Preoperative MRI Model for Predicting Inferior Vena Cava Wall Invasion in Renal Cell Carcinoma with Tumor Thrombus

Abstract

Purpose

To develop and evaluate a preoperative MRI-based model for predicting inferior vena cava (IVC) wall invasion in renal cell carcinoma (RCC) with IVC tumor thrombus (IVCTT) and to compare its performance with that of individual MRI features and radiologists’ subjective assessments.

Materials and Methods

This single-center study with retrospective and prospective components included individuals who underwent or were scheduled to undergo surgery for RCC with IVCTT (retrospective training set, n = 173, January 2005–December 2023; prospective temporal validation set, n = 44, January 2024–September 2025). Histopathology served as the reference standard. Quantitative (tumor, vessel, and thrombus measurements) and qualitative (signal and morphologic characteristics) MRI features were assessed. Two fellowship-trained abdominal radiologists independently provided subjective assessments of IVC wall invasion, and interobserver agreement was assessed. Variables significant at univariable analysis were entered into multivariable logistic regression to identify predictors of IVC wall invasion. Diagnostic performance was compared using receiver operating characteristic (ROC) curve analysis and DeLong tests.

Results

A total of 217 individuals were included (mean age, 57 years ± 12 [SD], 166 male). Four independent predictors of IVC wall invasion were identified: bland thrombus (odds ratio [OR] = 3.32 [95% CI: 1.38, 8.03]), lumbar vein diameter (>5.25 mm) (OR = 2.64 [95% CI: 1.23, 5.69]), ipsilateral renal vein ostium diameter (>19.20 mm) (OR = 3.64 [95% CI: 1.73, 7.63]), and thrombus craniocaudal length (>46.95 mm) (OR = 3.08 [95% CI: 1.43, 6.63]). The multivariable model incorporating these predictors achieved area under the ROC curve (AUC) values of 0.81 (95% CI: 0.75, 0.88) and 0.84 (95% CI: 0.73, 0.96) in the training and validation sets, respectively, significantly outperforming the best individual MRI predictor (AUC = 0.71) and radiologists’ subjective assessments (AUC = 0.66) (all P < .05).

Conclusion

The multiparametric MRI-based model demonstrated good discriminatory performance for predicting IVC wall invasion and outperformed individual MRI features and subjective radiologist assessment.

Keywords: MR Imaging, Urinary, Kidney, Renal Cell Carcinoma, Magnetic Resonance Imaging, Inferior Vena Cava Tumor Thrombus, Venous Wall Invasion

Supplemental material is available for this article.

© RSNA, 2026

See also commentary by Katwal and Lee in this issue.

Summary

A preoperative multiparametric MRI-based model demonstrated high performance for predicting inferior vena cava wall invasion in renal cell carcinoma with tumor thrombus, outperforming individual MRI features and radiologists’ subjective assessments.

Key Points

• ■ In a combined retrospective and prospective study of 217 individuals, four MRI features independently predicted inferior vena cava wall invasion: bland thrombus, larger lumbar vein and ipsilateral renal vein diameters, and longer tumor thrombus craniocaudal length (odds ratio = 3.32, 2.64, 3.64, and 3.08, respectively; all P < .05).

• ■ The multiparametric MRI model incorporating these four variables demonstrated stable discrimination in the training and temporal validation sets (area under the receiver operating characteristic curve, 0.81 and 0.84, respectively).

• ■ The model outperformed independent MRI predictors and radiologists’ subjective assessments (DeLong test: all P < .05).

Introduction

Renal cell carcinoma (RCC), one of the most common urinary system malignancies (1,2), represents a distinct and surgically challenging clinical entity when complicated by renal vein or inferior vena cava (IVC) tumor thrombus (IVCTT). IVCTT occurs in approximately 4%–15% of patients with newly diagnosed RCC and is associated with an adverse prognosis; in particular, IVC wall invasion has been linked to an approximately 56% increase in the risk of cancer-specific death (3,4). In patients without distant metastases, radical nephrectomy combined with thrombectomy remains the standard treatment (5,6). However, this procedure is complex and risky because of the requirement for precise vascular control, the risks of intraoperative embolism and hemodynamic instability, and the frequent need for a multidisciplinary team (7). Moreover, the surgical strategy depends fundamentally on whether the tumor thrombus has invaded the IVC wall. Accurate preoperative identification of IVC wall invasion is therefore critical, as confirmed invasion often necessitates partial or segmental caval resection or vascular reconstruction, substantially increasing procedural difficulty, perioperative morbidity, and mortality (8,9). In current clinical practice, IVC wall invasion is still frequently determined intraoperatively, thereby limiting preoperative risk stratification and surgical planning. A reliable preoperative imaging-based assessment of IVC wall invasion could facilitate individualized surgical strategies, appropriate multidisciplinary preparation, and improved patient counseling.

Among imaging modalities, US offers a restricted field of view and is highly operator dependent. Contrast-enhanced CT is widely used in the preoperative evaluation of RCC with IVCTT because of its excellent spatial resolution, robust multiplanar reconstructions, and reliable depiction of enhancement patterns (10,11). Nevertheless, the assessment of subtle IVC wall involvement remains challenging, particularly when the tumor thrombus is closely apposed to the caval wall without obvious contour disruption. MRI, with its comparatively high soft tissue contrast and multiparametric capabilities, offers additional information that may improve the evaluation of thrombus-wall relationships and vascular wall integrity. However, prior studies have reported variable diagnostic performance of MRI in helping detect IVC wall invasion, likely reflecting heterogeneous imaging criteria, subjective interpretations, and limited sample sizes (12–14).

