Differentiating pulmonary sarcoidosis from mediastinal lymph node metastasis via conventional computed tomography features and quantitative imaging parameters: a retrospective study

Abstract

Background

Pulmonary sarcoidosis and mediastinal lymph node metastasis (MLNM) secondary to lung cancer frequently present with overlapping imaging features on chest computed tomography (CT), posing considerable diagnostic challenges. Although histopathological confirmation remains the gold standard, noninvasive CT-based strategies may provide supplementary information to aid in pretest probability estimation. This study evaluated the ability of a multivariate diagnostic framework comprising certain conventional quantitative CT parameters to differentiate pulmonary sarcoidosis from MLNM.

Methods

This retrospective, single-center study included 125 patients with histopathologically confirmed diagnoses who underwent standardized chest CT between January 2022 and December 2024, comprising 40 patients with pulmonary sarcoidosis and 85 with lung cancer involving MLNM. Three conventional CT features were measured for the single largest mediastinal lymph node per patient: short-axis diameter (mm), contrast enhancement value [in Hounsfield units (HU)], and anatomical nodal station classified according to the International Association for the Study of Lung Cancer (IASLC) lymph node station map. Baseline characteristics were compared via the independent samples t-test for continuous variables and the Chi-squared test for categorical variables. Three combinatorial multivariate logistic regression models were constructed and internally validated via repeated 10-fold cross-validation. Diagnostic performance was evaluated through receiver operating characteristic (ROC) curve analysis, with the area under the curve (AUC), sensitivity, specificity, and Youden index being calculated for each model. The 95% confidence intervals (CIs) for AUCs were computed via the DeLong method. A two-sided P value <0.05 was considered statistically significant.

Results

Significant differences were observed between the sarcoidosis and lung cancer groups in terms of age (51.16±12.85 vs. 64.24±11.26 years; t=−4.628; P<0.001), sex distribution (female: 70.0% vs. 22.3%; χ²=17.563; P<0.001), and lymph node imaging characteristics. Among individual CT parameters, contrast enhancement value achieved the highest discriminative performance (AUC =0.785; 95% CI: 0.745–0.825; sensitivity =66.3%; specificity =90.1%; optimal cutoff =30.501 HU), while that of anatomical nodal distribution alone was poor (AUC =0.505; 95% CI: 0.460–0.549). The inverse of short-axis diameter demonstrated modest performance (AUC =0.586; 95% CI: 0.543–0.629). The composite model integrating inverse short-axis diameter, enhancement value, and nodal location achieved the highest overall performance, with an AUC of 0.798 (95% CI: 0.749–0.868), a sensitivity of 66.2%, a specificity of 94.8%, and a Youden index of 0.610 (P<0.001).

Conclusions

Although the composite CT-based model demonstrated high specificity, the moderate sensitivity indicates that approximately one-third of malignant cases would not be correctly identified by imaging criteria alone. These CT-derived quantitative parameters are thus not a reliable substitute for histopathological confirmation. Patients with indeterminate mediastinal lymphadenopathy should be referred for tissue sampling—preferably via endobronchial ultrasound-guided transbronchial needle aspiration (EBUS-TBNA)—regardless of CT imaging findings. The proposed model should be considered a supplementary, hypothesis-generating research tool.

Introduction

Pulmonary sarcoidosis and mediastinal lymph node metastasis (MLNM) secondary to lung cancer are two distinct pathological conditions that commonly involve the mediastinal lymph nodes. Despite their different etiologies—sarcoidosis being a systemic granulomatous inflammatory disease and lung cancer a malignant epithelial neoplasm—these entities often exhibit overlapping radiological features, particularly in the form of mediastinal lymphadenopathy (1,2). On chest computed tomography (CT), both conditions may present with enlarged mediastinal lymph nodes, making reliable differentiation challenging in routine clinical settings (3). Although histopathological confirmation remains the diagnostic gold standard, mediastinal lymph node biopsy is invasive, technically complex, and associated with procedural risks (4). Sampling errors and interobserver variability may further limit diagnostic reliability in certain clinical contexts. These considerations highlight the potential value of accurate, reproducible, and noninvasive imaging strategies as supplementary tools in the evaluation of mediastinal lymphadenopathy.

Previous studies have evaluated conventional CT-based parameters—including lymph node short-axis diameter, contrast enhancement patterns, and anatomical distribution—in their ability to differentiate benign from malignant mediastinal lymphadenopathy (5). For example, bilateral symmetrical hilar and mediastinal lymphadenopathy with homogeneous postcontrast enhancement is a recognized imaging pattern associated with pulmonary sarcoidosis, whereas asymmetrical nodal enlargement with heterogeneous enhancement and irregular margins is more commonly observed in malignant disease. Although these qualitative and quantitative CT features offer a degree of diagnostic utility, their sensitivity and specificity remain suboptimal, particularly in early-stage disease or atypical presentations. Furthermore, few studies have systematically combined multiple conventional CT parameters within a validated multivariate diagnostic framework specifically for the purpose of differentiating pulmonary sarcoidosis from MLNM in patients with lung cancer.

