Clinical-MRI-IVIM combined model for differentiating thymic epithelial tumors from anterior mediastinal lymphoma in large masses

Abstract

Objectives

To evaluate the diagnostic value of clinical and intravoxel incoherent motion (IVIM) parameters for differentiating large anterior mediastinal invasive masses (≥ 3 cm), particularly thymic epithelial tumors (TETs) and lymphomas.

Methods

This prospective study included 92 patients (mean age, 44.8 ± 15.5 years) with pathologically confirmed TETs (n = 52) or lymphomas (n = 40) who underwent routine MRI, diffusion-weighted imaging, and IVIM sequences. Quantitative parameters, including the apparent diffusion coefficient (ADC), true diffusion coefficient (D), pseudo-diffusion coefficient (D*), and perfusion fraction (f), were measured. Univariate and multivariate logistic regression analyses were used to construct predictive models. Diagnostic performance was evaluated using receiver operating characteristic (ROC) analysis, with area under the curve (AUC) compared by the DeLong test. Internal validation was performed using bootstrap resampling, and multicollinearity was assessed using variance inflation factors.

Results

Compared with TET, lymphoma was associated with younger age, higher female proportion, higher LDH, larger volume, more frequent pleural and/or lung invasion, and higher D * values (all p < 0.05). Age, sex, LDH level, lung and/or pleural infiltration, and f in the IVIM model were risk predictive factors for distinguishing lymphoma from TET. Among all models, the clinical MRI-IVM model achieved the highest diagnostic performance numerically (AUC = 0.946, 95% CI: 0.901–0.991); However, compared with the clinical MRI model (AUC = 0.929, 95% CI: 0.873–0.985), the increment brought by adding IVIM parameters was not significant (DeLong test, P = 0.316). Calibration analysis showed good consistency between predicted probabilities. The apparent AUC of Bootstrap’s internal validation was 0.946, the estimation deviation was 0.017, and the standard error was 0.022, indicating limited optimism in the model’s performance.

Conclusions

The clinical-MRI-IVIM model had shown relatively good diagnostic performance in preoperative differentiation of large anterior mediastinal TETs and lymphoma. However, there is no statistical difference in the increment benefit generated by IVIM parameters in the composite model. Therefore, in order to elucidate the clinical efficacy of IVIM, further large-scale, prospective, and multi-center studies are necessary.

Introduction

Anterior mediastinal invasive tumors include thymic epithelial tumors (TETs), lymphomas, germ cell tumors, and other rare malignancies. Among them, TETs and lymphoma are the most common. Their management differs substantially: TETs are usually treated surgically, while primary mediastinal lymphomas are treated with systemic therapy. Therefore, accurate differential diagnosis is crucial for optimizing treatment plans. The commonly used non-invasive diagnostic tools include computed tomography (CT), magnetic resonance imaging (MRI), and fluorine-18 fluorodeoxyglucose positron emission tomography/CT (18 F-FDG PET/CT) [1, 2]. However, the overlap of imaging appearances, especially for large masses (≥ 3 cm) with necrosis or cystic changes, often limits the specificity of these imaging methods. Nowadays, diagnosis still relies on invasive tissue sampling, such as CT guided biopsy. Therefore, this prompts us to explore advanced non-invasive identification methods and models that can provide benefits for clinical diagnosis.

IVIM analysis of multi b-value diffusion MRI separates microcirculatory perfusion effects from true molecular diffusion by fitting a biexponential model to the diffusion-weighted imaging (DWI) signal at multiple b-values. IVIM is more accurate in evaluating the complexity of microstructure in tissues. The quantification of its derived parameters has been proven to be very meaningful in diagnosis [3]. The commonly reported IVIM parameters are D (true/slow diffusion) related to cell density, D* (pseudo-diffusion associated with microcirculation) associated with microcirculation perfusion, and f (perfusion fraction) correlated with tumor angiogenesis and microvascular density [4–6]. Previous research on IVIM in tumor diagnosis and predictive classification had mainly focused on areas such as head and neck, breast, prostate, liver and lung, which has high value [7–12]. There is relatively little research on mediastinum, and the differentiation of invasive tumors in the anterior mediastinum is still a research gap [13, 14].

This prospective study focused on the differential diagnosis of large anterior mediastinal masses (≥ 3 cm), especially TETs and lymphoma, aimed to evaluate whether the biological significance of IVIM parameters provided incremental value in the differential diagnosis of large anterior mediastinal tumors, and to explore the diagnostic value of the combined model.