To address these limitations, we systematically integrated multiple quantitative and qualitative MRI-derived features with the guidance of previous literature and clinical experience. The objective of our study was to develop and evaluate a preoperative MRI-based predictive model for IVC wall invasion in RCC with IVCTT and to compare its performance with that of individual MRI features and subjective assessment by radiologists.

Materials and Methods

Study Design and Sample

This single-center study included a retrospectively collected training set and a prospectively collected temporal validation set. The study was approved by the institutional review board of PLA General Hospital (approval no. S2024-034-01). The requirement for informed consent was waived for the retrospective dataset, whereas written informed consent was obtained from all prospectively enrolled individuals.

Patients who underwent radical nephrectomy with IVC thrombectomy between January 2005 and December 2023 were identified from the institutional database. The inclusion criteria were (a) age 18–85 years, (b) postoperative pathologic confirmation of RCC, (c) complete multiparametric MRI of the kidneys performed within 2 weeks before surgery, and (d) complete clinical, laboratory, and pathologic data. The exclusion criteria were (a) preoperative systemic therapy (targeted, immunotherapy, chemotherapy, or radiation therapy), (b) nondiagnostic MRI due to artifacts or missing sequences, (c) a tumor thrombus confined to the renal vein without extension into the IVC, as such individuals require a fundamentally different surgical strategy than those with IVC involvement, and (d) previous ipsilateral nephrectomy or another primary malignancy, to ensure anatomic homogeneity of the study sample focused on de novo RCC with IVC thrombus.

Among the 721 individuals who underwent surgery screening, 173 met all the eligibility criteria and were consecutively enrolled as the retrospective training set. According to the histopathologic findings, the patients were categorized into IVC wall invasion and non–IVC wall invasion groups. An additional 44 participants who were consecutively enrolled between January 2024 and September 2025 served as a temporal validation set. All cases in both sets were pathologically confirmed. Figure 1 shows the inclusion and exclusion flowcharts for both datasets.

Figure 1:

Flowcharts of study sample inclusion and exclusion for the (A) training set (retrospective) and (B) validation set (prospective). Note: The exclusion criterion “No IVC tumor thrombus” (n = 201) refers to cases where the tumor thrombus was confined to the renal vein and did not extend into the IVC. IVC = inferior vena cava, RCC = renal cell carcinoma.

MRI Acquisition

MRI examinations were performed using a 3.0-T GE Discovery MR 750 or a 1.5-T GE Signa HDxT scanner (GE HealthCare) with an eight-channel phased-array body coil. The 1.5-T scanner was primarily used for examinations conducted between 2005 and 2019, after which the 3.0-T system was used. In the retrospective training set (n = 173), 113 examinations (65.32%) were performed with the 1.5-T scanner, and 60 (34.68%) were performed with the 3.0-T scanner. All the participants in the prospective validation set (n = 44) were scanned with the 3.0-T system. The scanning range extended from the right atrium to the lower pole of the right kidney. All patients fasted for at least 4 hours before examination.

The scanning sequences and parameters are summarized in Table S1. A precontrast scan was performed before enhanced imaging. The contrast agent gadobenate dimeglumine (MultiHance; Shanghai Bracco Sine Pharmaceutical) was injected using a power injector at a dose of 0.1 mmol per kilogram of body weight and a flow rate of 1.5 mL/sec, followed by a 20-mL saline flush at the same rate. Axial images for the corticomedullary, nephrographic, and delayed phases were acquired at 25–30 seconds, 60–70 seconds, and 240 seconds after contrast agent injection, respectively.

Image Analysis

In this study, as is standard in the field, the term tumor thrombus specifically refers to the neoplastic venous extension of RCC and is distinct from a bland thrombus. Preoperative Digital Imaging and Communications in Medicine (ie, DICOM)–format MR images were imported into RadiAnt DICOM Viewer software (version 2020.2; Medixant). Two fellowship-trained abdominal radiologists (W.X.W. and S.P.Z., with 2 and 5 years of postfellowship experience, respectively, each performing more than 3000 renal MRI interpretations annually) independently reviewed all the imaging studies while blinded to patients’ clinical or pathologic information, to minimize bias. This blinding procedure was consistently applied to both the training set and the independent validation set.

Quantitative parameters

The radiologist assessments were performed on contrast-enhanced axial, coronal, and sagittal sequences according to a standardized measurement protocol to ensure reproducibility. The following parameters were measured: (a) Maximum primary tumor diameter, defined as the largest dimension identified in any of the three orthogonal planes. (b) Tumor thrombus craniocaudal length, measured on the coronal contrast-enhanced image that displayed the longest continuous segment of the thrombus. The proximal extent was defined as the most cephalad point within the IVC lumen, and the distal extent was defined as the most caudad point of the tumor thrombus. (c) Maximum short-axis diameter of the thrombus, measured on the axial image where the thrombus appeared largest perpendicular to its long axis. (d) Renal vein ostium diameter, measured on coronal contrast-enhanced images at the ostium where the renal vein joins the IVC, as the vertical distance between the inner walls. (e) Anteroposterior (AP) diameter of the IVC at the renal vein ostium level, measured on axial images. (f) Maximum AP diameter of the IVC, recorded at the widest point along its course on axial images. (g) Maximum lumbar vein diameter, measured on venous phase or delayed phase sagittal reconstructions as the maximum short-axis internal diameter at the most dilated segment of the lumbar vein anterior to the L1–L5 vertebral bodies.