This study thus aimed to evaluate the diagnostic utility of conventional quantitative CT imaging parameters—specifically lymph node short-axis diameter, contrast enhancement value, and anatomical nodal distribution—and to determine whether a multiparametric combination of these features could improve the noninvasive characterization of imaging differences between pulmonary sarcoidosis and MLNM in lung cancer. Importantly, the proposed model is not intended to replace histopathological confirmation, which remains the only reliable means of establishing a definitive diagnosis in patients with indeterminate mediastinal lymphadenopathy. All such patients should be referred for tissue sampling via endobronchial ultrasound-guided transbronchial needle aspiration (EBUS-TBNA), irrespective of CT imaging findings (6). We present this article in accordance with the STARD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1-2581/rc).

Methods

Study design and patient selection

This single-center retrospective study was conducted at Heilongjiang University of Chinese Medicine between January 2022 and December 2024. A total of 127 patients were initially screened, 2 of whom were excluded: one due to poor CT image quality caused by motion artifacts and one due to a coexisting intrathoracic condition that could interfere with image interpretation. The remaining 125 patients with histopathologically confirmed diagnoses were included in the final analysis, comprising 40 cases of pulmonary sarcoidosis and 85 cases of lung cancer with MLNM.

The inclusion criteria were as follows: (I) availability of diagnostic-quality chest CT imaging with both non-contrast-enhanced and contrast-enhanced phases and (II) definitive histopathological diagnosis of either pulmonary sarcoidosis or lung cancer with MLNM based on EBUS-TBNA, CT-guided biopsy, or surgical pathology. Meanwhile, the exclusion criteria were as follows: (I) poor CT image quality due to motion artifacts or incomplete scanning and (II) coexisting intrathoracic conditions that could interfere with image interpretation, including active tuberculosis, other granulomatous diseases, or concurrent neoplastic processes. All included patients had complete clinical, imaging, and pathological data available for analysis.

CT image acquisition

All patients underwent chest CT examination with a Revolution Frontier CT scanner (GE HealthCare, Chicago, IL, USA) under the following standardized acquisition parameters: tube voltage, 120 kVp; automatic tube current modulation, 100–250 mA; slice thickness, 1.25 mm; reconstruction interval, 1.0 mm; and matrix size, 512×512. Contrast-enhanced imaging was performed following the intravenous administration of iodinated contrast medium, with images being acquired during both arterial and venous phases. All images were archived in Digital Imaging and Communications in Medicine (DICOM) format and retrieved for analysis from the institutional picture archiving and communication system (PACS) (7).

CT feature measurement

For each patient, the single largest mediastinal lymph node, determined according to the maximal short-axis diameter and visibility on CT, was selected as the target lesion for analysis. This approach was adopted to ensure analytical consistency and to minimize within-patient statistical redundancy. We acknowledge, however, that analysis of the single largest node may not fully capture the multistation and bilaterally symmetrical distribution patterns characteristic of pulmonary sarcoidosis.

Two radiologists with 5 or more years of thoracic imaging experience who were blinded to the final pathological diagnosis independently performed manual measurement and assessment of each target lymph node. Interobserver discrepancies were resolved by consensus discussion. Three conventional CT imaging features, including short-axis diameter (mm), contrast enhancement value [in Hounsfield units (HU)], and contrast enhancement value (HU), were recorded for each target lymph node. The short-axis diameter was measured on the axial CT image at the level of maximum nodal cross-section and perpendicular to the long axis of the node. Contrast enhancement was defined as the difference between the attenuation value on the venous phase postcontrast image and the corresponding noncontrast image at the same nodal level. It was measured with a circular region of interest (ROI) being placed within the lymph node parenchyma, with areas of necrosis or calcification being avoided. Anatomical nodal distribution was classified according to the International Association for the Study of Lung Cancer (IASLC) lymph node station map, with stations categorized as follows: stations 1–4, upper mediastinal; stations 5–6, aortopulmonary; station 7, subcarinal; stations 8–9, lower mediastinal; and stations 10–11, hilar/lobar.

No automated feature extraction, texture analysis, or high-dimensional computational imaging procedures were performed in this study.

Model construction and validation

A multivariate logistic regression framework was employed for diagnostic model construction. Based on prior literature and clinical reasoning, including established CT size criteria for mediastinal lymph nodes and lung cancer staging recommendations, the inverse of short-axis diameter (rather than the raw diameter) was used as a predictor variable, reflecting the inverse relationship between nodal size and the likelihood of a sarcoidosis diagnosis in this cohort (8,9). Contrast enhancement value and anatomical nodal location were entered as additional predictor variables.

Three combinatorial logistic regression models were constructed and compared: Model 1, comprising the inverse short-axis diameter and nodal location; Model 2, comprising contrast enhancement value and nodal location; and Model 3 (the composite multiparametric model), comprising inverse short-axis diameter, contrast enhancement value, and nodal location.