Materials and methods

This prospective study was approved by the ethics committee of our institution, and the Institutional Review Board number was NCC-012590.

Patients

We prospectively analyzed patients who reported anterior mediastinal masses from July 2017 to July 2024. The inclusion criteria were as follows: (1) MRI including routine sequences, DWI, and multi-b-value IVIM performed on the same scanner; (2) No prior treatment before MRI; (3) accurate pathological diagnosis of TETs or lymphoma obtained by surgical resection specimen or percutaneous biopsy. The exclusion criteria were as follows: (1) tumor < 3.0 cm on prior CT; (2) poor-quality MRI/DWI images; (3) unavailable or non-diagnostic pathological diagnosis (Fig. 1).

Fig. 1.

Flowchart of the study population. Numbers in parentheses are the numbers of patients. TETs, Thymic Epithelial Tumors; LDH, Lactic Dehydrogenase; DWI, Diffusion-Weighted Imaging; IVIM, Intravoxel Incoherent Motion; MRI, Magnetic Resonance Imaging

Finally, a total of 92 patients (46 females and 46 males) with anterior mediastinal mass confirmed by pathology were enrolled in this prospective study for data analysis. A total of 22 and 70 individuals obtaining accurate pathological results underwent surgical resection and CT guided percutaneous puncture, respectively.

Imaging protocol

All MRI scans were performed on a 3.0-T whole-body scanner equipped with a 32-channel coil (Discovery MR750; GE Healthcare, Milwaukee, WI, USA). The routine lung imaging protocol included the following sequences: axial propeller T2-weighted imaging with fat suppression (T2WI/FS), axial fast spin echo T1-weighted imaging, DWI, DWI with multiple b-values.

DWI and IVIM was performed using a single-shot echo-planar imaging (SS-EPI) sequence with axial bilateral orientation. Imaging parameters included a repetition time (TR) of 5900 ms and an echo time (TE) of 56 ms. The field of view (FOV) was 384 × 384 mm², with a matrix size of 128 × 128, yielding a reconstructed voxel size of 2.0 × 2.0 × 4.0 mm³. A total of 28 slices were acquired with a section thickness of 4 mm. Spectral adiabatic inversion recovery (SPAIR) was applied for fat suppression. Two signal averages (NSA = 2) were used. The receiver bandwidth was 250 kHz, and the total acquisition time was 54 s. DWI was applied using b-values of 0 and 800 s/mm². IVIM was applied using eight b-values: b = 0, 50, 100, 200, 400, 600, 800, and 1000 s/mm². Constrained by the total scan time to minimize patient motion, additional signal averaging was not pursued, although it could improve the signal-to-noise ratio (SNR) as well as robustness for eddy current correction. For each non-zero b-value, the diffusion gradient was applied successively along each of the three orthogonal directions, producing a trace-weighted image to minimize the influence of diffusion anisotropy [13].

Quantitative image analysis

All images were successfully acquired. Subsequently, quantitative were calculated on an offline workstation (GE Advantage Workstation AW4.6; GE Healthcare). Including, parametrical maps of apparent diffusion coefficient (ADC).

Quantitative parameters from the IVIM model were as follows:

D, slow diffusion coefficient (mm²/s), reflecting the slow diffusion motion of water molecules (tissue diffusion);

D*, fast diffusion coefficient (mm²/s), reflecting the rapid diffusion of microcirculatory perfusion in the capillary network;

f, the perfusion fraction (0 ≤ f ≤ 1) represents the volume ratio of microcirculatory perfusion contribution.

The mean ADC was determined on the basis of the assumption of a mono-exponential relationship between signal intensity S_b and b-value in the DWI model:

The IVIM model was fitted to generate parameter maps of D, D*, and f based on the multi-b diffusion images on a voxel-by-voxel basis with a Levenberg-Marquardt nonlinear fitting algorithm, using a dual exponential mode equation as follows [15]:

S was strength when applying a diffusion sensitive gradient (b-value), S0 represents strength at b = 0 (without diffusion gradient), b was the diffusion sensitivity factor (unit: s/mm ²), reflecting the sensitivity of gradient fields to diffusion motion.

Two radiologists specializing in chest radiology analyze images without knowing the pathological results. Before further parameter analysis, the reader evaluated the imaging quality. The poor image quality of DWI with multiple b-values was mainly due to excessive motion artifacts or significant signal loss [16]. After measuring the average axis diameter on T2WI/FS images, firstly, using DWI and ADC images as references, a single ROI was defined in the diffusion restricted solid part of the tumor, with a lesion coverage of at least 75%, avoiding cystic or necrotic components. Then the system automatically links other parameter maps and synchronously located the ROI to the ADC, D, D *, and f maps at the same level for parameter extraction. (Fig. 2).