Qualitative parameters

The imaging features evaluated were as follows: (a) Complete IVC occlusion, defined as the absence of intraluminal contrast enhancement. (b) Associated bland thrombus, an intraluminal filling defect that is contiguous with the enhancing tumor thrombus. Its defining characteristic is the absence of visible enhancement in all postcontrast phases, especially during the portal venous and delayed phases. This diagnosis is further supported by other imaging features, such as relatively homogeneous signal intensity on T2-weighted images and the absence of restricted diffusion at diffusion-weighted imaging (15). (c) Thrombus extravasation beyond the vascular contour, representing tumor extension outside the expected normal vessel boundary. (d) Vessel wall abnormalities, including wall irregularity, uneven thickening or thinning, focal discontinuity of the wall contour, or abnormal focal high signal on T2-weighted images and arterial phase or delayed phase contrast-enhanced images that suggest tumor infiltration. (e) Thrombus contour regularity, classified as regular or irregular. (f) Retrograde venous flow thrombus: refers to a tumor thrombus growing in the direction opposite to venous return, such as in the contralateral renal vein, lumbar veins, and gonadal veins. (g) Mayo classification stage, assigned according to established criteria (16).

The assessment criteria are illustrated in Figure 2. The average measurements from the two observers were used for quantitative parameters. Disagreements in qualitative assessments were resolved through consensus.

Figure 2:

Schematic illustrations of MRI parameter measurements and qualitative signs. (A) Quantitative parameter measurements. 1. Craniocaudal length of the tumor thrombus on a coronal image. 2. Ipsilateral renal vein ostium diameter on a coronal image. 3. Contralateral renal vein ostium diameter on a coronal image. 4. Lumbar vein diameter on a sagittal image. 5. Anteroposterior diameter of the inferior vena cava at the renal vein ostium level on an axial image. 6. Maximum anteroposterior diameter of the inferior vena cava on an axial image. 7. Maximum short-axis diameter of the tumor thrombus (green line) on an axial image, defined as the maximum of its anteroposterior and transverse diameters. 8. Maximum transverse diameter of the tumor thrombus on an axial image. (B) Qualitative MRI signs assessed in the study. IVC = inferior vena cava.

Clinical and Pathologic Data

Clinical data, including demographics (age, sex, and body mass index), clinical presentation, laboratory results, and pathologic findings (tumor subtype and TNM stage [17] and histologic confirmation of IVC wall invasion) were collected from electronic medical records. The pathologic diagnosis of wall invasion was based on microscopic confirmation of tumor cell infiltration into the vascular wall in those who underwent venous wall resection (Fig 3). For all cases, including those in the validation set, the pathologist performing this assessment was blinded to the imaging findings and the radiologists’ evaluations. For those who underwent complete thrombectomy without venous wall resection, the determination of noninvasion was based on the intraoperative surgical assessment of a completely nonadherent thrombus-vein interface, which represents the standard clinical criterion in such cases.

Figure 3:

Representative imaging, gross specimen, and histopathologic findings (from left to right) in a 63-year-old male patient with renal cell carcinoma with an inferior vena cava (IVC) tumor thrombus. (A) Preoperative contrast-enhanced MRI: Panels 1, 2, and 4 are coronal images, and panel 3 is a sagittal image. They show the craniocaudal diameter of the tumor thrombus (67.5 mm), the associated thrombus, a dilated lumbar vein (8.6 mm), and the diameter of the ipsilateral renal vein orifice (28.0 mm), respectively. (B) Gross specimen after radical nephrectomy: 1 shows the resected right kidney and the transected segment of the IVC containing the tumor thrombus; 2 presents a cross-sectional view of the tumor thrombus within the IVC. (C) Histopathologic micrograph: The red box highlights tumor cells infiltrating the muscular layer of the venous wall, indicating tumor thrombus invasion of the venous wall. (Hematoxylin-eosin stain; original magnification, ×100.) PT = primary tumor, TT = tumor thrombus.

Statistical Analysis

All statistical analyses were performed using SPSS software (version 26.0; IBM). Continuous variables are expressed as means ± SDs or medians with IQRs in parentheses and were compared using the t test or the Mann–Whitney U test, respectively; categorical variables were compared using the χ² test or the Fisher exact test. Interobserver reproducibility for quantitative variables was evaluated using the intraclass correlation coefficient, and interobserver agreement for qualitative variables was assessed using the Cohen κ statistic.