No automated feature selection procedures (e.g., least absolute shrinkage and selection operator regression) were employed. All three candidate features were entered into the respective models based on a priori clinical relevance and routine CT availability.

Model performance was internally validated via repeated 10-fold cross-validation, in which the dataset was randomly partitioned into 10 equal subsets; the model was trained on 9 subsets and tested on the remaining subset in each iteration, with this procedure repeated to reduce overfitting and provide a more reliable estimate of out-of-sample predictive performance.

Assessment of diagnostic performance

The diagnostic performance for each individual CT parameter and each combinatorial model was evaluated via receiver operating characteristic (ROC) curve analysis. The following metrics were calculated: area under the curve (AUC), sensitivity, specificity, and Youden index (defined as sensitivity + specificity − 1). Optimal diagnostic cutoff values for continuous parameters were determined by maximizing the Youden index. The 95% confidence interval (CIs) for AUCs were calculated via the DeLong method.

Statistical analysis

Statistical analyses were performed with R version 3.5.1 (The R Foundation for Statistical Computing, Vienna, Austria). Continuous variables are expressed as the mean ± standard deviation and were compared between groups via the independent samples t-test after approximate normality was confirmed. Categorical variables are summarized as frequencies and percentages and were compared via the Pearson Chi-squared test or Fisher exact test as appropriate based on expected cell counts. A two-sided P value of less than 0.05 was considered statistically significant for all analyses. ROC curve analyses and logistic regression modeling were performed with the pROC and stats packages in R, respectively.

Ethical considerations

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments and was approved by the Ethics Committee of Heilongjiang University of Chinese Medicine (Harbin, China). The requirement for written informed consent was waived by the Ethics Committee due to retrospective nature of the analysis and the use of fully anonymized patient data.

Results

Patient enrollment and grouping

A total of 127 patients were initially screened between January 2022 and December 2024. Two patients were excluded from the final analysis: one due to poor CT image quality caused by motion artifacts and one due to a coexisting intrathoracic condition (concurrent active tuberculosis) that could interfere with CT image interpretation. The remaining 125 patients with histopathologically confirmed diagnoses were allocated into two groups: pulmonary sarcoidosis (n=40) and lung cancer with MLNM (n=85). All 125 patients underwent standardized chest CT imaging and were included in the final analysis. The patient enrollment and allocation process is illustrated in Figure 1.

Figure 1.

Study flowchart of patient enrollment, group allocation, image processing, and analysis. ROC, receiver operating characteristic.

Imaging examples of representative cases

Representative CT images of pulmonary sarcoidosis and lung cancer with MLNM are illustrated in Figures 2,3, respectively. Each figure displays images from three acquisition phases presented from left to right: arterial phase, venous phase, and noncontrast phase.

Figure 2.

Representative CT images of a patient with pulmonary sarcoidosis. From left to right: arterial phase, venous phase, and noncontrast phase. Note the bilaterally symmetrical enlargement of mediastinal and hilar lymph nodes with homogeneous post-contrast enhancement and well-defined margins, consistent with the characteristic CT appearance of pulmonary sarcoidosis. CT, computed tomography.

Figure 3.

Representative CT images of a patient with lung cancer and mediastinal lymph node metastasis. From left to right: arterial phase, venous phase, and noncontrast phase. Note the asymmetrical nodal distribution with heterogeneous post-contrast enhancement, irregular margins, and focal areas of central necrosis visible on the venous phase image, consistent with malignant lymph node involvement. CT, computed tomography.

In the sarcoidosis case shown in Figure 2, the mediastinal lymph nodes demonstrate bilaterally symmetrical enlargement with homogeneous postcontrast enhancement and well-defined margins, involving the bilateral hilar and mediastinal stations. In contrast, the metastatic lymph nodes in the lung cancer case shown in Figure 3 exhibit asymmetrical distribution with heterogeneous enhancement, irregular margins, and focal areas of central necrosis visible particularly on the venous phase images. These qualitative radiological distinctions are consistent with the quantitative differences in enhancement value and nodal distribution observed across the study cohort and support the rationale of incorporating these features as candidate diagnostic parameters.

Baseline characteristics of the study population

A total of 125 patients were included in the final analysis: 40 with pulmonary sarcoidosis and 85 with lung cancer accompanied by MLNM. Significant differences were observed between the two groups for both demographic and imaging characteristics. The mean age of patients in the sarcoidosis group was significantly lower than that in the lung cancer group (51.16±12.85 vs. 64.24±11.26 years; P<0.001). The sex distribution also differed markedly between groups: 70.0% and 22.3% of patients were female in the sarcoidosis group and lung cancer group, respectively (P<0.001) (Table 1). These demographic differences are consistent with the established epidemiological profiles of both conditions: pulmonary sarcoidosis predominantly affects younger individuals with a female predominance, whereas lung cancer with mediastinal involvement is more commonly observed in older patients. It should be noted, however, that these demographic variables were not incorporated as covariates in the diagnostic model, which represents a potential source of confounding that is further discussed in the Limitations section.