Fig. 2.

ROI putting method. Using DWI (A) and ADC (B) images as references, place the region of interest in the diffusion restricted area (Fig. 1) to achieve at least 75% coverage (avoiding cystic necrosis). Next, locate the same region of interest in the corresponding quantitative map and record the average diffusion parameters between voxels within the ROI, including the slow diffusion coefficient (D) map (C), fast diffusion coefficient (D *) map (D), and perfusion score (f) map (E). DWI, Diffusion-Weighted Imaging; ADC, Apparent Diffusion Coefficient; ROI, Region of Interest

Interobserver agreement

To assess the reproducibility of quantitative IVIM measurements, two chest radiologists (with 4 and 10 years of experience in thoracic MRI, respectively) independently placed ROIs and extracted IVIM parameters (D, D*, and f). Interobserver agreement was evaluated using the intraclass correlation coefficient (ICC) with a two-way mixed-effects model and absolute agreement definition in SPSS (version 23.0, IBM, Chicago, IL, USA). An ICC value < 0.40 was considered poor, 0.40–0.59 fair, 0.60–0.74 good, and ≥ 0.75 excellent reliability. Since interobserver agreement was good to excellent (ICC ≥ 0.60 for all parameters), the mean values of the two radiologists’ measurements were used for further statistical analyses.

Statistical analyses

Statistical analyses were performed using R software (version 4.3.0; R Foundation for Statistical Computing, Vienna, Austria). The variables included in the analysis did not have missing data; therefore, a complete case analysis was conducted. The normality of continuous variables was evaluated using the Shapiro Wilk test, and homogeneity of variance was evaluated with the Levene’s test. Normally distributed continuous variables are presented as mean±standard deviation, whereas non-normally distributed variables are reported as median (interquartile range). Continuous variable comparison between groups was performed using analysis of t test or Kruskal Wallis test. Categorical variables are presented as counts and percentages and were compared using the chi-square test or Fisher’s exact test, as appropriate. Univariate logistic regression analysis was first performed to evaluate the association between each candidate clinical, conventional MRI, and IVIM parameters and tumor type. Variables with P < 0.20 in univariate logistic regression were entered into the multivariable logistic regression analysis, a threshold chosen to avoid excluding potentially relevant predictors at the screening stage. A forward stepwise selection procedure was then used to identify the final independent predictors. Multicollinearity among variables retained in the final model was assessed using variance inflation factors (VIFs). Receiver operating characteristic (ROC) curve analysis was performed to evaluate the diagnostic performance of individual IVIM parameters and prediction models, with the area under the curve (AUC) used as the primary metric. The AUCs of correlated models were compared using the DeLong test. Internal validation was conducted using nonparametric bootstrap resampling with 1,000 iterations, and model calibration was evaluated using calibration curves. Optimal cutoff values were determined based on the Youden index, and corresponding sensitivity, specificity, positive predictive value, and negative predictive value were calculated. A two-sided P value < 0.05 was considered statistically significant.

Results

Clinical features

Among the 92 included patients, there were 46 males and 46 females (44.8 ± 15.5 years). One lymphoma patient and one B2 thymoma patient both had lung cancer. Clinical symptoms were present in 41 (44.6%) patients, including cough (n = 19), chest-back pain (n = 19), chest tightness (n = 10), fever (n = 3), and blepharoptosis and fatigue sputum (n = 3). A total of 47 and 45 patients underwent plain and enhanced MRI examinations, respectively. Finally, 22 and 70 patients were confirmed by surgical pathology and CT-guided aspiration biopsy, respectively. There were 40 cases of lymphoma and 52 cases of TETs. TETs included 27 cases of thymoma, 23 cases of thymic cancer, and 2 cases of neuroendocrine tumors (Table 1).

Table 1.