Receiver operating characteristic (ROC) curve analysis was used to evaluate the diagnostic performance of individual MRI parameters; cutoff values were determined by maximizing the Youden index. Variables with P < .05 in univariable analysis were entered into the multivariable logistic regression (forward likelihood ratio method). Odds ratios (ORs) and their 95% CIs were calculated from the multivariable model. For categorical predictors, the reference categories were defined as follows: absence of bland thrombus, smaller vessel diameters (dichotomized at the Youden-derived cutoff), and shorter thrombus craniocaudal length (dichotomized at the Youden-derived cutoff). Variance inflation factors (VIFs) were calculated to assess multicollinearity, with a VIF < 3 indicating acceptable collinearity. The final multivariable logistic regression model was constructed using the significant predictors identified, with model performance quantified by the area under the ROC curve (AUC) and compared using the DeLong test. Statistical significance was defined as P < .05.

Results

Clinical Characteristics of the Study Sample

For the retrospective training set, 721 patients who underwent surgery were initially screened. After applying the inclusion and exclusion criteria, 548 were excluded, resulting in a final retrospective training set of 173 patients (Fig 1A). For the prospective temporal validation set, an additional 44 participants were consecutively enrolled between January 2024 and September 2025 (Fig 1B). The main reasons for exclusion in the retrospective cohort included: absence of preoperative MRI within 2 weeks before surgery (n = 211), poor MRI quality (n = 23), tumor thrombus confined to the renal vein without IVC extension (n = 201), non-RCC pathology (n = 25), and receipt of preoperative antitumor therapy (n = 88). In the prospective cohort, exclusions were due to poor MRI quality (n = 4), tumor thrombus confined to the renal vein without IVC extension (n = 18), non-RCC pathology (n = 9), and preoperative antitumor therapy (n = 23).

This study included 217 individuals with RCC and IVCTT, comprising 166 male (76.5%) and 51 female individuals (23.5%), with a mean age of 57 years ± 12. Individuals were categorized into an invasion group (n = 98) and a noninvasion group (n = 119) according to the presence or absence of IVC wall invasion. There was no evidence of a difference in the baseline characteristics between the training set and the validation set (P > .05). In the training set, intergroup differences were observed in symptoms of flank/abdominal pain, pathologic histologic type (clear cell RCC vs non–clear cell RCC), Mayo classification, and T stage (all P < .05). In the validation set, intergroup differences were noted in age, positive urinary white blood cells, Mayo classification, and T stage (all P < .05). For the other clinicopathologic characteristics, there was no evidence of a difference between the two groups (Table 1).

Table 1:

Comparison of Clinical Characteristics between IVC Wall Invasion and Noninvasion Groups of Patients with RCC with IVCTT

Variable	Training Set (n = 173)				Validation Set (n = 44)
	IVC Wall Invasion Group (n = 73)	No–IVC Wall Invasion Group (n = 100)	Test Statistic	P Value	IVC Wall Invasion Group (n = 25)	No–IVC Wall Invasion Group (n = 19)	Test Statistic	P Value
Age (y)	53 ± 12	57 ± 12	−1.85*	.07	63 ± 12	56 ± 11	−2.25*	.03
Sex			0.38†	.54			0.02†	.90
Male	54 (74.0)	78 (78.0)			20 (80.0)	14 (73.7)
Female	19 (26.0)	22 (22.0)			5 (20.0)	5 (26.3)
BMI	24.20 (22.10, 26.86)	25.20 (22.73, 27.45)	−1.16*	.25	24.46 (23.67, 25.26)	24.18 (21.28, 27.08)	−0.22*	.83
Laterality			1.89†	.17			0.11†	.74
Right	51 (69.9)	79 (79.0)			17 (68.0)	12 (63.2)
Left	22 (30.1)	21 (21.0)			8 (32.0)	7 (36.8)
Hematuria	27 (37.0)	41 (41.0)	0.29†	.59	7 (28.0)	9 (47.4)	1.75†	.19
Flank/abdominal pain	33 (45.2)	22 (22.0)	10.48†	<.001	7 (28.0)	6 (31.6)	0.07†	.80
Diabetes mellitus	13 (17.8)	24 (24.0)	0.96†	.33	3 (12.0)	4 (21.1)	0.16†	.69
Hypertension	29 (39.7)	42 (42.0)	0.09†	.76	16 (64.0)	11 (57.9)	0.17†	.68
Smoking history	35 (47.9)	45 (45.0)	0.15†	.70	10 (40.0)	7 (36.8)	0.05†	.83
Alcohol use history	42 (57.5)	43 (43.0)	3.57†	.06	9 (36.0)	8 (42.1)	0.17†	.68
Positive urinary RBCs	34 (46.6)	45 (45.0)	0.04†	.84	6 (24.0)	7 (36.8)	0.86†	.36
Positive urinary WBCs	13 (17.8)	21 (21.0)	0.27†	.60	1 (4.0)	7 (36.8)	5.78†	.02
Positive urinary protein	27 (37.0)	37 (37.0)	0.00†	>.99	5 (20.0)	3 (15.8)	0.00†	>.99
Histologic subtype			7.71†	.005			0.02†	.90
Clear cell RCC	53 (72.6)	89 (89.0)			20 (80.0)	14 (73.7)
Non–clear cell RCC	20 (27.4)	11 (11.0)			5 (20.0)	5 (26.3)
Mayo stage			…‡	.003			…‡	.007
I	9 (12.3)	35 (35.0)			3 (12.0)	10 (52.6)
II	54 (74.0)	57 (57.0)			21 (84.0)	9 (47.4)
III	5 (6.8)	2 (2.0)			0 (0.0)	0 (0.0)
IV	5 (6.8)	6 (6.0)			1 (4.0)	0 (0.0)
T stage			…‡	<.001			…‡	<.001
T3b	0 (0.0)	93 (93.0)			0 (0.0)	11 (57.9)
T3c	70 (95.9)	1 (1.0)			24 (96.0)	6 (31.6)
T4	3 (4.1)	6 (6.0)			1 (4.0)	2 (10.5)
N stage			0.17†	.69			0.02†	.89
N0	65 (89.0)	87 (87.0)			25 (100.0)	18 (94.7)
N1	8 (11.0)	13 (13.0)			0 (0.0)	1 (5.3)
M stage			1.08†	.30			0.02†	.89
M0	61 (83.6)	89 (89.0)			25 (100.0)	18 (94.7)
M1	12 (16.4)	11 (11.0)			0 (0.0)	1 (5.3)

