Preoperative CT-Based Radiomics for Predicting Post-Hepatectomy Liver Failure and Assessing Liver Regeneration after ALPPS Stage I in Hepatitis B Patients

Preoperative CT-Based Radiomics for Predicting Post-Hepatectomy Liver Failure and Assessing Liver Regeneration after ALPPS Stage I in Hepatitis B Patients | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Preoperative CT-Based Radiomics for Predicting Post-Hepatectomy Liver Failure and Assessing Liver Regeneration after ALPPS Stage I in Hepatitis B Patients Ji Xu Guo, Shiqin pang, Deyang Huang, Tao Liu, Li Li, Sichen Feng, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9143445/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 5 You are reading this latest preprint version Abstract Aiming: This study aimed to develop a clinical–radiomics model based on preoperative dual-phase computed tomography to predict post-hepatectomy liver failure (PHLF) following stage I associating liver partition and portal vein ligation for staged hepatectomy (ALPPS) in patients with hepatitis B. Methods This retrospective study included 90 patients. Mixed-effects models were employed to assess the dynamic regeneration of the future liver remnant (FLR). Radiomics features were extracted using volumes of interest defined for both whole-FLR (wFLR) and partial-FLR (pFLR). Four machine learning algorithms, combined with nested cross-validation, were used to construct stable clinical-radiomics models. Interpretability analyses, as well as mediation analyses, were also performed. Results The incidence of grade B or C PHLF was 24.4%. The liver generation model demonstrated a significantly lower daily growth rate in the PHLF group compared to the non-PHLF group (3.67% vs. 5.61% per day, P = 0.003). The clinical, radiologic, and combined model based on wFLR achieved AUCs of 0.791, 0.892, and 0.894, respectively; those based on pFLR achieved AUCs of 0.791, 0.769, and 0.828. No significant difference in model performance was observed between the two segmentation strategies (P = 0.858), though pFLR segmentation substantially reduced workload. A preliminary mediation analysis suggested that 86.2% of the radiomics score’s total effect on PHLF was mediated by liver regeneration rate (OR 1.146, P < 0.001). Conclusion The proposed clinical–radiomics models effectively predict PHLF after ALPPS Stage I in patients with hepatitis B. The effect of the radiomics score on PHLF is mediated by impaired liver regeneration. ALPPS liver regeneration radiomics prediction hepatitis B Figures Figure 1 Figure 2 Figure 3 Introduction Associating liver partition and portal vein ligation for staged hepatectomy (ALPPS) is a surgical strategy that aims to promote rapid growth of the future liver remnant (FLR)( 1 ). Compared with conventional portal vein embolization (PVE), ALPPS can increase FLR by over 40% within 7–14 days, offering a substantial advantage( 2 , 3 ). However, its high rates of postoperative complications and mortality, particularly post-hepatectomy liver failure (PHLF), limit its widespread clinical adoption( 4 , 5 ). In patients with hepatitis B virus (HBV) infection, liver fibrosis or cirrhosis significantly impairs hepatic regeneration following ALPPS( 4 – 6 ). Therefore, predicting PHLF risk after stage I of ALPPS holds clinical significance in this population. Current predictive models primarily rely on clinical indicators such as the Child–Pugh score, MELD score, and FLR to the standard liver volume (SLV) ratio. However, they do not fully leverage the medical imaging information, leading to limited predictive accuracy ( 3 , 7 ). Recent advances in artificial intelligence have led to new approaches for preoperative assessment and risk prediction in liver surgery. Specifically, the integration of machine learning (ML) and radiomics has enabled high-throughput extraction and analysis of detailed quantitative imaging features from Computed Tomography (CT) or Magnetic Resonance Imaging (MRI) scans( 8 ). This approach has been applied in multiple clinical contexts within hepatobiliary surgery. For example, combined radiomics–clinical models have successfully predicted postoperative liver failure in patients with large hepatocellular carcinoma (> 10 cm), demonstrating strong performance( 9 ). Similar to PVE, comprehensive models can accurately predict FLR growth, thereby informing optimal timing for surgery( 10 , 11 ). These findings suggest that radiomics features reflect the functional status of liver tissue at cellular and molecular levels, offering novel insights into assessing regenerative potential. This study aims to develop a predictive model for PHLF after stage I ALPPS in patients with HBV, performing a radiomics analysis and machine learning on preoperative clinical, non-contrast CT (NCCT), and portal venous phase CT (PVCT) data. Additionally, the present study compares the efficiency between whole-FLR (wFLR) and partial-FLR (pFLR) to determine whether predictive accuracy can be maintained while reducing computational workload. Such a model could assist surgeons in identifying high-risk patients preoperatively and uncover potential imaging biomarkers associated with liver regeneration. Materials and Methods Patients This retrospective study included patients with HBV infection who underwent the ALPPS Stage I in the First Affiliated Hospital of Guangxi