Clinical features and diagnosis of patients in this study

Clinical features
Age (years)	44.8 ± 15.5
Sex
Male	46 (50.0%)
Female	46 (50.0%)
Symptom
Negative	51 (55.4%)
Coughing	19
Chest-back pain	19
Chest tightness	10
Fever	3
Blepharoptosis and fatigue	3
History of Malignancy	2
Laboratory Examination
LDH ↑	29 (31.5%)
Diagnosis method
Surgery	22 (23.9%)
CT-guided aspiration biopsy	70 (76.1%)
Pathologic diagnosis
Lymphoma	40 (43.5%)
TETs	52 (56.5%)
Thymoma	27
Thymic carcinoma	23
Neuroendocrine tumors	22

Values are expressed as number, mean ± standard deviation. TETs, Thymic Epithelial Tumors; LDH, Lactic Dehydrogenase; DWI, Diffusion-Weighted Imaging; ADC, Apparent Diffusion Coefficient; IVIM, Intravoxel Incoherent Motion; MRI, Magnetic Resonance Imaging

Interobserver agreement

Table 2 summarized interobserver agreement for ADC, D, D*, and f. ADC exhibited the best inter observer consistency, while D * showed relatively lower reproducibility than other parameters, although the consistency was still within an acceptable range.

Table 2.

Interobserver agreement of the parameters in this study

Parameter	ICC value	95% CI (lower–upper)	Agreement level
ADC (×10⁻³ mm²/s)	0.82	0.73, 0.88	Excellent
D (×10⁻³ mm²/s)	0.75	0.62, 0.83	Good
D* (×10⁻³ mm²/s)	0.73	0.58, 0.81	Good
f (perfusion fraction)	0.80	0.64, 0.84	Excellent

ADC, Apparent Diffusion Coefficient; ICC, Intra-Class Correlation Coefficient; CI, Confidence Interval

Clinical and routine MRI imaging characters

Lymphoma and TETs tumors with large volumes (≥ 3.0 cm) in the anterior mediastinum were difficult to distinguish in terms of routine MRI imaging characteristics, and both exhibited the following signal characteristics: low signal on T1WI, uneven medium-high signal on T2WI, large areas of low signal necrosis, high signal on DWI, and uneven medium enhancement on contrast-enhanced scans (Fig. 3).

Fig. 3.

MRI imaging features of anterior mediastinal TETs and lymphoma. Patient 1, a 39-year-old male, was diagnosed with type B3 thymoma through pathological biopsy. The tumor has a long diameter of 13 centimeters and a short diameter of 8 centimeters. T1WI low signal (A), T2WI uneven high signal, extensive low signal necrosis (B), DWI high signal (C), enhanced scan uneven moderate enhancement, necrosis not enhanced (D). Patient 2, a 40-year-old male, was diagnosed with B-cell lymphoma through pathological biopsy. The tumor has a long diameter of 7.8 cm and a short diameter of 6.2 cm, with low signal intensity on T1WI (E), uneven medium high signal intensity on T2WI (F), extensive low signal necrosis, high signal intensity on DWI (G), and uneven medium enhancement on contrast-enhanced scan (H). Compared with lymphoma, thymoma has a higher degree of cystic degeneration and necrosis, higher T2WI signal intensity, and no significant differences in other sequences. MRI, Magnetic Resonance Imaging; DWI, Diffusion-Weighted Imaging

Table 3 showed the clinical and imaging features of the tumors in this study. In clinical features, age, sex, and elevated LDH levels were statistically significant in both groups (all P < 0.001). MRI findings demonstrated that lymphomas had a larger mean diameter and more frequent involvement of large blood vessels, lung, and pleural (P < 0.001, P = 0.048, 0.003). The ADC parameters in the DWI model had no statistical significance. In the IVIM model, the D* parameter of lymphoma was higher than that of TETs, with borderline statistical significance (P = 0.048). This finding was not adjusted for multiple comparisons and should therefore be interpreted cautiously.

Table 3.

Univariate and multivariate analysis for distinguishing anterior mediastinal TETs from lymphoma in this study

