Skip to main content
Download PDF

Development of a nomogram-based model incorporating radiomic features from follow-up longitudinal lung CT images to distinguish invasive adenocarcinoma from benign lesions: a retrospective study

BMC Pulmonary Medicine volume 24, Article number: 534 (2024) Cite this article

Abstract

Purpose

To develop and validate a radiomic model for differentiating pulmonary invasive adenocarcinomas from benign lesions based on follow-up longitudinal CT images.

Methods

This is a retrospective study including 336 patients (161 with invasive adenocarcinomas and 175 with benign lesions) who underwent baseline (T0) and follow-up (T1) CT scans from January 2016 to June 2022. The patients were randomized in a 7:3 ratio into training and test sets. Radiomic features were extracted from lesion volumes of interest on longitudinal CT images at T0 and T1. Differences in radiomic features between T1 and T0 were defined as delta-radiomic features. Logistic regression was used to build models based on clinicoradiological (CR), T0, T1, and delta radiomic features and compute signatures. Finally, a nomogram based on the CR, T0, T1 and delta signatures was constructed. Model performance was evaluated for calibration, discrimination, and clinical utility.

Results

The T1 radiomic model was superior to the other independent models. In the training set, it had an area under the curve (AUC) of 0.858), superior to the CR model (AUC 0.694), the T0 radiomic model (AUC 0.825), and the delta radiomic model (AUC 0.734). In the test set, it had an AUC of 0.817, again outperforming the CR model (AUC 0.578), the T0 radiomic model (AUC 0.789), and the delta radiomic model (AUC 0.647). The nomogram incorporating the CR, T0, T1 and delta signatures showed the best predictive performance in both the training (AUC: 0.906) and test sets (AUC: 0.856), and it exhibited excellent fit with calibration curves. Decision curve analysis provided additional validation of the clinical utility of the nomogram.

Conclusion

A nomogram utilizing radiomic features extracted from longitudinal CT images can enhance the discriminative capability between pulmonary invasive adenocarcinomas and benign lesions.

Peer Review reports

Introduction

The detection rate of lung nodules has greatly improved thanks to the widespread adoption of lung cancer screening and advancements in computed tomography (CT) technology [1]. While CT imaging can readily detect nodules as small as 2–3 mm in diameter, the clinical challenge lies in accurately differentiating between malignant and benign nodules, as the two have no distinct CT characteristics. Treatment strategies for malignant and benign nodules exhibit significant differences, and benign nodules have a better prognosis, necessitating accurate differentiation. Various organizations, such as the American College of Chest Physicians [2], the British Thoracic Society [3], and the American Fleischner Society [4, 5], have established guidelines for managing pulmonary nodules. Although guidelines recommend regular follow-up for most indeterminate nodules, excessive testing, additional costs, and unnecessary use of medical resources can result from multiple follow-ups that exacerbate patient anxiety [6]. Besides lung nodules cannot be identified as benign or malignant morphologically in the early stage. Moreover, in reality, the omission of early lung adenocarcinoma is already in the late stage when it is discovered, which lags behind clinical decision-making for lung adenocarcinoma [7]. Lung nodules are identified as malignant at an early stage, there is a remarkable improvement in the survival rate [8]. Therefore, it is crucial to quickly determine whether a nodule is benign or malignant.

Radiomics aims to obtain automated quantitative imaging signatures that can predict tumour behaviour noninvasively [9, 10]. Numerous CT radiomics studies on pulmonary nodules [11,12,13,14] have demonstrated the potential of radiomics in enhancing nodule assessment accuracy. However, these studies were limited to cross-sectional analysis and only examined radiomic features from single-time CT imaging, lacking a description of the temporal changes in the radiomic characteristics of pulmonary nodules. Most studies [15,16,17] have indicated that cancer is a dynamic disease characterized by increasing heterogeneity as it progresses. This heterogeneity [18] includes not only spatial heterogeneity but also temporal heterogeneity. Revealing the temporal and spatial variation features of tumours is imperative for accurately predicting their benign or malignant nature. Delta-radiomic features, which analyse changes in texture features obtained at two time points, can be utilized to investigate alterations in radiomic features of the same lesion over time [19]. Saeeds Alahmari et al. [20] demonstrated the potential use of delta radiomic features in screening pulmonary nodules for malignancy. To the best of our knowledge, there are no studies on radiomic models that integrate the baseline and follow-up CT data for differentiating pulmonary invasive adenocarcinomas from benign lesions. In this study, we developed a radiomic model for differentiating pulmonary invasive adenocarcinomas from benign lesions based on short-term follow-up longitudinal CT images.

Materials and methods

Subjects

This research was carried out following the guidelines of the Declaration of Helsinki. The institutional review board of our hospital granted approval for this study (2020-147-01). The clinical data, as well as the baseline and follow-up CT images (within a time interval of 3–6 months), were retrospectively collected from patients who underwent surgery for pulmonary nodules between January 2016 and January 2022 (Fig. 1).

