 2021;11:3633. [Crossref] [PubMed]
31. Wu F, Tian SP, Jin X, et al. CT and histopathologic characteristics of lung adenocarcinoma with pure ground-glass nodules 10 mm or less in diameter. Eur Radiol 2017;27:4037-43. [Crossref] [PubMed]
32. Fletcher JG, Levin DL, Sykes AG, et al. Observer Performance for Detection of Pulmonary Nodules at Chest CT over a Large Range of Radiation Dose Levels. Radiology 2020;297:699-707. [Crossref] [PubMed]
33. Chen C, Geng Q, Song G, et al. A comprehensive nomogram combining CT-based radiomics with clinical features for differentiation of benign and malignant lung subcentimeter solid nodules. Front Oncol 2023;13:1066360. [Crossref] [PubMed]
34. Lin RY, Zheng YN, Lv FJ, et al. A combined non-enhanced CT radiomics and clinical variable machine learning model for differentiating benign and malignant sub-centimeter pulmonary solid nodules. Med Phys 2023;50:2835-43. [Crossref] [PubMed]
35. Liu J, Qi L, Wang Y, et al. Development of a combined radiomics and CT feature-based model for differentiating malignant from benign subcentimeter solid pulmonary nodules. Eur Radiol Exp 2024;8:8. [Crossref] [PubMed]
36. Xu J, Liu L, Ji Y, et al. Enhanced CT-Based Intratumoral and Peritumoral Radiomics Nomograms Predict High-Grade Patterns of Invasive Lung Adenocarcinoma. Acad Radiol 2025;32:482-92. [Crossref] [PubMed]
37. Zhang L, Yankelevitz DF, Henschke CI, et al. Zone of transition: a potential source of error in tumor volume estimation. Radiology 2010;256:633-9. [Crossref] [PubMed]
38. Saeed N, Shridhar R, Almhanna K, et al. CT-based assessment of visceral adiposity and outcomes for esophageal adenocarcinoma. J Gastrointest Oncol 2017;8:833-41. [Crossref] [PubMed]