Values	Lymphoma group (n = 40)	TETs group (n = 52)	P value	Univariate Analysis	VIFs	Multivariate Analysis
				P value		P value	Exp(B)	95%CI
Age (year)	35.7 ± 14.6	51.8 ± 12.3	< 0.001	< 0.001	1.389	0.002	1.106	1.038,1.177
Sex (male/female)	11(27.5%) /29(72.5%)	35(67.3%) /17(32.7%)	< 0.001	< 0.001	2.165	0.002	11.264	2.422,52.383
LDH (↑)	24 (60%)	4 (7.7%)	< 0.001	< 0.001	1.467	< 0.001	27.961	4.893,159.797
Average diameter (cm)	8.3 (6.7, 10.2)	6.7 (5.8, 8.1)	< 0.001	0.004	1.800
Invasion
Large blood vessels	26 (65%)	23 (44.2%)	0.048	0.050	1.445
Pericardium	32 (80%)	40 (76.9%)	0.723	0.723
Pleura and lungs	27 (67.5%)	19 (36.5%)	0.003	0.004	1.770	0.024	5.408	1.253,23.335
Pericardial and/or pleural effusion (+)	23 (57.5%)	24 (46.2%)	0.280	0.280
Mediastinal lymphadenopathy (+)	23 (57.5%)	20 (38.5%)	0.070	0.071	1.607
DWI model (mm²/s)
Standard ADC value (/103)	1.103 ± 0.637	1.137 ± 0.497	0.574	0.774
IVIM model (mm²/s)
D (/103)	0.835 ± 0.423	0.869 ± 0.386	0.558	0.692
D*	0.033 ± 0.059	0.016 ± 0.014	0.048	0.104	1.220
f	0.408 ± 0.223	0.472 ± 0.235	0.189	0.186	1.744	0.046	24.697	1.052, 579.946

Categorical variables are expressed as number; Normally distributed continuous variables are presented as mean±standard deviation, whereas non-normally distributed variables are reported as median (interquartile range); Bold font represents a statistically significant P-value (P < 0.05).TETs, Thymic Epithelial Tumors; VIFs, Variance Inflation Factors; LDH, Lactic Dehydrogenase; DWI, Diffusion-Weighted Imaging; ADC, Apparent Diffusion Coefficient; IVIM, Intravoxel Incoherent Motion; MRI, Magnetic Resonance Imaging; CI, Confidence Interval

Differential prediction model for anterior mediastinal lymphoma and TETs

In univariate logistic regression analysis, variables with P < 0.20 were selected as candidate variables for multivariate logistic regression (age, sex, LDH level, average diameter, mediastinal Large blood vessels and lung/pleural invasion, mediastinal lymphadenopathy, IVIM-D*, IVIM-f). Using forward stepwise multivariate logistic regression, age, sex, LDH levels, lung and/or pleural infiltration, and IVIM-f were retained as independent predictive factors for distinguishing lymphoma from TET. Although D* showed significant inter group differences, it was not retained in the final multivariate model after positive stepwise selection (Table 3). Assessment of multicollinearity revealed low to moderate correlations among predictors. All variance inflation factors ranged from 1.220 to 2.165, indicating the absence of significant multicollinearity in the final multivariable model. Together, these analyses support the robustness of the model and mitigate concerns regarding overfitting.

Table 4 summarized the diagnostic performance of different models. The use of conventional MRI alone could only moderately distinguish TETs from lymphoma (AUC = 0.726, 95% CI: 0.617, 0.835). Comparing with MRI model, the MRI-IVIM model showed no significant improvement in diagnostic performance (AUC = 0.727, Delong test Z= -0.070, P = 0.944); the Clinical-MRI model showed significantly higher performance was observed (AUC = 0.929, Delong test Z = -3.767, P < 0.001); When IVIM parameters were further integrated into this clinical-MRI model, the best overall AUC was achieved (AUC = 0.946, Delong test Z=-3.954, P < 0.001)(Fig. 4). In the combined model, IVIM parameters did not provide significant increment value for the Clinical-MRI model (Delong test: Z= -1.003, p-value = 0.316).

Table 4.

The diagnostic efficacy of various models distinguishing anterior mediastinal TET from lymphoma in this study

Model	AUC	95% CI	Sensitivity	Specificity
MRI	0.726	0.617, 0.835	55.0%	86.5%
MRI-IVIM	0.727	0.623, 0.832	47.5%	88.5%
Clinical-MRI	0.929	0.873, 0.985	87.5%	92.3%
Clinical-MRI-IVIM	0.946	0.901, 0.991	90.0%	92.3%

MRI, Magnetic Resonance Imaging; DWI, Diffusion-Weighted Imaging; IVIM, Intravoxel Incoherent Motion; AUC, Aera Under The Curve; CI, Confidence Interval

Fig. 4.