The inclusion criteria were as follows: (1) confirmation of lung adenocarcinoma or benign lesions through surgical pathology; and (2) availability of baseline and follow-up CT images (within a time interval of 3–6 months) before surgery. The exclusion criteria were as follows: (1) lesion diameter exceeding 30 mm; (2) presence of more than one nodule in a single lung lobe; (3) poor quality chest CT images; and (4) previous or current history of malignancy. Ultimately, a total of 366 patients were enrolled in this study, comprising 161 patients with lung invasive adenocarcinoma (mean follow-up time: 129.52 days ± 27.81; 74 males and 87 females; aged 50.2 ± 11.59 years) and 175 patients with benign lesions (mean follow-up time: 136.01 days ± 37.64; 78 males and 97 females; aged 48.11 ± 10.36 years). The benign lesions included chronic inflammatory nodules (68, 38.9%), adenocarcinoma in situ (58, 33.1%), atypical adenomatous hyperplasia (13, 7.4%), adenoma (13, 7.4%), hamartoma (10, 5.7%), tuberculous granuloma (8, 4.6%), and sclerosing pneumocytomas (5, 2.9%). In the WHO histological classification of lung Tumors in 2021 [21], both atypical adenomatous hyperplasia and adenocarcinoma in situ were classified as glandular prodromal lesions for the first time, and they were excluded from lung adenocarcinoma.

The baseline CT scan was designated T0, while the follow-up CT scan (within a time interval of 3–6 months) was designated T1. General clinical information, including sex, age, smoking history, family history and tumour markers, was collected. The recorded tumour markers were carcinoembryonic antigen (CEA), cancer antigen 153 (CA153), neuron-specific enolase (NSE), squamous cell carcinoma-associated antigen (SCCA), cytokeratin fragment 21 − 1 (cyfra21-1), serum ferritin (SF), cancer antigen 125 (CA125) and progastrin-releasing peptide (ProGRP). The radiological morphological features of the lesions in the T0 and T1 CT images were independently assessed by two experienced chest radiologists (Z.M.W. and F.W., with 5 and 7 years of chest CT experience, respectively) without prior knowledge of the pathological results, including location, long diameter, short diameter, mean diameter, shape, lobulation, burr, pleural traction, air bronchogram, vacuole sign and vascular abnormalities [9]. The combined set of clinical and radiological morphological features are collectively referred to as clinicoradiological (CR) features.

Fig. 1
figure 1

Patient screening flowchart

Full size image

CT data acquisition

The CT scans were performed using the following scanners: GE Optima CT660 (USA), GE LightSpeed VCT (USA), and SOMATOM Definition Flash (Germany). Supplementary Table 1 provides detailed scan information. After deep inhalation, the patients held their breath while we acquired scans from the thoracic entrance to the lung base. The CT images were anonymized and exported in Digital Imaging and Communications in Medicine (DICOM) format through the Picture Archiving and Communication Systems (PACS).

ROI segmentation

After adjusting the window width to 1500 Hounsfield units (HUs) and setting the window level to -450 HU, the region of interest (ROI) of the lesions was semiautomatically segmented by a radiologist (Z.M.W.), who manually excluded large vessels, bronchi, and chest wall tissue. This segmentation process was performed using the open source software 3Dslicer (https://download.slicer.org). The entire lesion contour was delineated in a blinded manner without any knowledge of the pathological results, ensuring unbiased delineation. To assess observation consistency within and between observers, 30 random lesions were redrawn by both the above radiologist and another radiologist (F.W.) one month later.

Radiomic feature extraction and selection

The image was resampled to a spatial resolution of 1 mm × 1 mm × 1 mm before feature extraction. PyRadiomics was utilized for the extraction of radiomic features within ROIs. A total of 851 radiomic features were extracted for each ROI, including 107 original features and 744 wavelet features. We defined the difference in the radiomic features between T1 and T0 as the delta-radiomic features (Fig. 2A). Ultimately, a total of 2553 radiomic features (851 × 3) were obtained. Before feature selection, the radiomic features were normalized, and only features with an intraclass correlation coefficient (ICC) greater than 0.80 were retained. First, we employed the maximum correlation and minimum redundancy (mRMR) method to reduce dimensionality by eliminating redundant and irrelevant features while maximizing their correlation with classification variables. Second, we used least absolute shrinkage and selection operator (LASSO) with 5-fold cross-validation for further feature selection. Finally, stepwise forward multiple logistic regression was conducted to analyse the features, and only those with a significance level of p < 0.05 were selected for subsequent model establishment (Fig. 2B).