Note.—Continuous data are presented as means ± SDs or medians, with IQRs in parentheses, and categorical data as numbers, with percentages in parentheses. BMI = body mass index (calculated as weight in kilograms divided by height in meters squared), IVC = inferior vena cava, IVCTT = IVC tumor thrombus, RBC = red blood cell, RCC = renal cell carcinoma, WBC = white blood cell.

^*t value (normally distributed, continuous) or z score (non-normally distributed, continuous).

^† χ² value

^‡ Fisher exact test (no statistic reported).

Comparison of MRI Parameters and Diagnostic Efficacy for IVC Wall Invasion

Interobserver agreement for the assessed MRI parameters was good. The intraclass correlation coefficient values for the quantitative parameters ranged from 0.76 to 0.98 (P < .05), and the Cohen κ values for the categorical variables ranged from 0.69 to 0.93 (P < .05). The interobserver agreement for the overall subjective assessment of IVC wall invasion was moderate (Cohen κ = 0.62 [95% CI: 0.44, 0.80]). The AUC of the observers’ overall subjective assessment for predicting IVC wall invasion was 0.66 (95% CI: 0.58, 0.75), with a sensitivity, specificity, and accuracy of 63.0% (95% CI: 0.52, 0.74), 68.0% (95% CI: 0.59, 0.77), and 65.9% (95% CI: 0.58, 0.73), respectively. The quantitative parameters that were different between the invasion and noninvasion groups included the lumbar vein diameter, renal vein ostium diameter, AP diameter of the IVC at the renal vein ostium level, maximum IVC diameter, tumor thrombus craniocaudal length, and maximum short-axis diameter (all P < .05). Among the categorical variables, the presence of an associated bland thrombus, complete IVC occlusion, tumor thrombus extension beyond the vessel wall, vessel wall signal discontinuity, vessel wall irregularity, vessel wall abnormal thickening or thinning, abnormal arterial phase enhancement, and retrograde venous flow thrombus showed significant differences between groups (all P < .05, Table 2). For the maximum diameter of the primary renal tumor, tumor thrombus morphologic regularity, vessel wall signal abnormality on T2-weighted images, and delayed phase enhancement, there was no evidence of a difference between groups (Table 2). ROC curve analysis of individual parameters (Fig 4) indicated the limited efficacy of single MRI parameters in predicting IVC wall invasion. Among the quantitative parameters, the maximum IVC diameter had the highest diagnostic performance (AUC = 0.71 [95% CI: 0.63, 0.79]). Among the categorical variables, the presence of an associated bland thrombus had the best diagnostic value (AUC = 0.64 [95% CI: 0.56, 0.73]). The specific diagnostic performance of each parameter is detailed in Table 3.

Table 2:

Comparison of MRI Parameters between IVC Wall Invasion Group and Noninvasion Groups of Patients with RCC with IVCTT

Parameter	IVC Wall Invasion Group (n = 73)	No–IVC Wall Invasion Group (n = 100)	Statistical Value	P Value
Quantitative parameters
Maximum diameter of primary lesion (mm)	88.76 ± 27.52	83.40 ± 32.63	1.14*	.26
Lumbar vein diameter (mm)	6.43 ± 2.01	5.58 ± 1.93	2.81*	.006
Contralateral renal vein ostium diameter (mm)	12.76 ± 3.46	11.75 ± 3.15	2.00*	.047
Ipsilateral renal vein ostium diameter (mm)	20.90 (16.90, 24.00)	18.15 (15.40, 22.28)	−2.85†	.004
IVC anteroposterior diameter at renal vein ostium level (mm)	26.60 (21.75, 31.15)	20.35 (14.83, 28.30)	−3.77†	<.001
Maximum IVC anteroposterior diameter (mm)	37.30 (29.25, 41.20)	28.80 (23.18, 36.10)	−4.69†	<.001
Thrombus craniocaudal length (mm)	64.60 (44.50, 92.85)	39.25 (19.13, 64.78)	−4.34†	<.001
Maximum thrombus short-axis diameter (mm)	33.60 (25.50, 39.90)	23.60 (16.63, 32.78)	−4.34†	<.001
Qualitative parameters
Irregular shape of the tumor thrombus	30 (41.1)	39 (39.0)	0.08‡	.78
Associated bland thrombus	30 (41.1)	13 (13.0)	17.83‡	<.001
Complete IVC occlusion	37 (50.7)	26 (26.0)	11.11‡	.001
Thrombus extravasation beyond vessel contour	25 (34.2)	14 (14.0)	9.91‡	.002
Vessel wall signal discontinuity	37 (50.7)	29 (29.0)	8.41‡	.004
Vessel wall irregularity	45 (61.6)	37 (37.0)	10.28‡	.001
Vessel wall abnormal thickening or thinning	49 (67.1)	43 (43.0)	9.86‡	.002
T2-weighted imaging vessel wall signal abnormality	23 (31.5)	28 (28.0)	0.25‡	.62
Abnormal arterial phase enhancement of vessel wall	27 (37.0)	22 (22.2)	4.50‡	.03
Abnormal enhancement of the vascular wall in delayed phase	7 (9.6)	3 (3.0)	…§	.10
Retrograde venous flow thrombus	17 (23.3)	4 (4.0)	14.72‡	<.001