ROC curve of various models distinguishing anterior mediastinal TETs from lymphoma in this study. The ROC curves illustrated the diagnostic performance of four models for distinguishing TETs from lymphoma: the conventional MRI model (gray line), the combined MRI-IVIM model (blue line), the clinical-MRI model (green line), and the integrated clinical-MRI-IVIM model (red line). The corresponding AUC were: Conventional MRI, AUC = 0.726(0.617, 0.835); MRI-IVIM, AUC = 0.727(0.623, 0.832); Clinical-MRI, AUC = 0.929(0.873, 0.985); Clinical-MRI-IVIM, AUC = 0.946(0.901, 0.991). The Clinical-MRI-IVIM composite model showed the highest diagnostic efficacy, visually positioned toward the upper left corner of the graph. ROC, Receiver Operating Characteristic curve; TETs, Thymic Epithelial Tumors; IVIM, Intravoxel Incoherent Motion; MRI, Magnetic Resonance Imaging; AUC, Aera Under Curve

Table 5.

Univariate and multivariate analysis for distinguishing pathological classification of TETs in this study

	Thymoma group (n = 27)	Thymic carcinoma group (n = 25)	p value	Univariate Analysis	Multivariate Analysis
				P value	P value	Exp(B)	95%CI
Age (year)	49.1 ± 11.1	54.7 ± 13.0	0.103	0.099	0.031	0.939	0.887,0.994
Sex (male/female)	16/11	19/6	0.199	0.199
LDH (+)	2	2	0.936	0.936
Average diameter (cm)	7.2 (5.8, 8.3)	6.5 (5.8, 7.8)	0.486	0.531
Invasion
Large blood vessels	11	12	0.598	0.598
pericardium	17	23	0.013	0.023	0.008	0.088	0.014,0.529
Pleura and lungs	9	10	0.618	0.618
Pericardial and/or pleurale ffusion (+)	9	15	0.054	0.054
Mediastinal lymphadenopathy (+)	11	9	0.726	0.726
DWI model (mm²/s)
Standard ADC value (/103)	1.197 ± 0.556	1.073 ± 0.427	0.552	0.364
IVIM model (mm²/s)
D (/103)	0.897 ± 0.347	0.838 ± 0.430	0.416	0.581
D*	0.018 ± 0.015	0.015 ± 0.013	0.602	0.782
f	0.482 ± 0.267	0.461 ± 0.200	0.757	0.751

Categorical variables are expressed as number; Normally distributed continuous variables are presented as mean±standard deviation, whereas non-normally distributed variables are reported as median (interquartile range); Bold font represents a statistically significant P-value (P < 0.05).TETs, Thymic Epithelial Tumors; LDH, Lactic Dehydrogenase; DWI, Diffusion-Weighted Imaging; ADC, Apparent Diffusion Coefficient; IVIM, Intravoxel Incoherent Motion; MRI, Magnetic Resonance Imaging; CI, Confidence Interval

Internal validation using ordinary nonparametric bootstrap resampling (1,000 iterations) was performed for the clinical-MRI-IVIM model. The original AUC was 0.946, and the Bootstrap_mean_AUC was 0.962. Bootstrap analysis yielded a standard error of 0.022, with a small positive bias of 0.017, indicating stable model performance with limited optimism (Fig. 5a). Calibration analysis of Clinical-MRI-IVIM model showed good consistency between predicted probabilities and observed results (Fig. 5b).

Fig. 5.

Bootstrap validates the identification and calibration of clinical MRI IVIM prediction models. (a) The kernel density distribution of the AUC estimation valued obtained from 1000 guided resamples. This figure showed the sampling distribution of the model’s discriminative performance, displaying the raw AUC (0.946, red vertical line) and the average bootstrap AUC (0.962, black dashed line). The narrow approximate normal distribution confirmed that the predictive performance of the model was stable in resampling, with no evidence of significant bias or variance inflation. (b) The plot assesses the agreement between predicted probabilities and observed outcomes. The dashed line represents the apparent performance of the model on the original sample; the solid line denotes the bias-corrected calibration curve after 1,000 bootstrap repetitions; the grey-shaded area indicates the 95% C.L. The ideal perfect calibration is shown as a dashed diagonal line. Individual observations are displayed as short vertical ticks along the top of the plot. The model demonstrates good calibration, with the bias-corrected curve closely following the ideal line. AUC, Aera Under Curve; C.L., Confidence Limits

Differentiation of TET pathological classification

TETs in this study were divided into thymoma and thymic cancer (including neuroendocrine tumors) groups, with 27 and 25 cases, respectively. The clinical and imaging features were shown in Table 4. There was a statistically significant difference in pericardial invasion between the two groups (P = 0.013). Age and pericardial invasion were risk factors for differentiating subtypes. The AUC was 0.756 (0.624, 0.887), with sensitivity and specificity of 81.5% and 64.0%, respectively (Fig. 6). In the diagnosis of TETs pathological classification, all parameters of the IVIM model had no statistical significance. Due to the small sample size of subgroup analysis, the lack of significant IVIM differences should be explained as exploratory rather than deterministic.

