Sage Journals: Discover world-class research

Abstract

Introduction: Small-cell lung cancer (SCLC) is a leading cause of cancer-related mortality worldwide, with limited therapeutic outcomes and poor prognosis. Accurate diagnosis and optimal surgical decision-making remain critical challenges. This study aimed to develop and validate a clinical-radiomics nomogram integrating computed tomography (CT) radiomic features of the peritumoral region and clinical factors to improve SCLC diagnosis and guide surgical planning.

Methods: A retrospective cohort of 113 patients (54 SCLC, 59 non-small cell lung cancer) was analyzed. CT images were processed to extract 1050 radiomic features from both intratumoral and peritumoral (2-mm expanded) ROIs. Feature selection was performed using t-tests, LASSO regression, and mRMR analysis. Logistic regression models were constructed for original and expanded ROIs, and a clinical-radiomics nomogram was developed by combining significant radiomic features with independent clinical predictors (gender, smoking history, tumor diameter, glitch, and neuron-specific enolase levels). Model performance was evaluated using ROC curves, AUC, sensitivity, specificity, and CIC curves.

Results: The expanded ROI radiomics model outperformed the original ROI and clinical models, achieving higher accuracy (0.83 vs 0.76/0.70), sensitivity (0.80 vs 0.74/0.77), specificity (0.85 vs 0.75/0.65), and AUC (0.85 vs 0.76/0.71). The clinical-radiomics nomogram demonstrated superior diagnostic performance, with an AUC of 0.96 (95% CI: 0.88-1.00), accuracy of 0.91, sensitivity of 0.92, and specificity of 0.90. CIC analysis confirmed its clinical utility for surgical decision-making at intermediate-risk thresholds.

Conclusion: The integration of peritumoral radiomic features and clinical factors into a nomogram provides a non-invasive tool for SCLC diagnosis and surgical planning. The superiority of the expanded model substantiates the potential presence of SCLC in peri-tumoral tissues that may be imperceptible through conventional imaging, thereby offering guidance for surgical decision-making. This approach has potential for improving treatment outcomes and warrants further validation in multicenter studies.

Keywords

SCLC CT radiomics peritumoral radiology NSCLC

Introduction

Lung cancer, the leading cause of cancer-related mortality globally accounting for a substantial proportion of deaths in both men and women.^1,2 It can be predominantly categorized into 2 major subtypes: SCLC and non-small cell lung cancer (NSCLC).³ Among its subtypes, SCLC is a highly aggressive neuroendocrine tumor responsible for approximately 250 000 deaths annually.⁴ The unique clinical behaviors and treatment challenges of SCLC—such as its aggressive nature, rapid recurrence, early metastasis, and poor prognosis—emphasize the need for tailored therapeutic strategies.⁵

In recent years, surgical intervention has emerged as a viable treatment option for SCLC, with large retrospective studies suggesting that surgery may offer superior outcomes compared to chemoradiotherapy in select cases.^6,7 Several international guidelines from American, European, and Japanese medical associations recommend surgery with adjuvant chemotherapy for early-stage SCLC patients without mediastinal nodal disease.^8,9 In the study by Yang et al,¹⁰ 29.6% (681/2301) of SCLC patients were identified as eligible for surgery, characterized mainly by T1N0 tumors. Radiomics tools can further help identify patients who may be suitable candidates for potential neoadjuvant chemotherapy before surgery.

Complete resection of SCLC lesions while minimizing damage to healthy lung tissue and addressing peritumoral disease remains a significant clinical challenge. CT is the primary imaging modality for SCLC diagnosis, provides critical morphological information for screening and evaluation.¹¹ However, considerable overlap in CT features among lung cancer subtypes often complicates accurate diagnosis based solely on visual assessment.^12,13 Radiomics offers a solution by extracting high-dimensional quantitative imaging features for improved tumor characterization, potentially supporting diagnosis, staging, and treatment planning.^14,15 While CT-based radiomics has shown promise in guiding precise surgical decisions in SCLC, but most studies to date have focused on histological subtyping,¹⁶ and its application in surgical decision-making remains limited.

