CT Volumetry and Hepatic Vascular Deformation Mapping to Predict Post-Hepatectomy Liver Failure

Overview

Removing a large part of the liver (major hepatectomy) can cure primary and secondary liver cancers, but it carries the risk of post-hepatectomy liver failure (PHLF), a serious complication in which the liver left behind - the future liver remnant (FLR) - cannot meet the body's metabolic needs. PHLF occurs after roughly 5% to 15% of major hepatectomies and is the leading cause of postoperative death. Before surgery, surgeons routinely use CT scans to measure how much liver will remain (CT volumetry). Volume alone does not tell the whole story. After a large resection the remaining liver can rotate and shift into the empty space left behind, kinking or compressing the veins that drain it. The resulting congestion can make an apparently adequate remnant fail. This prospective, multicenter, observational cohort study tests whether adding two elements to standard CT volumetry improves the preoperative prediction of liver failure: 1. Hepatic vascular deformation mapping (VDM), a three-dimensional image-analysis technique that quantifies the geometry and displacement of the hepatic veins and the portal vein; and 2. A set of simple, reproducible measurements that any radiologist can make on a standard CT scan (hepatic vein diameters, congestion index, spleen volume, liver attenuation, and the distances from the veins to the planned resection plane). Consecutive adults undergoing major hepatic resection (three or more Couinaud segments) at several tertiary hepatobiliary centers will be enrolled. The preoperative CT scans already performed as part of routine care are analyzed centrally by a core imaging laboratory. There is no additional imaging, no extra hospital visit, and no study-specific intervention: every patient receives standard surgical care. Participants are followed for at least 30 days after surgery to record liver failure (International Study Group of Liver Surgery criteria, operationalized by the "50-50" rule on postoperative day 5) and radiologic evidence of hepatic congestion. Three nested prediction models are compared - volumetry alone; volumetry plus VDM; and volumetry plus VDM plus the simple radiology parameters - and the best-performing model is converted into a practical risk score. Because the study runs across several centers, the model can be validated by leaving one center out at a time, which gives an honest estimate of how well it would perform at a new hospital. The goal is a generalizable, easy-to-use tool that tells surgeons, before the operation, which patients are genuinely at risk of post-hepatectomy liver failure.