Model building and evaluation

Univariate logistic regression analysis was employed to evaluate how well the individual clinicoradiological features distinguished invasive adenocarcinomas from benign nodules in the training set. Those that were statistically significant were included in a multivariate logistic regression analysis to identify the final set of potential risk factors for invasive adenocarcinomas. The risk factors were utilized to construct a clinicoradiological (CR) model employing a logistic regression classifier, and then the corresponding CR signature was generated from the features’ weight coefficients.

The selected T0 radiomic features, T1 radiomic features and delta radiomic features were utilized to construct the T0 radiomic model (T0 model), T1 radiomic model (T1 model), and delta-radiomic model (delta model), respectively, using a logistic regression classifier. The corresponding signatures (T0-signature, T1-signature, and delta-signature) were constructed based on the coefficients of feature weights. The relevant feature names and formulas for calculating signatures are presented in Supplementary Table 2.

The CR signature, T0 signature, T1 signature, and delta signature were integrated into a logistic regression classifier to construct a predictive model, which was visualized as a nomogram.

Receiver operating characteristic (ROC) curves were plotted, and the area under each ROC curve (AUC) was used to quantify the discriminative efficacy of each model. Multiple comparisons of the curves were performed using the DeLong test with Bonferroni-adjusted p values. Additionally, we calculated the 95% confidence intervals (CIs) of the AUCs, as well as their sensitivity, specificity, and accuracy. To assess the calibration and goodness of fit of the nomogram, we drew a calibration curve and ran the Hosmer–Lemeshow test. Furthermore, decision curve analysis (DCA) was employed to evaluate the clinical utility of the nomogram (Fig. 2C).

Fig. 2
figure 2

Radiomics analysis process

Full size image

Statistical analysis

Statistical analyses were performed using IBM SPSS Statistics 26 (http://downloading IBM SPSS Statistics 26) and R4.2 software (http://www.r-project.org). Continuous variables are presented as mean ± standard deviation (SD) if normally distributed and were compared by the independent-sample t test. Otherwise, they were compared by the nonparametric Mann‒Whitney U test. The chi-square test was utilized to compare categorical variables. A bilateral p value < 0.05 was considered to indicate statistical significance. The “irr” and “mRMRe” packages were utilized for conducting the ICC and mRMR analyses, respectively. The “glmnet” package was employed for performing LASSO and constructing logistic regression models. The “rms” package was utilized to generate calibration curves, and the “pROC” package was employed for plotting ROC curves and calculating AUC values. The DeLong test was applied to compare AUCs, and the Hosmer–Lemeshow test was used to assess calibration performance.

Results

Clinicoradiological features

When we compared the clinicoradiological features between T0 and T1, no statistically significant differences were detected (all p > 0.05; Supplementary Table 3). The baseline T1 clinicoradiological features are summarized in Table 1. In the training set, there were significant differences between pulmonary invasive adenocarcinomas and benign lesions in the long diameter, short diameter, mean diameter, pleural traction, air bronchogram and vascular abnormality (p < 0.05). In the test set, significant differences were found in vascular abnormalities and pleural traction (p < 0.05).

Table 1 The clinicoradiological features of participants
Full size table

Construction and performance of the clinicoradiological model

Univariate logistic regression analysis revealed that long diameter, short diameter, mean diameter, pleural traction, air bronchogram presence, and vascular abnormalities were significantly related to adenocarcinoma (p < 0.05). According to the multivariate logistic regression analysis, long diameter (OR = 1.153; 95% CI: 1.076–1.235; p < 0.001) and vascular abnormalities (OR = 4.800; 95% CI: 1.763–13.067; p = 0.002) were identified as independent predictors of invasive adenocarcinoma (Table 2). The clinicoradiological (CR) model was developed by combining the two predictors, and the CR signature was computed from them.

Table 2 Univariate and Multivariate analysis of the clinicoradiological features
Full size table

The CR signatures exhibited significant differences between adenocarcinomas and benign nodules in both the training (0.706 ± 1.290 vs. -1.272 ± 1.840; p < 0.001) and test sets (0.742 ± 0.912 vs. 0.406 ± 0.945; p = 0.304). The AUC values of the CR model in the training set and test set were 0.694 (95% CI: 0.630–0.752) and 0.578 (95% CI: 0.457–0.658), respectively.

Construction and performance of the radiomics models

After eliminating features that were poorly reproducible or redundant, the remaining features were sorted using mRMR and screened by LASSO. Ultimately, a total of 6, 6, and 5 radiomic features with p < 0.05 were selected through stepwise forward multiple logistic regression for the development of independent radiomic models (the T0 model, T1 model, and delta model, respectively). These selected features were utilized to calculate the corresponding signatures (T0 signature, T1 signature, and delta signature).