Table 1. Comparison of baseline characteristics between the two patient groups.

Characteristic	Sarcoidosis group (n=40)	Lung cancer group (n=85)	χ2 (t)	P value
Sex
Male	12 (30.0)	66 (77.6)
Female	28 (70.0)	19 (22.3)	17.563	<0.001
Age (years)	51.16±12.85	64.24±11.26	−4.628†	<0.001

Continuous variables are presented as mean ± standard deviation and were compared via the independent samples t-test. Categorical variables are presented as frequencies (percentages) and were compared via the Chi-squared test or Fisher’s exact test, as appropriate. A two-sided P value <0.05 was considered statistically significant. ^†, t-test used for age comparison. A comparison of baseline demographic and clinical characteristics between patients with pulmonary sarcoidosis and patients with lung cancer with mediastinal lymph node metastasis was conducted.

Distribution of lymph node stations

Analysis of lymph node station involvement revealed distinct anatomical distribution patterns between the two groups (Table 2). In the sarcoidosis group, lymph node involvement was most frequently identified at hilar and lobar stations (stations 10 and 11), with additional involvement at the upper mediastinal stations (stations 2, 3, and 4), reflecting the characteristically symmetrical and bilateral hilar lymphadenopathy associated with sarcoidosis. In contrast, the lung cancer group demonstrated a predominance of lymph node involvement at station 4 (lower paratracheal), consistent with the established patterns of locoregional metastatic spread in lung cancer (10). These differences in anatomical nodal distribution between the two groups suggest that nodal location may have discriminatory value; in the ROC analysis, location alone yielded an AUC of 0.505, indicating that anatomical distribution is insufficient as a standalone diagnostic criterion and its utility is primarily realized when combined with other CT parameters.

Table 2. Comparison of lymph node station distribution between the two patient groups.

Lymph node station	Sarcoidosis group (n=336)	Lung cancer group (n=548)
Station 2	50 (14.8)	66 (12.0)
Station 3	52 (15.4)	54 (9.8)
Station 4	56 (16.7)	109 (19.8)
Station 5	24 (7.1)	21 (3.8)
Station 6	13 (3.9)	10 (1.8)
Station 7	15 (4.4)	82 (14.9)
Station 10	66 (19.6)	106 (19.3)
Station 11	60 (17.8)	100 (18.2)

Comparison of mediastinal lymph node station distribution classified according to the IASLC lymph node station map between the pulmonary sarcoidosis group and the lung cancer with mediastinal lymph node metastasis group. Data are presented as number or frequencies (percentages). Station involvement was determined based on the target lymph node selected for quantitative analysis in each patient. IASLC, International Association for the Study of Lung Cancer.

Diagnostic value of CT imaging features

ROC curve analysis was performed to evaluate the diagnostic performance of individual CT parameters and their combinations in differentiating pulmonary sarcoidosis from MLNM (Table 3 and Figure 4).

Table 3. ROC curve analysis for the diagnosis of pulmonary sarcoidosis based on lymph node short axis, enhancement value, and distribution.

Variable	Cutoff value	Youden index	Sensitivity (%)	Specificity (%)	AUC (95% CI)
Inverse of short-axis diameter	1.074	0.168	37.5	79.4	0.586 (0.543–0.629)
Enhancement value	30.501	0.663	66.3	90.1	0.785 (0.745–0.825)
Location	10.503	0.155	15.3	90	0.505 (0.460–0.549)
Inverse of short-axis location	2.016	0.608	65.5	94.2	0.788 (0.749–0.828)
Inverse of diameter location	1.822	0.165	46.9	69.1	0.589 (0.546–0.631)
Enhancement location	1.937	0.613	77.4	90.3	0.786 (0.745–0.826)
Composite index (triple inverse)	1.995	0.606	80.2	94.8	0.798 (0.749–0.868)

Diagnostic performance of individual CT imaging parameters and combinatorial logistic regression models for the differentiation of pulmonary sarcoidosis from mediastinal lymph node metastasis as evaluated via ROC curve analysis. Performance metrics include AUC with the 95% CI, sensitivity, specificity, Youden index, and optimal diagnostic cutoff value for each parameter and model. Optimal cutoff values were determined by maximizing the Youden index (sensitivity + specificity − 1). AUC, area under the curve; CI, confidence interval; CT, computed tomography; ROC, receiver operating characteristic.

Figure 4.

ROC curves of lymph node short axis, enhancement value, distribution, and combined parameters for the diagnosis of pulmonary sarcoidosis. Curves are shown for the following: inverse short-axis diameter alone (AUC =0.586); contrast enhancement value alone (AUC =0.785); anatomical nodal location alone (AUC =0.505); Model 1, comprising inverse short-axis diameter and nodal location (AUC =0.788); Model 2 comprising enhancement value and nodal location (AUC =0.786); and the composite Model 3 (“triple inverse” model), comprising inverse short-axis diameter, enhancement value, and nodal location (AUC =0.798; 95% CI: 0.749–0.868). The diagonal reference line represents chance-level discrimination (AUC =0.500). AUC, area under the curve; CI, confidence interval; ROC, receiver operating characteristic.