Recent research indicates that tumor’s physiological microenvironment, particularly the peritumoral region, harbors critical prognostic information that may be overlooked by analyses limited to the tumor itself.^17-20 To address this limitation, we incorporated peritumoral radiomic features extracted from the extended 2-mm region surrounding visible lung lesions on CT. Using these features, we developed a radiomics model aimed at differentiating SCLC from NSCLC. This study aims to provide a more comprehensive understanding by integrating both lesion and peritumoral radiomic features with clinical data to assess their potential for distinguishing SCLC from NSCLC. Furthermore, we explore the potential applications of these radiomic signatures in guiding surgical decision-making for SCLC patients. Ultimately, this approach aims to improve accuracy in characterizing lesion extent, facilitate early diagnosis, enable precise staging, and support the development of personalized treatment strategies.

Materials and Methods

Data Collection

This retrospective study enrolled 113 patients with histopathologically confirmed primary lung tumors, diagnosed between March 2022 and August 2023, comprising 54 cases of SCLC and 59 cases of NSCLC. Inclusion criteria included histopathological diagnosis of primary lung cancers and no prior treatment. Exclusion criteria included history of other primary malignancies, certain pathological types, incomplete clinical data, prior treatment, and lesions with a maximum diameter <1 cm.

Collected clinical data included demographic information (sex, age, smoking history, lymph node metastasis), tumor characteristics (tumor diameter, imaging features), and laboratory biomarkers, which included neuron-specific enolase (NSE), carcinoembryonic antigen (CEA), cancer antigen 125 (CA125), and cytokeratin 19 fragment (CYFRA21-1). These data were used to characterize the patient population and explore potential correlations between clinical features and radiomic characteristics.

The Ethics Committee of The First Affiliated Hospital of Zhejiang Chinese Medical University approved the study (2023-KLS-098-01). The requirement for informed consent was waived due to the retrospective study design.

CT Scanning Protocol

CT imaging was performed using Toshiba Aquilion ONE (Toshiba Medical Systems, Japan) and Siemens Somatom Sensation 64-slice scanners (Siemens Healthcare, Germany). Scanning parameters for the Toshiba system included: tube voltage 120 kV, tube current 200 mA, detector collimation 0.5 mm × 64 mm, matrix 512 × 512, slice thickness 1.0 mm, rotation time 0.4 s, and reconstruction kernel FC51. Parameters for the Siemens system were: tube voltage 120 kV, automatic tube current modulation, detector collimation 0.6 mm × 64 mm, matrix 512 × 512, slice thickness 0.75 mm, and reconstruction kernel B3H. Standard lung windows (window level: −600 HU, window width: 1300 HU) were used, with adjustments as needed to visualize internal lesion characteristics. Images were routinely transferred to the Picture Archiving and Communication System (PACS) for storage.

Lesion Segmentation and Radiomics Feature Extraction

Image Selection

CT images of selected patients were retrieved from the PACS. Images were reviewed by radiologists with over 10 years of diagnostic experience to ensure the presence of complete sequences without artifacts before segmentation.

Lesion and Peritumoral Region Segmentation

Thin-slice axial CT images in DICOM format were imported into 3D Slicer (version 5.3.0) for segmentation. Regions of interest (ROIs) were manually delineated on lung window images, with reference to mediastinal windows, to ensure accurate lesion coverage. Each lesion was outlined, carefully avoiding blood vessels, bronchi, and extrathoracic normal tissues. The original ROIs were semi-automatically segmented and extended by 2 mm outward, with manual adjustments to exclude normal structures. For lesions with unclear margins, only regions with significantly high density were delineated. In cases with multiple lesions, the largest lesion was selected for analysis. Segmentation and review were performed independently by 2 experienced radiologists, each with over 10 years of expertise in thoracic imaging. In cases of disagreement, a third radiologist with more than 15 years of experience in thoracic imaging adjudicated the final decision. The workflow of the study is illustrated in Figure 1.

Figure 1.

Overall Workflow of This Study

Radiomics Feature Extraction, Selection, and Model Construction

The reporting of this study conforms to TRIPOD guidelines.²¹

Using the pyradiomics module in 3D Slicer, 1050 radiomics features were automatically extracted for each ROI, including shape, first-order statistics, and texture features such as gray-level co-occurrence matrix (GLCM), gray-level size zone matrix (GLSZM), gray-level dependence matrix (GLDM), gray-level run-length matrix (GLRLM), and neighborhood gray-tone difference matrix (NGTDM). Advanced filtering techniques, including wavelet and Laplacian of Gaussian (LoG) filters, were applied to enhance feature extraction. Wavelet filters decomposed images into multiscale components, while LoG filters emphasized regions with high gray-level variation. Smaller sigma values highlighted finer textures, whereas larger values emphasized coarser patterns.^22,23