BACKGROUND AND RATIONALE Major hepatic resection is the only potentially curative treatment for many primary and secondary liver malignancies. Post-hepatectomy liver failure (PHLF) remains the most feared complication and the leading cause of postoperative mortality, occurring in approximately 5% to 15% of patients after major hepatectomy. Preoperative risk assessment currently rests on CT volumetry, which quantifies the future liver remnant (FLR) and confirms that sufficient functional hepatic mass will remain. Volumetry is, however, a static measurement: it does not capture the hemodynamic consequences of the resection. Hepatic venous congestion is an under-recognized contributor to PHLF. When major venous tributaries are sacrificed, or when the residual liver rotates and shifts into the empty subphrenic space after a large resection, the hepatic veins may kink or become compressed, producing outflow obstruction and functional impairment of an FLR that appeared volumetrically adequate. Vascular Deformation Mapping (VDM) is an image-analysis technique that uses deformable registration to quantify three-dimensional changes in vascular geometry, offering a way to identify "at-risk" vascular configurations before the operation. This study tests the hypothesis that integrating VDM-derived parameters and a set of simple, reproducible radiology parameters with conventional CT volumetry improves the preoperative prediction of PHLF. A multicenter design is essential: it accrues an adequate number of PHLF events for stable multivariable modeling, broadens the case-mix and the range of imaging and surgical practice represented, and permits internal-external cross-validation, which provides an honest estimate of how the model is expected to perform when transported to a new center. OBJECTIVES Primary objective: to determine whether the integration of CT volumetry with VDM-derived parameters improves the preoperative prediction of PHLF compared with volumetry alone. Secondary objectives: to correlate VDM and simple radiology parameters with radiologic evidence of postoperative hepatic congestion; to identify simple, clinically applicable radiology parameters suitable for routine use; to develop a practical scoring system for preoperative risk stratification; and to assess the transportability of the resulting model across participating centers by internal-external cross-validation. DESIGN AND SETTING Prospective multicenter observational cohort study conducted at tertiary referral centers specializing in hepatobiliary surgery. Each center prospectively enrolls consecutive eligible patients undergoing major hepatic resection. A central data-coordinating center oversees protocol harmonization, data integration, central image analysis, and quality control. The number of contributing centers is not fixed a priori; recruitment continues across sites until the target sample size is reached, with each center contributing consecutive eligible cases in order to minimize selection bias. To reduce the measurement heterogeneity inherent to multicenter data, all preoperative imaging undergoes central review by a core radiology laboratory using a single standardized analysis pipeline. Participating centers prospectively adopt a harmonized CT acquisition protocol; unavoidable deviations are documented and examined in sensitivity analyses. A common case report form, a data dictionary, and standard operating procedures for every measurement are distributed to all sites before data collection begins. No study-specific intervention is performed. Patients undergo standard-of-care major hepatic resection, and only routinely acquired imaging and clinical and laboratory data are collected for research analysis. IMAGING AND CENTRAL ANALYSIS Preoperative CT images are pseudonymized at source and transmitted to the central imaging core laboratory through a secure, de-identified data-transfer pipeline. Image processing and parameter extraction are performed centrally using specialized 3D modeling software (or, where centralization is impractical, by site readers trained and certified to a common standard operating procedure, with central audit). A random subset of at least 10% of studies from each center is independently double-read; inter-center, inter-observer, and intra-observer agreement are reported. Three parameter sets are extracted from the routinely acquired preoperative multiphasic contrast-enhanced CT: 1. VOLUMETRY - total liver volume (TLV, semi-automated segmentation on the portal venous phase); future liver remnant (FLR) volume, segmented according to the planned resection line; FLR ratio (FLR/TLV x 100); FLR-to-body-weight ratio; and resection volume (TLV minus FLR). 2. HEPATIC VASCULAR DEFORMATION MAPPING (VDM) - three-dimensional angle of the right, middle, and left hepatic veins to the long axis of the inferior vena cava (IVC); distance from the hepatic vein confluence to the IVC ostium; vascular displacement index (three-dimensional spatial shift of the major hepatic veins relative to fixed landmarks on post-resection simulation); and portal vein angulation relative to the planned resection plane. 3. SIMPLE RADIOLOGY PARAMETERS, measurable on any standard CT - right, middle, and left hepatic vein diameters at the IVC junction; portal vein diameter at the hilum; aortic diameter at the celiac trunk; congestion index (portal vein diameter / aortic diameter x 100); spleen volume; splenic vein diameter at its junction with the portal vein; IVC diameter at the level of the hepatic vein confluence; mean liver parenchymal attenuation on the portal venous phase; liver-to-spleen attenuation ratio; remnant liver attenuation homogeneity (standard deviation of Hounsfield units within the FLR); perpendicular distance from the right and middle hepatic vein ostia to the planned resection plane; diameter of the hepatic artery supplying the FLR; and remnant liver perfusion index (ratio of arterial-phase to portal venous-phase enhancement within the FLR). OUTCOME DEFINITIONS PHLF is defined according to the International Study Group of Liver Surgery (ISGLS) criteria and operationalized by the "50-50" rule: prothrombin time with INR greater than 1.7 together with serum bilirubin greater than 50 micromol/L on or after postoperative day 5. Postoperative hepatic congestion is identified on postoperative imaging as areas of delayed enhancement, mosaic attenuation, or periportal edema. Transient hepatic attenuation differences (THAD) are documented as a marker of vascular compromise. All outcomes and postoperative imaging findings are adjudicated centrally by two independent radiologists and a hepatobiliary surgeon, blinded to the center of origin and to the candidate VDM and radiology parameter values; disagreements are resolved by consensus or by a third adjudicator. SAMPLE SIZE The limiting quantity for a multivariable prediction model is the number of outcome events relative to the number of candidate predictors. A conventional events-per-variable (EPV) of 10 is applied. The maximal model (volumetry + VDM + simple radiology parameters) carries up to 10 candidate predictors and therefore requires 100 PHLF events; sizing to the largest of the three nested models ensures that all three comparisons are adequately powered. At an expected PHLF incidence of 12% (mid-range of the reported 5% to 15%), 100 / 0.12 = approximately 834 patients are required. An inflation factor of 15% is applied for between-center variation in case-mix, surgical practice, and imaging protocols, since center is modeled as a random effect rather than through survey-style design-effect inflation: 834 x 1.15 = approximately 960 patients. Allowing for up to 10% of records with incomplete follow-up or with imaging that fails central quality control: 960 / 0.90 = approximately 1,067 patients. The final target is approximately 1,070 patients, expected to yield approximately 128 PHLF events before adjustment for missingness (approximately 115 after). Adequacy is additionally confirmed against the criteria of Riley and colleagues for the minimum sample size required to develop a clinical prediction model . Where the two approaches diverge, the larger recommended sample is adopted. STATISTICAL ANALYSIS Continuous variables are summarized as mean and standard deviation or as median and interquartile range, as appropriate; categorical variables as frequencies and percentages. Univariate comparisons use the Student t test or the Mann-Whitney U test for continuous variables and the chi-square or Fisher exact test for categorical variables. The primary multivariable model is a mixed-effects logistic regression with center as a random intercept, which accounts for the clustered structure of multicenter data and for within-center correlation of outcomes. The incremental value of VDM and of the simple radiology parameters is assessed by comparing the area under the receiver operating characteristic curve (AUC-ROC) of three nested models: (1) volumetry alone; (2) volumetry + VDM; and (3) volumetry + VDM + simple radiology parameters. The pre-specified primary comparison is Model 2 versus Model 1. Calibration is assessed using calibration plots, the calibration slope, calibration-in-the-large, and the Hosmer-Lemeshow goodness-of-fit test. Between-center heterogeneity in baseline risk and in predictor effects is quantified. Model generalizability is evaluated by internal-external cross-validation (leave-one-center-out): the model is iteratively developed on all centers but one and validated on the omitted center, with discrimination (AUC-ROC) and calibration (calibration slope and calibration-in-the-large) assessed in each held-out center and pooled by random-effects meta-analysis. Inter-observer and intra-observer agreement for the radiologic measurements is evaluated using the intraclass correlation coefficient for continuous variables and Cohen kappa for categorical assessments, reported overall and stratified by center. Analyses are conducted in SPSS version 27, with mixed-effects modeling and internal-external validation. ETHICS AND DATA PROTECTION The protocol has been approved by the Medical Research Ethics Committee / Institutional Review Board of the Faculty of Medicine, Minia University (MUFMIRB approval number 2059:7/2026, dated 13 July 2026) and, prior to any enrollment, by the responsible research committee of each participating site. Written informed consent is obtained from every participant before enrollment, in accordance with local regulations and the Declaration of Helsinki. Patients undergo standard-of-care major hepatic resection; no study-specific intervention is performed. Data-sharing and transfer agreements govern the secure transmission of coded, de-identified data and imaging from each participating center to the coordinating center. No biological specimens are collected or stored, and no material is transferred outside the country.