Among the individual CT parameters, contrast enhancement value demonstrated the highest discriminative performance, with an AUC of 0.785, a sensitivity of 66.3%, and a specificity of 95.1%. The inverse of short-axis diameter yielded a modest AUC of 0.586, while anatomical nodal location alone demonstrated negligible discriminative value (AUC =0.505), performing at a near-chance level.

The combinatorial models demonstrated incrementally improved diagnostic performance compared with single-parameter analyses. Model 1, combining inverse short-axis diameter with nodal location, achieved an AUC of 0.788. Model 2, combining enhancement value with nodal location, yielded an AUC of 0.786. The highest overall performance was achieved by Model 3—the composite “triple inverse” model integrating inverse short-axis diameter, contrast enhancement value, and nodal location—which achieved an AUC of 0.798 (95% CI: 0.749–0.868), with a sensitivity of 66.2% and a specificity of 94.8% (Youden index =0.610).

These results demonstrate that combining multiple conventional CT parameters improves diagnostic discrimination as compared with any single feature used alone. However, it is essential to interpret these findings with appropriate caution in terms of clinical application. The composite model achieved a sensitivity of only 66.2%, indicating that approximately one-third of malignant cases would not be correctly identified by CT imaging criteria alone. This level of diagnostic sensitivity is categorically insufficient to justify withholding histopathological confirmation in any clinical scenario. The high specificity (94.8%) suggests that the model may provide supplementary support for a benign (sarcoidosis) diagnosis in appropriately selected cases, but it cannot reliably exclude malignancy and should not be used as a standalone decision-making tool.

Discussion

In this study, we evaluated the ability of a multiparametric diagnostic model incorporating three conventional CT imaging parameters—lymph node short-axis diameter, contrast enhancement value, and anatomical nodal distribution—to characterize the imaging differences between pulmonary sarcoidosis and MLNM in lung cancer. No automated feature extraction, radiomic analysis, or machine learning procedures were employed; the model was constructed with standard quantitative CT measurements within a multivariate logistic regression framework. Among the individual parameters, contrast enhancement value demonstrated the highest discriminative performance (AUC =0.785), consistent with its established role in differentiating benign from malignant lymph nodes based on vascular and perfusion characteristics. The composite “triple inverse” model achieved the highest overall diagnostic performance (AUC =0.798; sensitivity 66.2%; specificity 94.8%), with the multiparametric combination of conventional CT features outperforming any single imaging parameter in this diagnostic task.

However, these findings must be interpreted within a clinically and ethically appropriate framework. The moderate sensitivity of the composite model (66.2%) indicates that CT-based quantitative imaging parameters alone are insufficient for reliably distinguishing pulmonary sarcoidosis from MLNM with the diagnostic fidelity required for clinical decision-making. Approximately one-third of malignant cases would not be correctly identified by imaging criteria alone, and no CT-based model currently available can exclude metastatic lymph node involvement with acceptable confidence. The principal conclusion of this study is therefore that CT features cannot replace histopathological confirmation, and all patients with indeterminate mediastinal lymphadenopathy should be referred for tissue sampling via EBUS-TBNA regardless of imaging appearances. The model developed in this study should be regarded strictly as a research tool for characterizing CT imaging phenotypes and generating hypotheses for future investigation.

Comparison with previous studies

The diagnostic utility of conventional CT features in distinguishing sarcoidosis from malignancy has been well-documented. Bilateral and symmetrical hilar or mediastinal lymphadenopathy, along with homogeneous enhancement, is commonly associated with pulmonary sarcoidosis, whereas metastatic lymph nodes in lung cancer tend to exhibit asymmetric enlargement and heterogeneous contrast enhancement. These qualitative observations have formed the basis for noninvasive diagnostic impressions in clinical practice (11-13).

However, few studies have systematically integrated conventional CT features with radiomic analysis for the specific task of differentiating sarcoidosis from MLNM. The bulk of the related research has either focused on qualitative descriptors or examined radiomics in isolation, limiting their practical impact and generalizability.

Our findings are consistent with the previously reported imaging trends, confirming the diagnostic value of enhancement patterns and nodal distribution. Importantly, this study extended previous work by quantitatively assessing diagnostic performance through AUC analysis and validating a multifeature logistic regression model. Compared to previously described approaches, this methodological integration offers a more objective and reproducible framework for imaging-based differential diagnosis.