Significant differences in the radiomic signatures between invasive adenocarcinomas and benign lesions were detected in both the training (T0-signatures: 0.706 ± 1.290 vs. -1.272.± 1.840; p < 0.001; T1-signatures: 1.244 ± 2.206 vs. -1.823 ± 2.596; p < 0.001; Delta-signatures: 0.283 ± 1.223 vs. -5.828 ± 0.861; p < 0.001;) and test (T0-signatures: 0.118 ± 0.575 vs. -2.653 ± 0.521; p = 0.001; T1-signatures: 0.549 ± 0.964 vs. -0.605 ± 1.003; p < 0.001; Delta-signatures: 0.958 ± 0.429 vs. -0.173 ± 0.334; p = 0.501) sets.

The predictive performance of the T0 model, T1 model, and delta model are summarized in Table 3. The corresponding ROC curves are presented in Fig. 3. Notably, the T1 model performed the best, with AUCs of 0.858 (95% CI: 0.857-0.900) in the training set and 0.817 (95% CI: 0.749–0.854) in the test set.

Table 3 Prediction performance of the models in the training and validation sets
Full size table

CR model: clinicoradiological model; T0 model: T0 radiomic model; T1 model: T1 radiomics model; Delta model: Delta radiomics model.

Fig. 3
figure 3

ROC curves of five models in the training set (A)and validation set (B). T1 model: T1 radiomics model; T0 model: T0 radiomics model; Delta model: Delta radiomics model; CR model: clinicoradiological model

Full size image

Construction and evaluation of the nomogram

The predictive nomogram model was constructed from the CR signature, T0 signature, T1 signature, and delta signature (Fig. 4). The nomogram exhibited significantly better predictive performance than the CR model (p < 0.001), T0 model (p < 0.001), T1 model (p = 0.004) and delta model (p < 0.001) in the training set. In the test set, the nomogram demonstrated superior performance compared to the CR model (p < 0.001), T0 model (p = 0.041), and delta model (p < 0.001). The DeLong test p values of all models are presented in Supplementary Table 4.

Fig. 4
figure 4

The integrated nomogram incorporating the CR-signature, T0-signature, T1-signature, and Delta-signature

Full size image

The calibration curve demonstrated a strong concordance between the predicted probability derived from the nomogram and the corresponding actual probability (Fig. 5). DCA showed that when the threshold probability was greater than 20%, the nomogram had greater net benefit than the CR, T0, T1, and delta models in all datasets (Fig. 6).

Fig. 5
figure 5

Calibration curves of the nomogram in the training and validation sets (A, B). The x-axis indicates the predicted probability estimated by the nomogram, while the y-axis indicates the actual probability. Apparent probabilities and bias-corrected probabilities are indicated by dotted and solid lines, respectively

Full size image
Fig. 6
figure 6

Decision curves of 5 models in all data set. The net income is shown on the y-axis, and the probability threshold is shown on the x-axis

Full size image

Discussion

The present study developed a nomogram based on clinicoradiological features and radiomic features extracted from longitudinal CT images to discriminate between pulmonary invasive adenocarcinomas and benign lesions. The nomogram utilizing radiomic features extracted from longitudinal CT images can enhance the discriminative capability between pulmonary invasive adenocarcinomas and benign lesions.

Regarding clinicoradiological features, only the long diameter and vascular abnormalities were independent predictors of invasive adenocarcinoma. These two factors are crucial in the follow-up management of pulmonary nodules according to the 2017 guidelines published by the Fleischner Society in the United States, which aligns with previous studies [22, 23]. However, Qi LL et al. [24] reported that some invasive adenocarcinomas may display a slow-growing pattern, the nodules showing limited alterations during short-term follow-up. These findings are supported by our comparison of baseline and follow-up CT data (Supplementary Table 3). Additionally, due to the limited dimensions of the nodules, inherent measurement errors are inevitable. For example, Van Riel et al. [25] reported that the evaluation of pulmonary nodules in the NELSON cohort showed only moderate agreement between eight radiologists in both the inter- and intraobserver morphological classification. Consequently, achieving a standardized criterion for lung nodule assessment solely based on conventional morphological features is challenging.

Radiomics [26, 27] has been clinically validated to have good diagnostic efficacy in the evaluation of pulmonary nodules, which encompasses discrimination between benign and malignant nodules, preoperative nodule classification, and prognostic analysis. Less attention has been given to the application of radiomic features during the follow-up of pulmonary nodules. Our study demonstrated that the radiomic model had a higher AUC for follow-up chest CT scans than for baseline CT scans, suggesting the potential utility of radiomics for monitoring pulmonary nodules. This result was consistent with that of Digumarthy et al. [28], who reported that radiomic features measured from lung nodules on preoperative CT scans exhibit greater predictive capability for nodule malignancy than the same radiomic features measured on baseline CT scans. The following factors may explain why the preoperative radiomic model performed better than the baseline radiomic model. First, the preoperative examination was temporally closer to the nodule’s pathological diagnosis compared to the baseline examination, while the cellular composition of the nodules exhibited temporal evolution over time [29]. Second, the radiomic features of malignant nodules are more susceptible to changes than are the radiomic features of benign nodules [28]. Importantly, radiomic features, such as irregularity of shapes, infiltration patterns, heterogeneity levels, or the presence of necrosis, can be utilized to quantify the phenotypic differences of tumours observed in CT images [30].