Medical University between January 2018 and December 2024. This research was approved by the Ethics Committee of our institution (2025-E0957). Inclusion criteria were: ( 1 ) contrast-enhanced CT was performed within 14 days preoperatively, with operative records and postoperative imaging confirming standard ALPPS Stage I; ( 2 ) preoperative Child-Pugh class A or B liver function. Exclusion criteria were: ( 1 ) preoperative imaging indicating unresectable lesions in FLR, main portal vein tumor thrombosis, or extrahepatic metastasis; ( 2 ) postoperative imaging revealing ligation located at a secondary branch of the right portal vein; ( 3 ) non-standard ALPPS variants (e.g., Tourniquet ALPPS, Haro-ALPPS, or other hybrid procedures). Standard ALPPS procedure The ALPPS technique was performed in two stages ( 12 , 13 ). In Stage I, the right portal vein is ligated, the liver parenchyma is partitioned in situ along the planned resection plane (typically along Cantlie’s line), and a cholecystectomy is conducted. In Stage II, right hepatectomy or extended right hepatectomy was completed to achieve R0 resection. Definition of PHLF PHLF was defined according to the International Study Group of Liver Surgery (ISGLS) criteria: increased international normalized ratio (INR > 1.5) and hyperbilirubinemia (serum total bilirubin > 1.2 mg/dl) on or after postoperative day 5( 14 ). Specifically, PHLF was graded as A, B, or C based on clinical management and severity. The primary outcome of this study was clinically significant PHLF (grades B and C). Liver Volume Measurement and Regeneration Model SLV was calculated according to the formula described by Urata et al.( 15 ). FLR volume was measured based on radiological images. The preoperative estimated FLR was delineated using the actual in-situ partitioning plane confirmed by postoperative imaging registration. Postoperative FLR volumes were measured based on available imaging until the day before stage II or within 90 days after stage I. To assess the regeneration kinetics of FLR after Stage I, the relative volume was used as the primary metric. Liver regeneration was modeled first using linear mixed-effects (LME) models and log-transformed LME (log-LME) models fitted to longitudinal volume data collected at multiple time points. The model showing the lowest Akaike information criterion (AIC) was selected as the optimal model and used to calculate individual liver regeneration rates. All formulas are detailed in Supplementary Table S1 . Image Segmentation and Radiomics Feature Extraction CT images were acquired from different scanner models, tube voltage, and slice thickness (Supplementary Table S2). To eliminate inter-scanner variability, all images underwent standardized preprocessing: resampling to 1×1×1 mm³ using B-spline interpolation, clipping of Hounsfield units to [-100, 400], and gray-level discretization with a fixed bin width of 25 HU. Two types of volumes of interest (VOIs) were delineated for model construction and comparison: whole-FLR (wFLR) and partial-FLR (pFLR)(Fig. 1 ). The whole-FLR (wFLR) was first automatically segmented from preoperative portal venous phase CT (PVCT) images using TotalSegmentator v1.5.7 ( 16 ). Subsequently, the resulting segmentation was refined manually by a surgeon and a radiologist to simulate the in-situ partition plane during surgery. The remaining liver after virtual resection served as the wFLR-VOIs. For the pFLR-VOIs, a 2-cm radius sphere was manually segmented at the portal venous level in segments S2/3. Large intrahepatic vessels and obvious non-hepatic tissues were excluded from all VOIs. Registration between non-contrast CT(NCCT) and PVCT images was performed to ensure spatial consistency of VOIs across sequences. All final VOIs were independently reviewed and corrected by two senior radiologists with more than 5 years of experience in liver imaging. All steps were performed using ITK-SNAP v4.2.0( 17 ). Radiomics features were extracted using PyRadiomics v3.1.0 (Supplementary Table S3). Radiomics Feature Selection and Model Construction Multiple strategies were employed to ensure feature robustness and model stability due to the limitation of a relatively small sample size, including z-score normalization, intraclass correlation coefficients, and cross-validation(CV)(Fig. 1 ). Stable features were selected through stability selection based on 10 runs of repeated 5-fold cross-validation. Within each training fold, features were removed using the independent-samples t-test, Pearson correlation analysis and least absolute shrinkage and selection operator (LASSO) regression. The selection frequency of each feature was recorded across all iterations, with features showing a selection frequency of ≥ 50%. Therefore, these features were used to establish radiomics models. Four supervised machine learning classifiers were tested: logistic regression (LR), random forest (RF), support vector machine (SVM) with RBF kernel, and extreme gradient boosting (XGBoost). To reduce the risk of overfitting, a nested stratified CV framework was implemented, comprising an outer 5-fold stratified CV for performance estimation and class balance, as well as an inner 3-fold stratified CV for hyperparameter tuning. The classifier with the highest mean area under the receiver operating characteristic (ROC) curve (AUC) in the outer