Logistic regression models were constructed to calculate radiomics scores for each patient. The dataset was standardized and split into training and testing sets at a 7:3 ratio. Feature selection was performed using t-tests, least absolute shrinkage and selection operator (LASSO) regression, and maximum relevance minimum redundancy (mRMR). Logistic regression models were developed for original and expanded ROIs based on selected features. We also performed a five-fold cross-validation (internal validation), which is recognized as an effective way to generate unbiased patient-specific predictions.²⁴

Clinical and Clinical-Radiomics Nomogram Models

Independent clinical predictors for SCLC were identified through t-tests and LASSO regression, elucidating the key predictive factors. Following this, we developed a clinical logistic regression model incorporating the selected clinical variables. Additionally, these features were integrated with the radiomics score (rad scores) to construct a combined clinical-radiomics nomogram.

Statistical Analysis

Feature stability was evaluated using the Intraclass correlation coefficient (ICC). Here, an ICC of ≥0.75-0.89 was considered good reproducibility and an ICC ≥0.90 was considered excellent reproducibility as recommended by Koo et al.²⁵ Differences in continuous variables were analyzed using independent t-tests, while categorical variables were assessed with Fisher’s exact test or the chi-square test. Diagnostic performance was evaluated using receiver operating characteristic (ROC) curves and clinical impact curves (CIC). Key metrics included area under the curve (AUC), sensitivity, specificity, and accuracy. Statistical comparisons of ROC curves were performed using the DeLong method.²⁶ The Hosmer-Lemeshow test assessed nomogram calibration using the statsmodels package. Statistical analyses were conducted in Python (version 3.9.13), with significance set at P < .05.

Results

Reproducibility Analysis

Inter-observer reproducibility analysis of both the whole nodule and solid component volumes delineated by the 2 radiologists demonstrated good agreement, with all ICCs exceeding 0.75.

Comparison of Clinical Factors Between SCLC and NSCLC Patients

A comparative analysis of clinical characteristics between SCLC and NSCLC patients revealed significant differences in gender, smoking prevalence, maximum tumor diameter, tumor glitch, and NSE levels (P < .05). In contrast, no significant differences were observed between the 2 groups with respect to age, tumor location, margin clarity, pleural effusion, or biomarkers such as CEA, CA125, and CYFRA21-1 (P > .05) (Table 1).

Table 1.

Comparison of Clinical Factors Between SCLC and NSCLC Patients

	Clinical characteristics		SCLC	NSCLC	P	T Or chi-square value
Clinical data	Gender	Male	46 (85.18%)	28 (48.27%)	<.001	16.99
	Gender	Female	8 (14.82%)	30 (51.73%)	<.001	16.99
	Age		65.51 ± 9.19	61.04 ± 10.26	.05	1.953
	Smoke	Yes	32 (59.25%)	13 (22.41%)	<.001	15.79
	Smoke	No	22 (40.75%)	45 (77.69%)	<.001	15.79
	Maximum diameter		41.03 ± 21.46	25.56 ± 19.70	<.001	3.976
	Tumor location	Left lung	23 (42.59%)	29 (50.00%)	.43	0.617
	Tumor location	Right lung	31 (57.41%)	29 (50.00%)	.43	0.617
	Lymph node metastasis	Yes	31 (57.41%)	34 (58.62%)	.07	3.159
	Lymph node metastasis	No	23 (42.59%)	24 (41.48%)	.07	3.159
Imaging data	Tumor margin	Clear	28 (51.85%)	32 (55.17%)	.73	0.124
	Tumor margin	Unclear	26 (48.15%)	26 (44.83%)	.73	0.124
	Lobular sign	Yes	43 (79.62%)	51 (87.93%)	.23	1.429
	Lobular sign	No	11 (20.48%)	7 (12.17%)	.23	1.429
	Pleural effusion	Yes	15 (27.77%)	12 (20.68%)	.38	0.768
	Pleural effusion	No	39 (72.23%)	46 (79.32%)	.38	0.768
	Glitch	Yes	22 (40.74%)	40 (68.96%)	.003	9.014
	Glitch	No	32 (59.36%)	18 (31.04%)	.003	9.014
Laboratory data	NSE		17.32 ± 33.54	4.23 ± 3.50	.004	2.956
	CEA		68.90 ± 419.84	11.90 ± 26.90	.304	1.032
	CA125		32.24 ± 51.52	27.79 ± 32.69	.584	0.549
	CYFRA21-1		3.23 ± 3.04	8.47 ± 20.95	.072	1.817