The development and progression of primary lung cancer are closely linked to dynamic alterations in the tumour microenvironment, which adaptively remodel to promote tumorigenesis [31, 32]. However, these alterations in the tumour microenvironment are not detectable through conventional imaging techniques. Delta-radiomic features were proposed to reflect temporal changes in the microenvironment or alterations induced by external factors, such as chemotherapy or radiotherapy. These features can be utilized for prognostic inference and disease monitoring purposes [33]. In this study, the delta model was helpful for distinguishing benign lesions from invasive adenocarcinoma nodules. Aerts HJ et al. [31] reported that grey inhomogeneity (HLH) can be used to express the intratumoural heterogeneity of lung cancer. This may be the reason why our delta model is helpful for distinguishing benign from invasive adenocarcinoma nodules. However, the delta model exhibited inferior performance compared to that of other independent radiomics models in the present study. The following factors may explain this result. First, the present study included pure ground-glass opacities (Supplementary Table 5) and had a relatively short follow-up period. A prior subsequent investigation conducted on ground-glass nodules (GGNs) [34] reported that among 351 GGNs, 208 stayed stable over the first five years of follow-up, suggesting that most GGNs are inert. Additionally, Yanqing Ma et al. [35], with a longer follow-up duration, revealed that delta radiomics exhibited improved capability in distinguishing between multiple primary lung adenocarcinomas and solitary primary lung adenocarcinomas. Second, Hawkins et al. [15] found that the utilization of radiomic characteristics from the initial assessment can effectively predict cancer occurrence with a precision of 76.79% and an AUC of 0.81, indicating that these radiomic features exhibit significant heterogeneity within the lesions.

This study distinguishes itself from previous radiomic analyses [36,37,38] by integrating longitudinal features extracted from baseline and follow-up CT images to present a comprehensive overview of pulmonary nodules. Longitudinal changes in radiomic features, as suggested by some studies [22, 39], may provide more predictive value for lung cancer screening. The nomogram developed in this study included delta-radiomic features from longitudinal lung CT images, which captured the dynamic changes in the internal texture features of pulmonary nodules and enhanced diagnostic accuracy. The American Fleischner Society [5] recommends follow-up for most indeterminate nodules, and our nomogram enhances the discriminatory capacity between benign lesions and invasive adenocarcinomas during short-term surveillance. The nomogram demonstrated consistent and outstanding predictive performance across various CT scanners, as evidenced by subgroup analysis (Supplementary Table 6). Furthermore, the majority of features employed in our diagnostic model are derived from wavelet transform, which reduces data dimensionality while facilitating efficient feature extraction and enhances the discrimination of different textures in CT images [40,41,42].

Despite our satisfactory findings, there are still certain limitations inherent to this study. First, due to its retrospective nature and the inclusion of patients with only three to six months of follow-up, selection bias may have been present. Second, the present study only differentiated between lung invasive adenocarcinoma and benign nodules, while malignant nodules were limited to a single pathological type. The study was done in a single centre and lacked an external validation set. To address these limitations, we aim to enhance the robustness of our findings by expanding the sample size and extending the follow-up period in future research.

Conclusion

The current study successfully developed a nomogram that utilizes clinicoradiological and longitudinal CT radiomic signatures to differentiate between pulmonary invasive adenocarcinomas and benign lesions. The incorporation of follow-up CT radiomic signatures and longitudinal delta-radiomic signatures improved the performance of both individual clinicoradiological models and baseline CT radiomic signatures in distinguishing between pulmonary invasive adenocarcinomas and benign lesions.

Data availability

The datasets used or analyzed during the current study are available from the corresponding author on reasonable request.

Abbreviations

AUC:

Area under the curve

CT:

Computed tomography

CEA:

Carcinoembryonic antigen

CA153:

Cancer antigen 153

Cyfra21-1:

Cytokeratin fragment 21-1

CA125:

Cancer antigen 125

DCA:

Decision curve analysis

ICC:

Intraclass correlation coefficient

LASSO:

Least absolute shrinkage and selection operator

MRMR:

Maximum correlation and minimum redundancy

NSE:

Neuron specific enolase

ProGRP:

Progastrin-releasing peptide

ROIs:

Regions of interest

SCCA:

Squamous cell carcinoma associated antigen

SF:

Serum ferritin

References

  1. Silva M, Milanese G, Ledda RE, Pastorino U, Sverzellati N. Screen-detected solid nodules: from detection of nodule to structured reporting. Transl Lung Cancer Res. 2021;10(5):2335–46.