Fig. 6.

ROC curve of TETs pathological classification. Age and pericardial invasion were risk factors for predicting subtypes. AUC was 0.756(0.624, 0.887), sensitivity and specificity was 81.5% and 64.0%, respectively. In the diagnosis of TETs pathological classification, all parameters of the IVIM model had no statistical significance. ROC, Receiver Operating Characteristic curve; TETs, Thymic Epithelial Tumors; IVIM, Intravoxel Incoherent Motion; AUC, Aera Under Curve

Discussion

Among the invasive anterior mediastinal tumors, TETs and lymphoma had the highest incidence rate, and it was difficult to differentiate between them before operation, especially the larger ones (d≥3 cm) [1]. In this prospective study, we evaluated the subtle differences in D* value from IVIM quantitative parameters between anterior mediastinal TETs and lymphoma, and constructed a combined model for the differential diagnosis of invasive large masses (≥ 3 cm) in the anterior mediastinum, while exploring the incremental value of IVIM parameters in the differentiation of anterior mediastinal masses.

In this study, there was a mild statistically significant difference in D * values between lymphoma and TETs (p = 0.048), while there was no statistically significant difference in D and f values. In this study, the majority of TETs were in the late stage (T2 ~ T4 stage, 82.7%, 43/52), with most necrotic areas observed without effective blood flow perfusion. Even if our ROI attempts to avoid necrotic areas, some cases may still be affected, leading to a decrease in D * values. Although there have been reports that TETs tumor cells secrete vascular endothelial growth factor, which promotes angiogenesis and is closely related to the increased invasiveness of TETs [17–19]. But an increase in blood vessels does not necessarily mean effective perfusion. The incompleteness of newly generated vascular walls can cause leakage and damage to blood flow, and an abnormally high tortuosity vascular network can increase geometric resistance and delay blood flow [20, 21]. Therefore, the relatively low D * values observed in TETs may partially reflect microvascular structural abnormalities and low perfusion efficiency, rather than just a lack of microvessels. However, our explanation is still speculative based on literature and has not undergone direct pathological and IVIM related mechanistic studies. Meanwhile, due to the lower ICC of D * compared to other diffusion parameters (ICC = 0.73), this result should be interpreted with caution. Previous studies have also shown that the stability of D * values was poor and susceptible to interference from scanning parameters and physiological movements [14, 22]. Given the borderline P-value and multiple comparisons conducted, this finding may also be prone to false positives, and should be considered an exploratory hypothesis rather than a deterministic biomarker. In addition, D * was not retained as an independent predictor in the final multivariate model. This exploratory finding should be evaluated using standardized and larger cohorts for repeat experiments.

We constructed a clinical-MRI-IVIM composite model (AUC = 0.946, 95% CI: 0.901–0.991) to improve the differential diagnostic efficacy between anterior mediastinal lymphoma and TETs. However, comparing the AUC of the clinical-MRI model, it was found that the difference caused by adding IVIM parameters to the multivariate model was not statistically significant. This indicated that although IVIM can provide supplementary physiological information, its additional clinical utility in this diagnostic context appears to be limited. This may be due to our limited number of enrolled patients. To determine the stability of the clinical-MRI-IVIM composite model, we conducted Bootstrap resampling (1000 iterations) and calibration analysis, demonstrating the internal stability of the current dataset. However, this does not confirm the universality of the external population. The risk of model overfitting cannot be completely ruled out. Because this is a single center prospective study without independent external validation, the observed diagnostic performance may be optimistic when applied to different institutions, scanners, patient populations, or imaging protocols. In summary, the current research results have not determined the significant incremental diagnostic benefits of IVIM beyond conventional clinical and MRI features. A larger and more heterogeneous population and external validation are needed to better determine the role of IVIM in evaluating anterior mediastinal masses.

In further exploratory TETs subgroup analysis, no IVIM parameters showed statistically significant differences between thymoma and thymic cancer/neuroendocrine tumors; However, due to the limited sample size of the subgroups, these findings should be interpreted with caution. Li et al. [14] found that the ADC, D, and D * values of the low-risk group were higher than those of the high-risk group or thymic cancer group, but there was no difference between the high-risk group and the thymic cancer group. Compared to this retrospective study, our study was specifically designed prospectively for large anterior mediastinal masses (≥ 3 cm) and directly evaluated whether IVIM provided incremental value beyond clinical and conventional MRI. In our study, only 6 cases were low-risk thymoma (type A, AB, B1), and we did not classify them based on risk level. The microstructural features (perfusion and diffusion) of high-risk thymoma and thymoma cancer are relatively consistent, making it difficult to capture subtle differences between them.

