Hepatic and Splenic Microcirculatory Perfusion for Ruling Out High-Risk Varices in Patients With Hepatitis B-Related Cirrhosis

This retrospective cohort study aims to evaluate whether hepatic and splenic intravoxel incoherent motion (IVIM) diffusion-weighted MRI parameters can noninvasively identify and "rule out" high-risk esophagogastric varices (HRV) in patients with chronic hepatitis B virus (HBV)-related cirrhosis or advanced chronic liver disease. The study is based on the Beijing Friendship Hospital All-Disease Platform and will include consecutive eligible patients up to October 31, 2025. Patients who underwent both upper gastrointestinal endoscopy and abdominal MRI including IVIM sequences within a 6-month interval will be included. Endoscopic findings will serve as the reference standard, and participants will be categorized into HRV and non-HRV groups (the definition of HRV follows the registered outcome measures).

Image Acquisition, Parameter Extraction, and Quality Control

IVIM analysis will be performed using multi-b-value diffusion-weighted imaging data and fitted with a biexponential model to separate true molecular diffusion from microcirculatory perfusion effects. Quantitative parameters will be derived from both liver and spleen, including D (true diffusion coefficient), D* (pseudo-diffusion coefficient reflecting perfusion-related components), f (perfusion fraction), and ADC (apparent diffusion coefficient). To minimize the influence of focal vessels, bile ducts, and obvious artifacts on parameter fitting, parameter extraction will be conducted only after image quality assessment. Sequences with suspected artifacts or significant distortion will be excluded or flagged to ensure data reliability. Hepatic and splenic IVIM parameters will serve as key independent variables for subsequent diagnostic performance evaluation and model development.

Comparator Indices, Clinical Variables, and Stratified Analyses

Commonly used noninvasive indices, including Baveno VI criteria, FIB-4, and APRI (calculated from routine laboratory and clinical data), will be included as comparators. Additional covariates will include demographic characteristics, complete blood count, coagulation profile, liver function tests, and liver stiffness measurement (LSM), when available. These variables will be used to describe baseline characteristics, control for potential confounding, and construct combined predictive models. In addition to the binary HRV outcome, associations between IVIM parameters and variceal severity grading as well as key clinical indicators will be explored. Prespecified subgroup analyses will be conducted to assess the diagnostic consistency and robustness of IVIM across different clinical subpopulations.

Statistical Analysis, Model Development, and Evaluation of the "Rule-Out" Strategy

Continuous variables will be expressed as mean ± standard deviation or median (interquartile range) after testing for normality, and categorical variables will be presented as frequencies and percentages. Between-group comparisons will be performed using the independent-samples t test or Mann-Whitney U test according to data distribution, while categorical variables will be compared using the chi-square test or Fisher's exact test. The diagnostic performance of individual hepatic and splenic IVIM parameters and their combinations for HRV will first be evaluated using receiver operating characteristic (ROC) curves and area under the curve (AUC). AUC values will be compared with those of noninvasive indices (Baveno VI, FIB-4, and APRI) using the DeLong test.

Subsequently, a multivariable logistic regression model will be constructed to develop an IVIM-based combined prediction model. To support a clinically applicable "rule-out" strategy, model performance will be evaluated under a prespecified sensitivity threshold of ≥95%, and the proportions of potentially avoided endoscopic examinations and missed HRV cases will be calculated for different candidate models to balance clinical benefit and safety. Internal validation will be performed using bootstrap resampling (B = 1000) to obtain bias-corrected AUC estimates and assess model stability. Correlation analyses will be conducted to examine relationships between IVIM parameters and variceal grading, platelet count, LSM, FIB-4, and APRI. All statistical tests will be two-sided, and P < 0.05 will be considered statistically significant.