Pulmonary sarcoidosis is a systemic granulomatous disease with an estimated global incidence of 10–14.8 per 100,000 persons annually, with notable variation across geographic regions and ethnic populations (14,15). The disease predominantly affects individuals between the ages of 20 and 60 years, with a slight female predominance observed in many cohorts—an epidemiological pattern consistent with the demographic characteristics of the sarcoidosis group in our study (16). On CT imaging, sarcoidosis classically manifests as bilateral, symmetrical hilar and mediastinal lymphadenopathy and frequently involves stations 10 and 11, with homogeneous post–contrast enhancement and well-defined nodal margins (17). These imaging hallmarks, although not pathognomonic, provide an important foundation for the development of CT-based diagnostic criteria and informed the feature selection strategy employed in our study.

The CT characterization of malignant mediastinal lymphadenopathy in lung cancer has been extensively studied. Short-axis diameter exceeding 10 mm has historically served as a threshold criterion for nodal malignancy, although this metric alone provides limited sensitivity and specificity (18). Heterogeneous enhancement patterns, central necrosis, irregular margins, and asymmetrical station involvement are additional CT features associated with malignant lymphadenopathy, particularly in non-small cell lung cancer (19). Notably, several studies have demonstrated that a substantial proportion of lymph nodes that appear morphologically normal on CT may nonetheless harbor micrometastatic disease, underscoring the fundamental limitation of morphological CT assessment as a standalone diagnostic tool (20). These observations are consistent with the moderate sensitivity (66.2%) observed in our study and reinforce the conclusion that CT-based models cannot serve as a replacement for tissue confirmation.

Comparison with advanced imaging modalities

Beyond conventional CT, positron emission tomography-CT (PET-CT) has emerged as an important adjunct modality for the evaluation of mediastinal lymphadenopathy. PET-CT leverages the metabolic activity of lymph nodes—quantified by standardized uptake value—to differentiate hypermetabolic malignant nodes from less metabolically active benign lesions (21). However, pulmonary sarcoidosis is itself associated with markedly elevated fluorodeoxyglucose uptake in affected lymph nodes, which can closely mimic the PET-CT appearance of malignant mediastinal disease and lead to diagnostic misclassification (22). This phenomenon—sometimes referred to as “sarcoid mimicking malignancy” on PET-CT—further illustrates the diagnostic challenge addressed in our study and highlights the need for multimodal, integrative approaches as opposed to a reliance on any single imaging technique. Our CT-based model, while complementary to PET-CT in characterizing nodal morphology and enhancement, does not resolve the fundamental ambiguity between sarcoidosis and metastatic disease at the imaging level, further supporting the indispensability of histopathological confirmation.

Clinical implications

Although the proposed CT-based multiparametric model demonstrated a statistically significant discriminatory performance, it is essential to interpret these findings within a clinically and ethically appropriate framework. The combined model achieved a sensitivity of only 66.2%, meaning that approximately one-third of malignant cases could potentially be misclassified as benign by imaging criteria alone. Based on the sarcoidosis prevalence of approximately 32% in the cohort, this translates to an estimated 14 false-negative cases out of 125 patients—individuals whose lymph nodes may appear imaging-consistent with sarcoidosis but in fact harbor metastatic cancer. This high false-negative rate is clinically unacceptable and categorically insufficient to justify foregoing tissue diagnosis. Even lymph nodes demonstrating radiologically benign features may carry metastatic disease, and no currently available CT-based model can exclude malignancy with the fidelity required for clinical decision-making. Therefore, the principal conclusion of this study is that CT features alone cannot reliably distinguish pulmonary sarcoidosis from MLNM. All patients presenting with indeterminate mediastinal lymphadenopathy should be referred for histopathological confirmation, regardless of CT imaging findings. The current gold-standard approach for mediastinal lymph node sampling is EBUS-TBNA, as recommended by both the American Thoracic Society Interventional Pulmonary Guidelines and the 2020 Sarcoidosis Clinical Practice Guideline. Mediastinoscopy is no longer the recommended modality for this indication, and mediastinal lymph nodes are almost always accessible via bronchoscopy. It is exceedingly rare for a patient to be unable to tolerate bronchoscopy even in the setting of significant medical comorbidities. The CT-based model developed in our study should therefore be regarded strictly as a research tool for characterizing imaging phenotypes and generating hypotheses regarding the relationship between CT features and underlying pathology. It should not be used to guide decisions to forgo biopsy under any clinical circumstance.

Recent studies have examined various noninvasive techniques for differentiating between lung nodules and mediastinal lymph node metastases in patients with lung cancer. TBNA and EBUS have demonstrated high diagnostic accuracy and safety for assessing mediastinal and hilar lymph nodes (23). Additionally, diffusion-weighted magnetic resonance imaging and apparent diffusion coefficient values have been investigated regarding their ability to identify metastatic involvement in patients with non-small cell lung cancer (24). Gemstone spectral CT has also shown promise in this regard, with arterial phase spectral curve slope improving diagnostic efficiency in the preoperative evaluation of mediastinal lymphadenopathy (25).

