Habitat Analysis of Tumor Thrombus for Vena Cava Wall Invasion in Renal Cell Carcinoma

Background: Tumor thrombus in the inferior vena cava can invade the vessel wall. Wall invasion changes the surgical plan. Doctors must decide whether to resect or reconstruct the vein wall. Preoperative imaging has limited accuracy for wall invasion. Habitat imaging divides a tumor into subregions. Each subregion is a habitat. The method uses voxel-level image features. Habitat imaging can show intratumoral heterogeneity. No study has tested habitat imaging of tumor thrombus for wall invasion in renal cell carcinoma.

Objective: This study will measure the diagnostic performance of a habitat imaging model for inferior vena cava wall invasion. The study will compare the habitat model with three comparator models. The comparators are a conventional sign model, a whole-tumor radiomics model, and a clinical model. In an exploratory analysis, the study will test whether imaging habitats correspond to distinct spatial transcriptomic regions.

Design and setting: This is a single-center, retrospective, observational diagnostic accuracy and prediction model study. The study site is Peking University Third Hospital, Beijing, China. The study will use existing preoperative contrast-enhanced computed tomography (CT) and magnetic resonance imaging (MRI) images. The study will use existing clinical records and surgical pathology reports. The study will select participants from the institutional surgical cohort. The cohort includes patients with renal cell carcinoma and renal vein or inferior vena cava tumor thrombus.

Participants: The study will include adults who had radical nephrectomy with tumor thrombectomy at the study center. Each participant must have a preoperative contrast-enhanced CT image and a surgical pathology report. The pathology report must state the wall invasion status. Eligibility criteria are in the Eligibility section.

Index test: The study team will extract voxel-level local features from the tumor thrombus. The features will follow the Image Biomarker Standardisation Initiative (IBSI) definitions. The team will exclude percentile features and global shape features before analysis. The team will use stable features for global consensus clustering. The team will calculate the patient-level volume fraction of each habitat. The team will build a model for wall invasion from these habitat features.

Reference standard: The reference standard is the structured pathology report from surgery. Two pathologists will read each report independently. The pathologists will record wall invasion as present or absent. The pathologists will also record the wall invasion grade. The study team will calculate the agreement between the two readers. A third pathologist will resolve disagreements. The study team will not use intraoperative findings as the primary reference standard.

Statistical analysis: The primary metric is the area under the receiver operating characteristic curve (AUROC) with the 95% confidence interval. The study team will report sensitivity and specificity at a pre-specified threshold. The team will assess calibration with intercept, slope, and Brier score. The team will assess clinical utility with decision curve analysis. The team will use nested cross-validation and bootstrap for internal validation. The team will compare the habitat model with the three comparator models on the same participants. The team will report the differences with 95% confidence intervals.

Spatial transcriptomic validation: The study team will use available Visium spatial transcriptomic data from the study center. The team will compare imaging habitat composition with spatially resolved transcriptomic regions. The team will test habitat-associated gene expression and pathway signatures. The team will compare invasion and non-invasion regions in the paired sample. This analysis is exploratory.

Sample size: The planned sample size is 273 participants for the main imaging analysis. The study team used the prediction model framework to set this target. The study will use internal validation only. The study will not claim external validation. This validation sample is exploratory and is not part of the sample size calculation.

Missing data: The study team will code each missing or failed measurement as an independent failure reason. The team will not fill missing values with zero. The team will not impute missing values. The team will report the number and the characteristics of non-evaluable cases.

Reporting: The study will follow the STARD 2015 structure. The study team will also use the TRIPOD+AI framework for the prediction model. The team will use the CLEAR checklist and the METRICS criteria for the radiomics analysis. The team will report the exploratory spatial transcriptomic validation in the same article.

Ethics and data handling: The institutional human subjects review board reviewed and approved the study before data extraction. The consent procedure follows the review board determination. The study team will use only de-identified data. The team will not upload individual-level data to public cloud services or foreign platforms. This registration contains summary-level protocol information only.

Limitations: The study is single-center and retrospective. The study has no external validation. The reference standard depends on pathology reports and reader judgment. The study will report reader agreement. The selection of recent cases may confound scanner and protocol effects. Habitat boundaries may be unstable. The exploratory spatial transcriptomic validation uses a small sample. No habitat-specific reporting guideline exists.