Note.—Continuous data are presented as means ± SDs, or medians, with IQRs in parentheses, and categorical data as numbers, with percentages in parentheses. IVC = inferior vena cava, IVCTT = IVC tumor thrombus, RCC = renal cell carcinoma.

^*t value

^† z score

^‡ χ² value

^§ Fisher exact test (no statistic reported).

Figure 4:

Receiver operating characteristic (ROC) curves for predicting inferior vena cava wall invasion. (A) Comparison of the ROC curves for overall subjective assessment (AUC = 0.66 [95% CI: 0.58, 0.75]); individual predictors, including associated bland thrombus (AUC = 0.64 [95% CI: 0.56, 0.73]), lumbar vein diameter (AUC = 0.63 [95% CI: 0.54, 0.71]), ipsilateral renal vein ostium diameter (AUC = 0.63 [95% CI: 0.54, 0.71]), and tumor thrombus craniocaudal length (AUC = 0.69 [95% CI: 0.61, 0.77]); and the combined multivariable prediction model (AUC = 0.81 [95% CI: 0.75, 0.88]). For clarity, only the four independent predictors from the final multivariable model are shown; the AUC values of all other individual MRI parameters are provided in Table 3. All AUCs for quantitative parameters were calculated using continuous measurements, with dichotomized cutoffs used only for model building. (B) ROC curves for the combined multivariable prediction model on the training set (AUC = 0.81 [95% CI: 0.75, 0.88]) and the prospective validation set (AUC = 0.84 [95% CI: 0.73, 0.96]). AUC = area under the ROC curve.

Table 3:

Diagnostic Performance of MRI Parameters for Detecting IVC Wall Invasion in Patients with RCC with IVCTT

Parameter	AUC	Cutoff Value (mm)	Sensitivity (%)	Specificity (%)	P Value
Quantitative parameters
Lumbar vein diameter	0.63 [0.54, 0.71]	5.25	75.3 (55/73) [65.5, 85.2]	49.0 (49/100) [39.2, 58.8]	.006
Contralateral renal vein ostium diameter	0.60 [0.52, 0.69]	13.95	43.8 (32/73) [32.5, 55.2]	79.0 (79/100) [71.0, 87.0]	.047
Ipsilateral renal vein ostium diameter	0.63 [0.54, 0.71]	19.20	69.9 (51/73) [59.3, 80.4]	62.0 (62/100) [52.5, 71.5]	.004
IVC anteroposterior diameter at renal vein ostium level	0.67 [0.59, 0.75]	20.65	80.8 (59/73) [71.8, 89.9]	52.0 (52/100) [42.2, 61.8]	<.001
Maximum IVC anteroposterior diameter	0.71 [0.63, 0.79]	32.35	72.6 (53/73) [62.4, 82.8]	65.0 (65/100) [55.7, 74.3]	<.001
Thrombus craniocaudal length	0.69 [0.61, 0.77]	46.95	74.0 (54/73) [63.9, 84.0]	66.0 (66/100) [56.7, 75.3]	<.001
Maximum thrombus short-axis diameter	0.70 [0.62, 0.78]	26.50	74.0 (54/73) [63.9, 84.0]	61.0 (61/100) [51.4, 70.6]	<.001
Qualitative parameters
Associated bland thrombus	0.64 [0.56, 0.73]	NA	41.1 (30/73) [29.8, 52.4]	87.0 (87/100) [80.4, 93.6]	<.001
Complete IVC occlusion	0.62 [0.54, 0.71]	NA	50.7 (37/73) [39.2, 62.2]	74.0 (74/100) [65.4, 82.6]	.001
Thrombus extravasation beyond vessel contour	0.59 [0.51, 0.68]	NA	34.2 (25/73) [23.4, 45.1]	86.0 (86/100) [79.2, 92.8]	.001
Vessel wall signal discontinuity	0.61 [0.52, 0.69]	NA	50.7 (37/73) [39.2, 62.2]	71.0 (71/100) [62.1, 79.9]	.004
Vessel wall irregularity	0.62 [0.54, 0.71]	NA	61.6 (45/73) [50.5, 72.8]	63.0 (63/100) [53.5, 72.5]	.001
Vessel wall abnormal thickening or thinning	0.62 [0.54, 0.71]	NA	67.1 (49/73) [56.3, 77.9]	58.6 (58/100) [47.3, 66.7]	.002
Abnormal arterial phase enhancement of vessel wall	0.57 [0.48, 0.66]	NA	37.0 (27/73) [25.9, 48.1]	78. (78/100) [68.8, 85.2]	.034
Retrograde venous flow thrombus	0.60 [0.51, 0.68]	NA	23.3 (17/73) [13.6, 33.0]	96.0 (96/100) [92.2, 99.8]	<.001

Note.—Values in parentheses are numerators and denominators; values in brackets are 95% CIs. AUC = area under the receiver operating characteristic curve, IVC = inferior vena cava, IVCTT = IVC tumor thrombus, NA = not applicable, RCC = renal cell carcinoma.