Selection of Radiomics and Clinical Features

Feature selection was performed using t-tests and least absolute shrinkage and LASSO regression, which retained 11 features from the original ROI and 20 features from the expanded ROI. The efficacy of LASSO regression in selecting texture features within lesions, along with the detailed procedure and results of feature selection, are illustrated in Figures 2 and 3. To further optimize feature dimensions and improve model efficiency, the mRMR classifier was employed, identifying the most relevant and least redundant features. A total of 10 representative features were selected from each ROI configuration, with their respective weight distributions displayed in Figure 4. Moreover, t-tests and LASSO analysis identified gender, smoking history, maximum tumor diameter, glitch, and NSE levels as independent clinical predictors for SCLC.

Figure 2.

Selection of Radiomics Texture Features Using t-test and LASSO Regression

Figure 3.

Feature Selection Flowchart

Figure 4.

Feature Weights Selected From Original and Expanded ROIs. (A) Feature Weights Within the Original ROI. (B) Feature Weights Within the 2-mm Expanded ROI

Model Performance Comparison

The clinical model demonstrated limited diagnostic capability, with an accuracy of 0.70, sensitivity of 0.77, and specificity of 0.65, primarily attributable to its low specificity, which resulted in a higher false-positive rate. The original ROI radiomics model showed improved diagnostic performance, but remained suboptimal. In contrast, the expanded ROI model achieved higher accuracy (0.83), sensitivity (0.80), and specificity (0.84), capturing additional lesion information and enhancing discrimination ability. As illustrated in Figure 5, the expanded ROI model’s AUC in the test set was 0.85 (95% CI, 0.80-0.97), outperforming the clinical model (AUC = 0.71, 95% CI, 0.79-0.90) and the original ROI model (AUC = 0.76, 95% CI, 0.76-0.87).

Figure 5.

Model Performance Comparison. (A) AUC of the Clinical Model. (B) AUC of the Original ROI Model. (C) AUC of the Expanded ROI Model. (D) AUC of the Clinical-Radiomics Nomogram

A clinical-radiomics nomogram was developed by combining expanded ROI radiomics features with clinical predictors for SCLC prediction, as shown in Figure 6. This model demonstrated superior performance, with an AUC of 0.96 (95% CI, 0.88-1.00), accuracy of 0.91, sensitivity of 0.92, and specificity of 0.90 in the test set. Hosmer-Lemeshow tests indicated good calibration in both the test (P = .891) and training sets (P = .663). DeLong tests confirmed significant differences in diagnostic performance between models (P < .05). The diagnostic performance of each model is detailed in Table 2.

Figure 6.

Clinical-Radiomics Nomogram Constructed by Integrating Radiomics Features and Clinical Factors to Diagnose SCLC

Table 2.

Diagnostic Performance Metrics of Each Model in the Test Set

Model	Sensitivity	Specificity	Accuracy	AUC (95% CI)
Clinical model	0.77	0.65	0.70	0.71 (0.79-0.90)
Original ROI model	0.74	0.75	0.76	0.76 (0.76-0.87)
Expanded ROI model	0.80	0.85	0.83	0.85 (0.80-0.97)
Clinical-radiomics nomogram model	0.92	0.90	0.91	0.96 (0.88-1.00)

CIC analysis, as presented in Figure 7, revealed that the clinical-radiomics nomogram consistently predicted more true-positive cases at a high-risk threshold above 72%, whereas other models achieved optimal benefits at thresholds >80%. These findings suggest that the clinical-radiomics nomogram has the potential to improve treatment decision-making in clinical practice.

Figure 7.

The Clinical Impact Curves. (A) CIC of the Clinical Model. (B) CIC of the Original ROI Model. (C) CIC of the Expanded ROI Model. (D) CIC of the Clinical-Radiomics Nomogram

Discussion

Conventional single-energy CT imaging evaluates tumor characteristics based on attenuation, morphology, and invasiveness, with treatment response assessed through changes in tumor volume and density.²⁰ However, relying solely on morphological assessments often falls short in accurately determining tumor pathology, highlighting the need for more comprehensive evaluation methods. Radiomics enhances tumor characterization by extracting extensive quantitative imaging features, enabling a detailed depiction of tumor phenotypes and offering a more nuanced understanding of tumor biology.^22,23 Advanced statistical and machine learning methods can identify clinically significant features, providing high accuracy and non-invasive evaluation potential.^27-30

The biological significance of the peritumoral microenvironment in tumor growth and progression has been well established.³¹ However, radiomics studies on SCLC have primarily focused on intralesional features and prognosis,^32-34 with limited exploration of peritumoral characteristics. Integrating radiomics features from both the lesion and the peritumoral microenvironment may improve diagnostic accuracy and predictive efficiency, offering a novel perspective for surgical guidance by incorporating microscopic textural complexity and multidimensional data. This approach has the potential to provide a more comprehensive understanding of SCLC, enabling more accurate diagnosis and treatment planning.