This study has some limitations. Firstly, the sample size of this prospective cohort is moderate (lymphoma, n = 40, TET, n = 52), particularly in relation to the number of candidate predictive factors evaluated. Based on these group sizes, a two-sided comparison withα = 0.05 and 80% power would require an approximate standardized effect size of about 0.60. Insufficient ability to detect mild to moderate differences in IVIM parameters. Especially for TETs subgroup analysis, the queue is relatively small and lacks motivation for analysis. Therefore, the analysis results should be interpreted as exploratory rather than deterministic. Secondly, this was a single-center prospective study without external validation. Therefore, although the model exhibited good internal performance, its clinical applicability in the real world is still uncertain. Thirdly, the stability of D* values was poor and susceptible to interference from scanning parameters and physiological movements, resulting in measurement errors and slightly lower consistency between observers. Two signal averages were used in IVIM acquisition without further increasing the signal average. This may affect the stability of signal-to-noise ratio and IVIM parameter estimation, especially for D *. Future research on optimizing acquisition strategies may help improve the repeatability of IVIM parameter measurements. Fourthly, the variable selection process was based on forward stepwise multivariate logistic regression after single factor screening. Although this method helps establish a simplified model, stepwise regression may increase the risk of model instability and overfitting in queues with limited sample sizes. Overall, our observation of mediastinal tumors based on MRI is still preliminary and exploratory, and it is necessary to conduct a multicenter prospective cohort study and external validation to evaluate the effectiveness of IVIM parameters in non-invasive preoperative prediction of pathological and histological types of anterior mediastinal large masses.

Conclusion

The difference in D * values of IVIM between lymphoma and TETs was borderline, and its reproducibility was limited, which restriced its applicability as a biomarker. The clinical-MRI-IVIM model had shown relatively good diagnostic performance in preoperative differentiation of large anterior mediastinal TETs and lymphoma. In our exploratory study, IVIM parameters generated limited incremental value in the composite model, but there was no statistical difference. Therefore, in order to elucidate the clinical efficacy of IVIM and reduce unnecessary biopsies while guiding appropriate treatment strategies, further large-scale, prospective, and multi-center external validations are necessary.

Abbreviations

ADC

Apparent diffusion coefficient

AUC

Area under the curve

CT

Computed tomography

D

True diffusion coefficient

D*

Pseudodiffusion coefficient

DWI

Diffusion-weighted imaging

f

Perfusion fraction

FOV

Field of view

18F-FDG PET/CT

Fluorine-18 fluorodeoxyglucose positron emission tomography/computed tomography

ICC

Intraclass correlation coefficient

IVIM

Intravoxel incoherent motion

LDH

Lactate dehydrogenase

MRI

Magnetic resonance imaging

ROC

Receiver operating characteristic

ROI

Region of interest

SNR

Signal-to-noise ratio

SPAIR

Spectral adiabatic inversion recovery

SS-EPI

Single-shot echo-planar imaging

TE

Echo time

TETs

Thymic epithelial tumors

TR

Repetition time

T1WI

T1-weighted imaging

T2WI/FS

T2-weighted imaging with fat suppression

VIF

Variance inflation factor

Author contributions

J. C. and W.T. conducted the investigation, performed data curation and formal analysis, developed the methodology, and wrote the original draft. L.Q., F.L., and S.C. were responsible for conceptualization, validation, visualization, and supervision. Y.H. provided resources and software, and contributed to writing (review & editing). J.W. acquired funding, administered the project, and contributed to writing (review & editing). All authors reviewed and approved the final manuscript.

Funding

This study has received funding by the Noncommunicable Chronic Diseases-National Science and Technology Major Project [grant number 2024ZD0520000; 2024ZD0520002], National Natural Science Foundation of China (grant 81971616), Beijing Natural Science Foundation (grant 7222148).

Data availability

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

Declarations

Ethics approval and consent to participate

This prospective study was approved by the ethics committee of our institution, and the Institutional Review Board number was NCC-012590. Informed consent was obtained from all individual participants included in the study.

Consent for publication

Written informed consent for publication was obtained from all participants.

Competing interests

The authors declare no competing interests.