    Article  PubMed  PubMed Central  Google Scholar 

  2. Gould MK, Donington J, Lynch WR, Mazzone PJ, Midthun DE, Wiener RS, et al. Evaluation of individuals with pulmonary nodules: when is it lung cancer? Diagnosis and management of lung cancer, 3rd ed: American College of Chest Physicians evidence-based clinical practice guidelines. Chest. 2013;143(5 Suppl):eS93–120.

    Article  Google Scholar 

  3. Callister ME, Baldwin DR, Akram AR, Barnard S, Cane P, British Thoracic Society Standards of Care Committee. British Thoracic Society guidelines for the investigation and management of pulmonary nodules. Thorax. 2015;70(Suppl 2):ii1–54.

    Article  PubMed  Google Scholar 

  4. MacMahon H, Austin JH, Gamsu G, Herold CJ, Jett JR, et al. Fleischner Society. Guidelines for management of small pulmonary nodules detected on CT scans: a statement from the Fleischner Society. Radiology. 2005;237(2):395–400.

    Article  PubMed  Google Scholar 

  5. MacMahon H, Naidich DP, Goo JM, Lee KS, Leung ANC, Bankier AA, et al. Guidelines for management of Incidental Pulmonary nodules detected on CT images: from the Fleischner Society 2017. Radiology. 2017l;284(1):228–43.

    Article  PubMed  Google Scholar 

  6. Harris RP, Sheridan SL, Lewis CL, Barclay C, Vu MB, Brewer NT, et al. The harms of screening: a proposed taxonomy and application to lung cancer screening. JAMA Intern Med. 2014;174(2):281–5.

    Article  PubMed  Google Scholar 

  7. Pan Z, Hu G, Zhu Z, Tan W, Han W, Jin Z, et al. Predicting invasiveness of lung adenocarcinoma at chest CT with Deep Learning Ternary classification models. Radiology. 2024;311(1):e232057.

    Article  PubMed  Google Scholar 

  8. Lai Y, Wang Y, Wu Y, Wu M, Xing S, Liu W et al. Identification and Validation of Serum CST1 as a Diagnostic Marker for Differentiating Early-Stage Non-Small Cell Lung Cancer from Pulmonary Benign Nodules. Cancer Control. 2022 Jan-Dec;29:10732748221104661.

  9. Gillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures. They Are Data Radiol. 2016;278(2):563–77.

    Google Scholar 

  10. Lambin P, Rios-Velazquez E, Leijenaar R, Carvalho S, van Stiphout RG, Aerts HJ, et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer. 2012;48(4):441–6.

    Article  PubMed  PubMed Central  Google Scholar 

  11. Yi L, Peng Z, Chen Z, Tao Y, Lin Z, Zuo M, 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.

    Article  PubMed  PubMed Central  Google Scholar 

  12. She Y, Jin Z, Wu J, Deng J, Zhang L, Chen C, et al. Development and validation of a deep learning model for Non-small Cell Lung Cancer Survival. JAMA Netw Open. 2020;3(6):e205842.

    Article  PubMed  PubMed Central  Google Scholar 

  13. Xu QQ, Shan WL, Zhu Y, Huang CC, Bao SY, Guo LL. Prediction efficacy of feature classification of solitary pulmonary nodules based on CT radiomics. Eur J Radiol. 2021;139:109667.

    Article  PubMed  Google Scholar 

  14. Calheiros JLL, de Amorim LBV, de Lima LL, de Lima Filho AF, Ferreira Júnior JR, de Oliveira MC. The effects of Perinodular features on solid lung nodule classification. J Digit Imaging. 2021;34(4):798–810.

    Article  PubMed  PubMed Central  Google Scholar 

  15. Travis WD, Asamura H, Bankier AA, Beasley MB, Detterbeck F, Yatabe Y, et al. The IASLC Lung Cancer Staging Project: proposals for Coding T Categories for Subsolid Nodules and Assessment of Tumor Size in Part-Solid tumors in the Forthcoming Eighth Edition of the TNM classification of Lung Cancer. J Thorac Oncol. 2016;11(8):1204–23.

    Article  PubMed  Google Scholar 

  16. Scholten ET, de Jong PA, de Hoop B, van Klaveren R, van Amelsvoort-van S, Gietema HA, et al. Towards a close computed tomography monitoring approach for screen detected subsolid pulmonary nodules? Eur Respir J. 2015;45(3):765–73.

    Article  PubMed  Google Scholar 

  17. Dagogo-Jack I, Shaw AT. Tumour heterogeneity and resistance to cancer therapies. Nat Rev Clin Oncol. 2018;15(2):81–94.