In this study, we constructed 2 ROIs based on SCLC and NSCLC lesions respectively: the original ROI, which is based on visually identifiable SCLC or NSCLC lesion areas on CT images, and an expanded ROI with a 2-mm margin from the original ROI. Meanwhile, we identified clinical features such as gender, smoking history, maximum tumor diameter, glitch, and NSE levels as effective differentiators between SCLC and other lung cancers. After extracting the radiomics features from the 2 ROIs and performing dimensionality reduction, logistic regression models were established for the clinical variables, original ROI, and expanded ROI to evaluate their diagnostic performance. The results showed that compared to the clinical models, the radiomics-based model derived from the original ROI demonstrated superior performance compared to the clinical model, with higher accuracy (0.70 vs 0.76), specificity (0.65 vs 0.75), and sensitivity (0.77 vs 0.74). In previous studies investigating the role of radiomics in the differential diagnosis of lung cancer, Wu et al³⁵ reported an AUC of 0.72 for distinguishing lung cancer subtypes. Similarly, Junior et al³⁶ demonstrated an AUC of 0.71 in the training cohort when utilizing radiomics features from CT images to classify lung cancer subtypes, which is comparable to the performance of our original ROI model. In contrast, Liu et al reported a higher AUC of 0.82. The lower performance in our study compared to Liu et al³⁷ may be attributed to differences in sample size, which could have influenced the predictive performance of the radiomics models. However, our analysis of the expanded ROI showed that the results from our expanded ROI model outperformed theirs. Furthermore, compared to the radiomics model constructed using the original ROI, the expanded ROI model notably further improved accuracy (0.83 vs 0.76), sensitivity (0.80 vs 0.74), specificity (0.85 vs 0.75), and AUC (0.85 vs 0.76).A related conclusion has also been discussed in the study by Jia et al.³⁸ In their study, the expanded ROI demonstrated significantly higher accuracy (0.89 vs 0.82), sensitivity (0.85 vs 0.86), specificity (0.84 vs 0.79), and AUC (0.92 vs 0.86) compared to the original ROI. The advantage of the expanded ROI radiomics model over the original ROI model may lie in its ability to capture subtle features in the peritumoral microenvironment that are often undetectable by human observation, which also suggests that it may have captured certain peritumoral information. This finding is consistent with evidence suggesting that extending surgical margins improves overall survival in SCLC.^39,40

Moreover, we calculated the rad score derived from the expanded ROI model and integrated it with the clinical features used to construct the clinical model. Based on this, we developed a clinical-radiomics nomogram that achieved superior diagnostic performance (AUC = 0.96), significantly outperforming both the clinical model (AUC = 0.76) and the expanded ROI radiomics model alone (AUC = 0.85). Our findings align closely with those reported by Liu et al,³⁷ where their radiomics nomogram achieved an AUC of 0.94 compared to our 0.96. In the nomogram, rad scores and NSE hold significant diagnostic weight. NSE, a well-established neuroendocrine marker, continues to play an important role in distinguishing SCLC from NSCLC when combined with imaging-based predictors.^41,42 Similarly, in the studies by Liu et al³⁷ and Gigika et al,⁴² NSE was also identified as a crucial clinical marker. However, unlike previous studies where clinical-radiomics nomograms offered no significant advantages over radiomics models alone,^34,43 our approach demonstrated superior performance, likely due to the incorporation of peritumoral features. The nomogram’s visual representation facilitates clinical decision-making, providing an effective tool for individualized prediction of SCLC and NSCLC. CIC analysis further validates the clinical utility of the nomogram. In contrast to the clinical model and the 2 standalone radiomics models, the clinical-radiomics nomogram demonstrated high concordance between predicted and actual SCLC cases at the moderate-risk threshold (0.71), whereas the other 3 models required higher risk thresholds exceeding 0.8. This finding suggests that extending the surgical margin by an additional 2 mm could enhance SCLC resection and improve postoperative outcomes, highlighting its potential for supporting surgical risk assessment, postoperative prognosis prediction, and enabling more precise surgical planning.