Logistic Regression Analysis

For clinical applicability, relevant continuous variables were dichotomized based on the optimal cutoff values derived from the ROC curve analysis for subsequent modeling. A total of 15 MRI parameters that showed significant intergroup differences in univariable analysis (Table 2) were entered as candidate variables into a multivariable logistic regression analysis using the forward likelihood ratio selection method. To avoid overfitting given the 73 positive events (approximately seven variables allowed by the “events per variable” rule), this stepwise approach was employed. Multicollinearity was assessed using the VIF, with VIF ≥ 3 indicating high collinearity. Among all 15 candidate variables (including seven quantitative and eight categorical parameters as detailed in the Results section), maximum IVC diameter, AP diameter of the IVC at the renal vein ostium level, and maximum short-axis diameter were excluded due to multicollinearity (VIF ≥ 3). Of the remaining 12 variables, four were ultimately retained as independent predictors: associated bland thrombus, ipsilateral renal vein ostium diameter, lumbar vein diameter, and tumor thrombus craniocaudal length (Table 4). The other eight variables (namely, contralateral renal vein ostium diameter, complete IVC occlusion, tumor thrombus extension beyond the vessel wall, vessel wall signal discontinuity, vessel wall irregularity, vessel wall abnormal thickening or thinning, abnormal arterial phase enhancement, and retrograde venous flow thrombus) were not retained in the stepwise regression model (P ≥ .05). Four independent factors were ultimately identified to predict IVC wall invasion: the presence of an associated bland thrombus, ipsilateral renal vein ostium diameter, lumbar vein diameter, and tumor thrombus craniocaudal length (Table 4). Based on these factors, we constructed the final prediction model. The prediction formula for estimating the probability P of IVC wall invasion is as follows: P = e^[logit(P)]/{1 + e^[logit(P)]}, where logit(P) = −1.99 + (1.20 × X₁) + (0.97 × X₂) + (1.29 × X₃) + (1.12 × X₄). The variables are defined as follows: X₁ = associated bland thrombus (1 = present, 0 = absent), X₂ = lumbar vein dilatation (>5.25 mm, 1; otherwise, 0), X₃ = ipsilateral renal vein ostium dilatation (>19.20 mm, 1; otherwise, 0), and X₄ = thrombus craniocaudal length (>46.95 mm, 1; otherwise, 0). For research exploration, a free web-based calculator implementing this model is available at https://ivc-research-group.github.io/ivc-calculator/; it is not intended for clinical decision-making at this stage. The predictive model incorporating these four factors demonstrated favorable discriminative ability in the training set, with an AUC of 0.81 (95% CI: 0.75, 0.88) (Fig 4A). The model also exhibited excellent generalizability in the validation set, achieving an AUC of 0.84 (95% CI: 0.73, 0.96) (Fig 4B). DeLong test results confirmed that the diagnostic performance of this combined predictive model was greater than that of each individual predictor and the overall subjective assessment (all P < .001).

Table 4:

Multivariable Logistic Regression Analysis of MRI Parameters for Diagnosing IVC Wall Invasion

Parameter	Partial Regression Coefficient	Standard Error	Wald χ2 Value	OR	P Value
Constant	−1.99	0.45	19.72	0.14	<.001
Associated bland thrombus	1.20	0.45	7.12	3.32 (1.38, 8.03)	.008
Lumbar vein diameter	0.97	0.39	6.15	2.64 (1.23, 5.69)	.01
Ipsilateral renal vein ostium diameter	1.29	0.38	11.62	3.64 (1.73, 7.63)	.001
Thrombus craniocaudal length	1.12	0.39	8.20	3.08 (1.43, 6.63)	.004

Note.—The training set consisted of 173 patients, 73 of whom experienced IVC wall invasion. Values in parentheses are 95% CIs. IVC = inferior vena cava, OR = odds ratio.

Discussion

In our study, we developed and validated a multiparametric MRI-based model for identifying IVC wall invasion in patients with RCC and IVC tumor thrombus. The diagnostic performance of the model was greater than that of individual MRI parameters and radiologists’ subjective assessments, providing a potential tool to improve preoperative surgical planning. Because the presence of IVC wall invasion is a key determinant of the surgical approach—often necessitating complex vascular reconstruction—and serves as an important prognostic factor (18), its accurate preoperative prediction is crucial for optimizing surgical strategies and assessing patient outcomes. We identified lumbar vein diameter, craniocaudal thrombus length, renal vein ostium diameter, and the presence of an associated bland thrombus as independent predictors of IVC wall invasion. The combined model incorporating these four factors achieved a significantly greater AUC (0.81 in the training set and 0.84 in the validation set) than the best individual MRI predictor (AUC = 0.71) and radiologists’ subjective assessments (AUC = 0.66) (all P < .05), underscoring the value of integrating multidimensional information. While previous studies have explored the relationship between RCC with IVC tumor thrombus and venous wall invasion, precise preoperative assessment remains a clinical challenge (19). Our findings address this gap by providing a validated multiparametric approach derived from a large, uniform MRI dataset.