    Article  PubMed  CAS  Google Scholar 

  18. Paracchini L, Mannarino L, Craparotta I, Romualdi C, Fruscio R, Marchini S, et al. Regional and temporal heterogeneity of epithelial ovarian cancer tumor biopsies: implications for therapeutic strategies. Oncotarget. 2016;12(24):2404–17.

    Article  PubMed  PubMed Central  Google Scholar 

  19. Nardone V, Reginelli A, Grassi R, Boldrini L, Vacca G, Cappabianca S, et al. Delta radiomics: a systematic review. Radiol Med. 2021;126(12):1571–83.

    Article  PubMed  Google Scholar 

  20. Alahmari SS, Cherezov D, Goldgof D, Hall L, Gillies RJ, Schabath MB. Delta Radiomics improves Pulmonary Nodule Malignancy Prediction in Lung Cancer Screening. IEEE Access. 2018;6:77796–806.

    Article  PubMed  PubMed Central  Google Scholar 

  21. Nicholson AG, Tsao MS, Beasley MB, Borczuk AC, Brambilla E, Travis WD, et al. The 2021 WHO classification of lung tumors: impact of advances since 2015. J Thorac Oncol. 2022;17(3):362–87.

    Article  PubMed  Google Scholar 

  22. Zhou C, Zhao X, Zhao L, Liu J, Chen Z, Fang S. Deep learning-based CT imaging in the diagnosis of treatment effect of Pulmonary nodules and Radiofrequency ablation. Comput Intell Neurosci. 2022;2022:7326537.

    Article  PubMed  PubMed Central  Google Scholar 

  23. Shi Z, Wang Y, He X. Differential diagnosis of solitary pulmonary nodules with dual-source spiral computed tomography. Exp Ther Med. 2016;12(3):1750–4.

    Article  PubMed  PubMed Central  Google Scholar 

  24. Qi LL, Wang JW, Yang L, Huang Y, Zhao SJ, Wu N, et al. Natural history of pathologically confirmed pulmonary subsolid nodules with deep learning-assisted nodule segmentation. Eur Radiol. 2021;31(6):3884–97.

    Article  PubMed  Google Scholar 

  25. Nam JG, Goo JM. Evaluation and management of Indeterminate Pulmonary nodules on chest computed tomography in asymptomatic subjects: the principles of Nodule guidelines. Semin Respir Crit Care Med. 2022;43(6):851–61.

    Article  PubMed  Google Scholar 

  26. She Y, Zhang L, Zhu H, Dai C, Xie D, Chen C, et al. The predictive value of CT-based radiomics in differentiating indolent from invasive lung adenocarcinoma in patients with pulmonary nodules. Eur Radiol. 2018;28(12):5121–8.

    Article  PubMed  Google Scholar 

  27. Zhao W, Xu Y, Yang Z, Sun Y, Li C, Li M, et al. Development and validation of a radiomics nomogram for identifying invasiveness of pulmonary adenocarcinomas appearing as subcentimeter ground-glass opacity nodules. Eur J Radiol. 2019;112:161–8.

    Article  PubMed  Google Scholar 

  28. Digumarthy SR, Padole AM, Rastogi S, Price M, Mooradian MJ, Kalra MK, et al. Predicting malignant potential of subsolid nodules: can radiomics preempt longitudinal follow up CT? Cancer Imaging. 2019;19(1):36.

    Article  PubMed  PubMed Central  Google Scholar 

  29. Tao G, Shi D, Yu L, Chen C, Zhang Z, Yu H, et al. Longitudinal prediction of lung nodule invasiveness by sequential modelling with common clinical computed tomography (CT) measurements: a prediction accuracy study. Transl Lung Cancer Res. 2022;11(5):845–57.

    Article  PubMed  PubMed Central  Google Scholar 

  30. Coroller TP, Grossmann P, Hou Y, Rios Velazquez E, Leijenaar RT, Aerts HJ, et al. CT-based radiomic signature predicts distant metastasis in lung adenocarcinoma. Radiother Oncol. 2015;114(3):345–50.

    Article  PubMed  PubMed Central  Google Scholar 

  31. Aerts HJ, Velazquez ER, Leijenaar RT, Parmar C, Grossmann P, Lambin P, et al. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun. 2014;5:4006.

    Article  PubMed  CAS  Google Scholar 

  32. Bissell MJ, Hines WC. Why don’t we get more cancer? A proposed role of the microenvironment in restraining cancer progression. Nat Med. 2011;17(3):320–9.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  33. Egeblad M, Nakasone ES, Werb Z. Tumors as organs: complex tissues that interface with the entire organism. Dev Cell. 2010;18(6):884–901.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  34. Lee SW, Leem CS, Kim TJ, Lee KW, Chung JH, Lee CT, et al. The long-term course of ground-glass opacities detected on thin-section computed tomography. Respir Med. 2013;107(6):904–10.