The accuracy of EBUS-TBNA in diagnosing mediastinal lymph node evaluation has been well-established in large prospective series and meta-analyses. EBUS-TBNA has a sensitivity ranging from 88% to 93% and a specificity approaching 100% for the diagnosis of malignant mediastinal lymphadenopathy in lung cancer, with a complication rate of less than 1% in experienced centers (26). For pulmonary sarcoidosis, EBUS-TBNA with rapid on-site cytological evaluation has demonstrated a diagnostic yield exceeding 80%, and thus it is the preferred minimally invasive procedure for establishing a tissue diagnosis in patients with suspected sarcoidosis and mediastinal lymphadenopathy (27). These data confirm that EBUS-TBNA is both highly accurate and minimally invasive, further reinforcing the recommendation that all patients with indeterminate mediastinal lymphadenopathy should undergo bronchoscopic tissue sampling and that a reliance on imaging-based models alone is insufficient.

In this context, we developed a noninvasive diagnostic model based on conventional CT-derived parameters, including short-axis diameter, enhancement values, and lymph node location, for distinguishing pulmonary sarcoidosis from MLNM. The proposed model demonstrated high specificity (up to 94.8%) and acceptable sensitivity, indicating its potential utility in guiding clinical decision-making and reducing reliance on invasive procedures such as mediastinoscopy or EBUS-guided biopsy. By leveraging standard imaging features that are readily available in routine CT, this strategy supports individualized diagnostic planning, particularly in cases of indeterminate thoracic lymphadenopathy.

It is important to emphasize that the proposed CT-based model is not intended to replace histopathological confirmation, which remains the diagnostic gold standard. Rather, it should be regarded as a noninvasive supportive tool for estimating pretest probability, guiding individualized follow-up strategies, and informing decision-making in patients who are considered poor candidates for invasive procedures due to significant comorbidities, high procedural risk, or anatomically inaccessible lesions. In such clinical scenarios, the model may assist in prioritizing the need for biopsy and reducing unnecessary invasive interventions.

The combined diagnostic model demonstrated high specificity (94.8%) but only moderate sensitivity (66.2%), a performance profile that carries distinct clinical implications. High specificity implies that when the model yields a positive result for sarcoidosis, it does so with considerable confidence, thus supporting a benign diagnosis and potentially reducing the need for invasive confirmation in appropriately selected patients. However, the moderate sensitivity indicates that the model has limited capacity to exclude malignancy, and a substantial proportion of lung cancer cases with mediastinal metastasis may not be correctly identified by imaging features alone. Therefore, this model should not be used as a standalone tool to rule out malignant lymphadenopathy. Its primary clinical utility lies in increasing diagnostic confidence for benign (sarcoid) etiology in patients presenting with indeterminate mediastinal lymphadenopathy, particularly in settings where invasive biopsy carries elevated risk.

Limitations

Several limitations of this study should be acknowledged in the interpretation of the findings. First, the retrospective, single-center design might have introduced selection bias and limit the generalizability of the results to broader and more heterogeneous patient populations. Moreover, the study was conducted at a single tertiary military medical center, and the patient population may not be representative of community-based or multicenter clinical settings with different disease prevalence distributions, referral patterns, or imaging protocols.

Second, the sample size was relatively small, particularly in the sarcoidosis group (n=40), which may reduce the statistical power and stability of the diagnostic model. A larger sample size would provide more reliable estimates of model performance and allow for more robust subgroup analyses. In particular, the relatively low prevalence of sarcoidosis in this cohort (approximately 32%) should be considered in the extrapolation of the model performance metrics—including positive and negative predictive values—to clinical settings with a different disease prevalence.

Third, model development and validation were performed exclusively via internal repeated 10-fold cross-validation, without an independent external test cohort. In the absence of external validation, the proposed diagnostic model should be regarded as exploratory and hypothesis-generating. The reported performance characteristics—including AUC, sensitivity, and specificity—were derived from internal cross-validation only and may not be applicable to other institutions, patient populations, or CT acquisition protocols. Prospective, multicenter validation studies are therefore essential before the model can be considered for broader clinical adoption.

Fourth, although CT feature assessment was conducted by two experienced thoracic radiologists, manual measurement and visual interpretation of lymph node parameters remain subject to interobserver variability. Common statistics of interobserver agreement (e.g., intraclass correlation coefficient for continuous measurements and Cohen’s kappa for categorical assessments) were not examined in this study, which represents a methodological limitation. The use of standardized measurement protocols or semiautomated segmentation tools in future studies may improve reproducibility and facilitate broader clinical applicability.

Fifth, significant differences in age and sex distribution were observed between the two groups, consistent with the established epidemiological profiles of pulmonary sarcoidosis and lung cancer. However, these demographic variables were not incorporated as covariates in the multivariate logistic regression model, which represents a potential source of confounding. It is possible that some of the observed differences in CT imaging features between groups partially reflect underlying demographic characteristics rather than purely disease-specific imaging phenotypes. In future studies, it may be helpful to include age and sex as covariates in model construction or to stratify analyses with these variables to better isolate the independent diagnostic contribution of imaging features and assess model robustness across demographic subgroups.