Despite its promising findings, this study has limitations. It is a single-center retrospective study with a limited sample size. The number of SCLC cases is relatively small, potentially introducing bias and reducing generalizability. Lymph node status is a key prognostic factor for non-metastatic lung cancer,^44-47 with radiomics providing superior accuracy in detecting metastasis through primary tumor analysis compared to size-based CT criteria.^48,49

In future studies, we plan to conduct a prospective multi-center validation study involving 3-5 medical centers across different regions, targeting a sample size increase of at least 50%. This will include the establishment of standardized imaging protocols and centralized standard operating procedures (SOPs) for image processing and feature extraction. Additionally, we will implement training for research personnel to ensure technical standardization across centers. To address clinical integration, we will develop decision support tools integrated with PACS systems to provide real-time radiomics-based risk stratification for lymph node involvement, directly informing surgical candidacy and operative approach selection. The prospective validation will specifically focus on surgical cohorts to evaluate how radiomics predictions influence treatment decisions and patient outcomes. We will also incorporate comprehensive mediastinal staging data to evaluate the model’s performance in predicting lymph node involvement, which represents a critical prognostic factor that can significantly influence treatment planning and patient outcomes. We aim to refine the diagnostic and decision-making framework combining CT radiomics with clinical features for SCLC.

Conclusions

This study developed an efficient clinical-radiomics nomogram by integrating CT radiomics and clinical features of SCLC patients, demonstrating excellent diagnostic performance. Compared to traditional clinical models, the expanded ROI radiomics model significantly improved diagnostic accuracy and specificity, highlighting the importance of peritumoral radiomic features in SCLC diagnosis. The nomogram not only achieved a high AUC in the test set but also showed superior predictive performance at intermediate-risk thresholds, offering a precise, non-invasive tool with potential utility in individualized treatment planning. Future studies should incorporate longitudinal patient survival data and prospective surgical outcome assessments to comprehensively evaluate the clinical utility of this nomogram.

Footnotes

ORCID iD

Junna Wang

Ethial Approval

The study was reviewed and approved by the Ethics Committee of The First Affiliated Hospital of Zhejiang Chinese Medical University, located in Zhejiang Province, China. The approval number is 2023-KLS-098-01, and the approval date is July 19, 2023. The ethical review confirmed compliance with the Helsinki Declaration and the “Measures for the Ethical Review of Biomedical Research Involving Humans”.

Patient Consent

Not applicable. The requirement for informed consent was waived due to the retrospective study design.

Author Contributions

Conceptualization, Shiwei Wang; Data curation, Hao Zheng; Formal analysis, Lanqi Fu; Investigation, Yujie Ding; Methodology, Shucheng Huang; Resources, Junna Wang; Software, Jie Lin; Supervision, Junna Wang; Visualization, Yuan Dong; Writing – original draft, Jie Lin.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was funded by Medical and Health Research Project of Zhejiang Province (2023KY853); Zhejiang Traditional Chinese Medicine Science and Technology Project (2023ZL393).

Declaration of Conflicting Interest

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Appendix

References

Schabath

Cote

. Cancer progress and priorities: lung cancer. Cancer Epidemiol Biomarkers Prev. 2019;28(10):1563-1579.

Wang

Zimmermann

Parikh

Mansfield

Adjei

. Current diagnosis and management of small-cell lung cancer. Mayo Clin Proc. 2019;94:1599-1622.

Travis

Brambilla

Burke

Marx

Nicholson

. Introduction to the 2015 World Health Organization classification of tumors of the lung, pleura, thymus, and heart. J Thorac Oncol. 2015;10:1240-1242.

Rudin

Poirier

. Small-cell lung cancer in 2016: shining light on novel targets and therapies. Nat Rev Clin Oncol. 2017;14:75-76.

Bunn

Jr. Minna

Augustyn

, et al.

Small cell lung cancer: can recent advances in biology and molecular biology be translated into improved outcomes?

J Thorac Oncol. 2016;11:453-474.

Combs

Hancock

Boffa

Decker

Detterbeck

Kim

. Bolstering the case for lobectomy in stages I, II, and IIIA small-cell lung cancer using the National Cancer Data Base. J Thorac Oncol. 2015;10(2):316-323.

Gaspar

McNamara

Gay

, et al. Small-cell lung cancer: prognostic factors and changing treatment over 15 years. Clin Lung Cancer. 2012;13(2):115-122.