    Article  PubMed  Google Scholar 

  35. Ma Y, Li J, Xu X, Zhang Y, Lin Y. The CT delta-radiomics based machine learning approach in evaluating multiple primary lung adenocarcinoma. BMC Cancer. 2022;22(1):949.

    Article  PubMed  PubMed Central  Google Scholar 

  36. Palumbo B, Bianconi F, Palumbo I, Fravolini ML, Minestrini M, Spanu A, et al. Value of shape and texture features from 18F-FDG PET/CT to discriminate between Benign and Malignant Solitary Pulmonary nodules: an experimental evaluation. Diagnostics (Basel). 2020;10(9):696.

    Article  PubMed  CAS  Google Scholar 

  37. Sun Y, Li C, Jin L, Gao P, Zhao W, Li M, et al. Radiomics for lung adenocarcinoma manifesting as pure ground-glass nodules: invasive prediction. Eur Radiol. 2020;30(7):3650–9.

    Article  PubMed  PubMed Central  Google Scholar 

  38. Zhuo Y, Zhan Y, Zhang Z, Shan F, Shen J, Yu M, et al. Clinical and CT Radiomics Nomogram for Preoperative differentiation of Pulmonary Adenocarcinoma from Tuberculoma in Solitary Solid Nodule. Front Oncol. 2021;11:701598.

    Article  PubMed  PubMed Central  Google Scholar 

  39. Ardila D, Kiraly AP, Bharadwaj S, Choi B, Reicher JJ, Shetty S, et al. Author correction: end-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography. Nat Med. 2019;25(8):1319.

    Article  PubMed  CAS  Google Scholar 

  40. Kang E, Min J, Ye JC. A deep convolutional neural network using directional wavelets for low-dose X-ray CT reconstruction. Med Phys. 2017;44(10):e360–75.

    Article  PubMed  CAS  Google Scholar 

  41. Choe J, Lee SM, Do KH, Lee G, Lee JG, Seo JB, et al. Deep learning-based Image Conversion of CT Reconstruction Kernels improves Radiomics Reproducibility for Pulmonary nodules or masses. Radiology. 2019;292(2):365–73.

    Article  PubMed  Google Scholar 

  42. Liao F, Liang M, Li Z, Hu X, Song S. Evaluate the malignancy of pulmonary nodules using the 3-D Deep Leaky Noisy-OR Network. IEEE Trans Neural Netw Learn Syst. 2019;30(11):3484–95.

    Article  PubMed  Google Scholar 

Download references

Acknowledgements

The authors thank American Journal Experts (AJE) for assisting in the preparation of this paper.

Funding

This work was funded by Clinical major innovative characteristic technology project of the Second Affiliated Hospital of Army Medical University (2018JSLC0016), and talent Project of Chongqing (Dong Zhang, CQYC202103075, cstc2022ycjh-bgzxm0082).

Author information

Authors and Affiliations

  1. Department of Radiology, XinQiao Hospital of Army Medical University, Chongqing, 400037, China

    Zhengming Wang, Fei Wang, Yan Yang, Weijie Fan, Li Wen & Dong Zhang

  2. Department of Medical imaging, Luzhou People’s Hospital, Luzhou, 646000, China

    Fei Wang

Authors
  1. Zhengming Wang
    View author publications

    You can also search for this author inPubMed Google Scholar

  2. Fei Wang
    View author publications

    You can also search for this author inPubMed Google Scholar

  3. Yan Yang
    View author publications

    You can also search for this author inPubMed Google Scholar

  4. Weijie Fan
    View author publications

    You can also search for this author inPubMed Google Scholar

  5. Li Wen
    View author publications

    You can also search for this author inPubMed Google Scholar

  6. Dong Zhang
    View author publications

    You can also search for this author inPubMed Google Scholar

Contributions

Author contributions DZ and LW were guarantors of integrity of the entire study and contributed to study concepts and design. ZMW and FW were involved in experimental studies/data analysis. YY and FW were involved in statistical analysis. WMZ and WJF contributed to manuscript preparation and manuscript editing. DZ and LW reviewed and edited the final version of the manuscript. All authors read and approved the final manuscript.

Corresponding author

Correspondence to Dong Zhang.

Ethics declarations

Ethics approval and consent to participate

This retrospective study was approved by the Medical Ethics Committee of XinQiao Hospital of Army Medical University(No: 2020-147-01), and patient informed consent was waived. All methods were implemented in accordance with the approved regulations and the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary Material 1

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Wang, Z., Wang, F., Yang, Y. et al. Development of a nomogram-based model incorporating radiomic features from follow-up longitudinal lung CT images to distinguish invasive adenocarcinoma from benign lesions: a retrospective study. BMC Pulm Med 24, 534 (2024). https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12890-024-03360-8

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12890-024-03360-8

Keywords

BMC Pulmonary Medicine

ISSN: 1471-2466

Contact us