Sixth, CT feature analysis in this study was restricted to the single largest mediastinal lymph node per patient. Although this approach ensures analytical consistency and avoids the statistical complexity associated with multiple correlated measurements per subject, it may not adequately capture the characteristic multistation and bilaterally symmetrical lymphadenopathy that is a hallmark of pulmonary sarcoidosis. Restricting analysis to a single node might therefore have underestimated the spatial distribution information that carries meaningful diagnostic value in sarcoidosis. Future studies incorporating multinode analysis, spatial distribution metrics, or graph-based nodal mapping approaches may provide more comprehensive and diagnostically informative characterization of lymph node involvement patterns in patients with sarcoidosis or malignant mediastinal disease.

Seventh, only three conventional CT parameters were examined in this study, and other potentially informative imaging features—such as lymph node shape descriptors, margin characteristics, internal architecture (e.g., calcification and necrosis), and dynamic contrast enhancement kinetics—were not systematically assessed. Furthermore, advanced computational imaging approaches such as radiomics, which enables the extraction of high-dimensional texture and shape features from segmented nodal volumes, were not employed and represent an important avenue for future investigation.

Finally, this study focused exclusively on the differentiation of pulmonary sarcoidosis from lung cancer with MLNM. The diagnostic performance of the proposed CT parameters in distinguishing sarcoidosis from other causes of benign mediastinal lymphadenopathy—such as active tuberculosis, fungal infection, or sarcoid-like reactions associated with immune checkpoint inhibitor therapy—remains unknown and should be specifically investigated.

Future directions

Future research should prioritize the external validation of the proposed diagnostic model in multicenter, prospective cohorts encompassing a diversity of clinical environments, CT acquisition systems, patient demographics, and disease prevalence distributions. Such studies will be critical for rigorously assessing the generalizability and clinical reliability of the CT-based multiparametric approach described here.

Future investigations should also assess the incremental diagnostic value of additional conventional CT parameters not examined in this study, including lymph node shape descriptors, margin regularity, internal architecture features (calcification, necrosis, and fat content), and dynamic contrast enhancement kinetics. Incorporating these features into an expanded multivariate model may further improve the diagnostic discrimination between sarcoidosis and malignant mediastinal disease.

Advanced computational imaging approaches—particularly radiomics, involving the systematic extraction of high-dimensional, mathematically derived image features from segmented lymph node volumes—represent a promising complementary strategy that could be integrated with the conventional CT parameters evaluated in our study (28,29). Radiomic texture and shape features may capture subvisual tissue heterogeneity imperceptible to human observers, potentially providing incremental diagnostic information beyond standard quantitative CT measurements.

Comparative studies involving other benign causes of mediastinal lymphadenopathy—including active tuberculosis, nontuberculous mycobacterial infection, fungal lymphadenitis, and sarcoid-like reactions associated with immune checkpoint inhibitor therapy—would help clarify the disease-specificity of the CT-based indicators identified in our study and establish their utility in broader real-world diagnostic contexts.

Standardization of CT image acquisition protocols, lymph node segmentation procedures, and feature measurement methodologies across institutions will be essential to facilitating meaningful multicenter comparisons and promoting the development of externally validated, clinically deployable diagnostic tools. The development and validation of semiautomated or fully automated lymph node segmentation and measurement pipelines may substantially improve reproducibility and reduce observer-dependent variability.

Finally, prospective studies incorporating demographic variables—including age, sex, smoking history, and occupational exposure—as covariates alongside CT imaging features may provide a more comprehensive and robust diagnostic framework that can account for the confounding effects of patient-level characteristics observed in our retrospective analysis.

Conclusions

Conventional CT-derived quantitative imaging parameters—including lymph node short-axis diameter, contrast enhancement value, and anatomical nodal distribution—exhibited statistically significant differences between pulmonary sarcoidosis and MLNM in patients with lung cancer. The composite multiparametric logistic regression model achieved an AUC of 0.798, with high specificity (94.8%) but only moderate sensitivity (66.2%), indicating that approximately one-third of malignant cases would not be correctly identified by CT criteria alone. These findings confirm that CT-based imaging features cannot replace histopathological confirmation and that all patients with indeterminate mediastinal lymphadenopathy should undergo tissue sampling via EBUS-TBNA regardless of imaging appearances. The proposed model should be regarded strictly as a hypothesis-generating research tool. Prospective, multicenter validation studies are required before any CT-based diagnostic framework can be considered for clinical implementation.

Supplementary

The article’s supplementary files as

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. The study was approved by the Ethics Committee of Heilongjiang University of Chinese Medicine (Harbin, China). The requirement for written informed consent was waived by the Ethics Committee due to the retrospective nature of the study and the use of fully anonymized patient data.

Data Sharing Statement

Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1-2581/dss

qims-16-06-472-dss.pdf ^{(74KB, pdf)}

DOI: 10.21037/qims-2025-1-2581