Rudin

Ismaila

Hann

, et al. Treatment of small-cell lung cancer: American society of clinical oncology endorsement of the American college of chest physicians guideline. J Clin Oncol. 2015;33(34):4106-4111.

National Comprehensive Cancer Network . NCCN clinical practice guidelines in oncology: small cell lung cancer (version 2.2015). NCCN Guidelines; 2015.

10.

Yang

CFJ

Chan

Speicher

, et al. Long-term survival after surgery compared with concurrent chemoradiation for node-negative small cell lung cancer. Ann Surg. 2018;268(6):1105-1112.

11.

Coroller

Agrawal

Huynh

, et al. Radiomic-based pathological response prediction from primary tumors and lymph nodes in NSCLC. J Thorac Oncol. 2017;12:467-476.

12.

Toyokawa

Kozuma

Matsubara

, et al. Radiological features of the surgically resected small-sized small-cell lung cancer on computed tomography. Anticancer Res. 2017;37(2):877-881.

13.

Linning

Zhan

Han

, et al. Radiomics for classification of lung cancer histological subtypes based on nonenhanced computed tomography. Acad Radiol. 2019;26(9):1245-1252.

14.

Huang

Liu

, et al. Radiomics signature: a potential biomarker for the prediction of disease-free survival in early-stage (I or II) non-small cell lung cancer. Radiology. 2016;281:947-957.

15.

Chamberlin

Kocher

Waltz

, et al. Automated detection of lung nodules and coronary artery calcium using artificial intelligence on low-dose CT scans for lung cancer screening: accuracy and prognostic value. BMC Med. 2021;19:1-14.

16.

Zhou

, et al. Effectiveness of CT radiomic features combined with clinical factors in predicting prognosis in patients with limited-stage small cell lung cancer. BMC Cancer. 2024;24(1):170.

17.

Wang

Kato

Oyama-Manabe

, et al. Identifying triple-negative breast cancer using background parenchymal enhancement heterogeneity on dynamic contrast-enhanced MRI: a pilot radiomics study. PLoS One. 2015;10:e0143308.

18.

Keller

Chen

Conant

Kontos

. Breast density and parenchymal texture measures as potential risk factors for estrogen-receptor positive breast cancer. Proc SPIE. 2014;9035:90351D.

19.

Shimada

Ishii

Hishida

Yoshida

Nishimura

Nagai

. Extratumoral vascular invasion is a significant prognostic indicator and a predicting factor of distant metastasis in non-small cell lung cancer. J Thorac Oncol. 2010;5:970-975.

20.

Braman

Etesami

Prasanna

, et al. Intratumoral and peritumoral radiomics for the pretreatment prediction of pathological complete response to neoadjuvant chemotherapy based on breast DCE-MRI. Breast Cancer Res. 2017;19(1):1-14.

21.

Collins

Reitsma

Altman

Moons

. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ (Pract Obs Ed). 2015;350:g7594.

22.

L Lambin

Rios-Velazquez

Leijenaar

, et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer. 2012;48:441-446.

23.

van Griethuysen

JJM

Fedorov

Parmar

, et al. Computational radiomics system to decode the radiographic phenotype. Cancer Res. 2017;77:e104-e107.

24.

Simon

Subramanian

Menezes

. Using cross-validation to evaluate predictive accuracy of survival risk classifiers based on high-dimensional data. Briefings Bioinf. 2011;12:203-214.

25.

Koo

. A guideline of selecting and reporting intraclass correlation coefficients for reliability research. J Chiropr Med. 2016;15(2):155-163.

26.

DeLong

Clarke-Pearson

. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44:837-845.

27.

Lin

Zheng

Jia

, et al. A meta-analysis of MRI radiomics-based diagnosis for BI-RADS 4 breast lesions. J Cancer Res Clin Oncol. 2024;150(5):254.

28.

Zhou

Wang

. Spectral CT imaging of lung cancer: quantitative analysis of spectral parameters and their correlation with tumor characteristics. Acad Radiol. 2018;11:1398-1404.

29.

El Naqa

. The role of quantitative PET in predicting cancer treatment outcomes. Clin Transl Imaging. 2014;4:305-320.

30.

Gillies

Kinahan

Hricak

. Radiomics: images are more than pictures; they are data. Radiology. 2016;278:563-577.

31.

Kim

Stolarska

Othmer

. The role of the microenvironment in tumor growth and invasion. Prog Biophys Mol Biol. 2011;106(2):353-379.