When our results are compared with those of prior studies, several important consistencies and discrepancies become apparent. In our univariable analysis, we observed significant differences in parameters such as the AP diameter of the IVC at the renal vein ostium, complete IVC occlusion, and maximum IVC diameter; however, these parameters were not retained as independent predictors in the final multivariable model. These findings differ from those of Psutka et al (22) and Overholser et al (23), who reported specific diameter thresholds (eg, an ostial diameter ≥ 24 mm) as independent predictors of wall invasion, whereas another study similarly identified an ostial AP diameter > 1.7 cm as an independent risk factor (24). These discrepancies may be attributed to heterogeneity in prior study designs, including limited sample sizes, mixed imaging modalities (CT and MRI), and broader inclusion criteria (eg, inclusion of non-RCC thrombi or thrombi confined to the renal vein), all of which could compromise the generalizability of single-parameter thresholds (20,21). By relying on a uniform MRI protocol and multivariable modeling, our study suggests that while diameter-related features are associated with invasion, they are less robust than the independent predictors identified in our final model.

Consistent with prior literature (25), we found that although multiple individual imaging indicators differed between groups, their standalone predictive performance remained suboptimal. Regarding the assessment of bland thrombi, it is worth noting a methodological distinction: Whereas Pei et al (25) incorporated bland thrombus length as a continuous variable, we treated its presence or absence as a binary predictor. This choice was driven by the primary clinical requirement of determining thrombus presence for surgical decision-making and by considerations of model robustness. Furthermore, while the predictive model developed by Alayed et al (26) achieved a high AUC of 0.91, that study was based on a small sample size (n = 24) and lacked independent validation. In contrast, our model was developed and validated in a larger sample, supporting its robustness and potential for clinical application.

To better interpret the biologic and hemodynamic significance of the independent predictors identified in our study, several mechanistic considerations merit discussion. First, the lumbar veins serve as major collateral pathways when the IVC is obstructed; therefore, lumbar vein dilatation observed at MRI indicates significant venous hypertension and obstruction (27), which may contribute to a hemodynamic milieu conducive to vascular involvement, although the underlying mechanisms warrant further investigation. Second, the craniocaudal length of the thrombus reflects the tumor burden and duration of contact with the vessel wall. A longer thrombus implies a larger surface area for adhesion and may cause local hemodynamic disturbances (eg, turbulence and stasis) (28) that promote invasion, although direct evidence remains limited. Biologically, the ability to form a lengthy thrombus often indicates an aggressive tumor phenotype (29). Third, irregular widening or a “trumpet-shaped” appearance at the ipsilateral renal vein ostium suggests that the tumor is not merely floating but has infiltrated through the renal vein wall into the IVC confluence, a site clinically known for frequent invasion (13). Finally, the formation of a bland thrombus is often a consequence of endothelial injury and stasis; concurrently, inflammatory mediators released during thrombosis may further promote tumor cell adhesion and invasion into the damaged vascular wall (30,31).

Our study had several limitations. First, the single-center design may limit the generalizability of the findings. Before the model can be considered for clinical translation, it requires multicenter external validation and calibration assessment to ensure that the predicted probabilities of invasion match the true risk, thereby avoiding potential underestimation or overestimation. Broader applicability needs to be confirmed in future studies. Second, owing to the low incidence of RCC with IVCTT, the sample sizes—particularly those of the prospective validation set—remained relatively limited. This resulted in wide CIs for the estimated AUC and some predictors, reflecting a degree of uncertainty in precise effect size estimation. However, given the rarity of the disease, this study sample represents an important collection of high-quality data. Third, as this was a retrospective study without routine radical resection of the caval wall in all patients, there remains a theoretical risk of occult microscopic invasion in cases managed with thrombectomy alone. Although our strict intraoperative criteria aimed to minimize ambiguous cases, the absence of pathologic confirmation for noninvasion cases is an inherent limitation of such surgical series. Fourth, retrospective data were acquired using two different MRI systems (1.5 T and 3.0 T). However, the parameters assessed were based on morphology rather than signal intensity and thus less susceptible to field strength variability. Furthermore, all prospective validation cases were uniformly scanned with a 3.0-T system to ensure consistency. Finally, the parameters were measured manually, which may have introduced interobserver variability. Future studies could explore artificial intelligence–based automated segmentation, radiomics analysis, and continued patient enrollment to further validate and refine the robustness of the model.

In conclusion, the multiparametric MRI-based model identified in our study offers a robust and noninvasive solution for predicting IVC wall invasion. By integrating the lumbar vein diameter, craniocaudal thrombus length, renal vein ostium diameter, and presence or absence of a bland thrombus, this tool outperforms individual imaging parameters and subjective assessment. Its clinical application has the potential to guide surgeons in accurately determining the need for venous wall resection, thereby optimizing surgical planning, reducing intraoperative uncertainty, and potentially improving patient outcomes.

Disclosures of conflicts of interest

Please see ICMJE form(s) for author conflicts of interest. These have been provided as supplemental materials