32.

Zhou

Wang

, et al. Evaluation of the prognosis in patients with small-cell lung cancer treated by chemotherapy using tumor shrinkage rate-based radiomics. Eur J Med Res. 2024;29(1):401.

33.

Chen

Jiang

, et al. Progression-free survival prediction in small cell lung cancer based on radiomics analysis of contrast-enhanced CT. Front Med. 2022;9:833283.

34.

Shah

Selby

Mukherjee

, et al. Machine learning radiomics model for early identification of small-cell lung cancer on computed tomography scans. JCO Clin Cancer Inform. 2021;5:746-757.

35.

Parmar

Grossmann

, et al. Exploratory study to identify radiomics classifiers for lung cancer histology. Front Oncol. 2016;6:71.

36.

Junior

Koenigkam-Santos

Cipriano

Fabro

Azevedo-Marques

. Radiomics-based features for pattern recognition of lung cancer histopathology and metastases. Comput Methods Progr Biomed. 2018;159:23-30.

37.

Liu

Zhao

, et al. Exploratory study of a CT radiomics model for the classification of small cell lung cancer and non-small-cell lung cancer. Front Oncol. 2020;10:1268.

38.

Jia

Zheng

Lin

, et al. Optimizing diagnosis and surgical decisions for chronic osteomyelitis through radiomics in the precision medicine era. Front Bioeng Biotech. 2024;12:1315398.

39.

Liu

Chen

Dang

. The role of surgery in stage I to III small cell lung cancer: a systematic review and meta-analysis. PLoS One. 2018;13(12):e0210001.

40.

Martucci

Baiano

Fontanella

, et al. Surgery in small-cell lung cancer. Cancers. 2021;13(3):390.

41.

Wang

Peng

Huang

, et al. The combination of CA125 and NSE is useful for predicting liver metastasis of lung cancer. Dis Markers. 2020;2020:8850873.

42.

Gkika

Büttner

Owczarek

, et al. Immunohistochemistry and radiomic features for survival prediction in small cell lung cancer. Front Oncol. 2020;10:1161.

43.

Grossmann

Stringfield

El-Hachem

, et al. Defining the biological basis of radiomic phenotypes in lung cancer. eLife. 2017;6:e23421.

44.

Dingemans

A-MC

Früh

Ardizzoni

, et al. Small-cell lung cancer: ESMO clinical practice guidelines for diagnosis, treatment, and follow-up. Ann Oncol. 2021;32(7):839-853.

45.

Ernani

Ganti

. Surgery for limited-stage small cell lung cancer: ready for prime time? J Thorac Dis. 2017;9(10):3576-3578.

46.

Stinchcombe

. Current treatments for surgically resectable, limited-stage, and extensive-stage small cell lung cancer. Oncologist. 2017;22(12):1510-1517.

47.

Detterbeck

Jones

Kernstine

Naunheim

American College of Chest Physicians . Invasive mediastinal staging of lung cancer: ACCP evidence-based clinical practice guidelines. Chest (Am Coll Chest Physicians). 2007;132(3 Suppl):202S-220S.

48.

de Langen

Raijmakers

Riphagen

Paul

Hooekstra

. The size of mediastinal lymph nodes and its relation with metastatic involvement: a meta-analysis. Eur J Cardio Thorac Surg. 2006;29:26-29.

49.

Andersen

Harders

Ganeshan

Thygesen

Torp Madsen

Rasmussen

. CT texture analysis can help differentiate between malignant and benign lymph nodes in the mediastinum in patients suspected for lung cancer. Acta Radiol. 2016;57(6):669-676.

Peritumoral Radiomic Features on CT for Differential Diagnosis in Small-Cell Lung Cancer: Potential for Surgical Decision-Making

Abstract

Keywords

Introduction

Materials and Methods

Data Collection

CT Scanning Protocol

Lesion Segmentation and Radiomics Feature Extraction

Image Selection

Lesion and Peritumoral Region Segmentation

Radiomics Feature Extraction, Selection, and Model Construction

Clinical and Clinical-Radiomics Nomogram Models

Statistical Analysis

Results

Reproducibility Analysis

Comparison of Clinical Factors Between SCLC and NSCLC Patients

Selection of Radiomics and Clinical Features

Model Performance Comparison

Discussion

Conclusions

Footnotes

ORCID iD

Ethial Approval

Patient Consent

Author Contributions

Funding

Declaration of Conflicting Interest

Appendix

References