loop and the most stable learning curve was selected as the final radiomics model and used to calculate radiomics score for each patient. Univariate and multivariable LR analyses were performed on clinical variables to construct the clinical model and generate clinical scores. A combined clinical-radiomics model was then developed by LR. The predictive performance of all models was assessed using AUC, with differences compared by the DeLong test. Calibration was assessed with calibration curves. Clinical utility was evaluated by decision curve analysis (DCA). Interpretability To enhance the interpretability of the radiomics models, SHAP (SHapley Additive exPlanations) was used to assess the contribution of individual radiomics features to the model’s predictions. Additionally, Pearson correlation analysis was applied to explore the relationship between liver regeneration rate and radiomics features. Moreover, causal mediation analysis was performed to assess whether the liver regeneration rate mediates the relationship between preoperative radiomic scores and PHLF. Statistical Analysis Prediction models were developed using Python (version 3.9). All other statistical analyses were performed using R software (version 4.2.1). The ‘lattice’ package was used to plot liver volume growth curves for individual patients, and the ‘nlme’ package was used to construct liver growth models. A significance level of p < 0.05 in the two-sided tests was considered statistically significant. Results Clinical characteristics PHLF grade B/C was found in 22 patients (24.4%) (Table 1). Compared with the non-LF group, the LF group exhibited significantly higher MELD scores (P < 0.001), aspartate aminotransferase (P = 0.049), and total bilirubin levels (P = 0.036). Liver Volume and Regeneration Model The preoperative mean SLV was 1206 cm³, and the mean estimated FLR was 353 cm³, resulting in a mean FLR/SLV ratio of 29% (Table 1, Fig. 2 ). After Stage I, the patients underwent a comprehensive evaluation, with 68 patients achieved sufficient FLR to proceed to Stage II. In this group, the median interval between two stages was 15.5 days (IQR: 13–26 days), with the FLR volume exhibiting a mean percentage increase of 152% from baseline (IQR: 142%-170%). Nonetheless, 7 patients developed PHLF but ultimately proceeded to Stage II after treatment and liver functional recovery. However, 22 patients did not undergo Stage II surgery due to PHLF after treatment (n = 15), tumor progression (n = 5), and insufficient liver regeneration (n = 1). The dynamic changes in liver volume are displayed in Fig. 2 , revealing a characteristic pattern of rapid FLR growth in the early postoperative phase, followed by a deceleration toward a plateau. Subsequently, a liver regeneration model was established between the LF and non-LF group by using LME and Log-LME, with a particular focus on the early postoperative period (within 10 and 15 days). The results demonstrated that the non-LF group had a significantly higher daily FLR growth rate than the LF group throughout the full period and during the early phase (Table 2). Among the models tested, the 10-day early Log-LME exhibited the best goodness-of-fit and was used to calculate the individual liver regeneration rates. Radiomics Feature Extractor and Model During VOI segmentation, the average time required for wFLR-VOIs was 0.5–1 hour with the assistance of the TotalSegmentator plugin, whereas the process only required 1 to 2 minutes for pFLR-VOIs. Following the stable feature selection process, 5, 10, 4, and 3 stable features significantly associated with LF were identified from wFLR-NCCT, wFLR-PVCT, pFLR-NCCT, and pFLR-PVCT, respectively (Supplementary Table 4). Pearson correlation analysis revealed that the correlation coefficients between wFLR and pFLR features were all below 0.8, indicating weak linear correlations (Supplementary Fig. 1). Finally, based on the outer-layer average scores of nested stratified CV and learning curves, LR was selected as the optimal machine learning model for all four models (Supplementary Table 5, Supplementary Fig. 2). Hence, a radiomics score was generated. In comparison, the other machine learning models exhibited varying degrees of underfitting or overfitting. Clinical Prediction Model Univariate analysis revealed that preoperative ALT (P = 0.044) and MELD score (P = 0.003) were associated with the occurrence of PHLF. The variance inflation factor (VIF) for both variables was less than 5. Multivariate logistic regression analysis identified MELD score (P = 0.006) as an independent predictor of PHLF, while ALT showed no statistical significance (Supplementary Table S6). A logistic regression (LR) clinical prediction model was constructed using the MELD score, and a clinical model score was calculated. Comprehensive Prediction Model Through nested CV, the AUCs of the clinical model, radiologic models, and the comprehensive model were 0.791, 0.892, and 0.894 for wVOIs, and 0.791, 0.769, and 0.828 for pVOIs, respectively (Table 3). Calibration curves and decision curve analysis (DCA) further confirmed the clinical usefulness of the comprehensive model (Fig. 1 ). Furthermore, DeLong’s test indicated no significant difference in predictive performance between the integrated models constructed using wVOIs and pVOIs segmentation (P = 0.238). The results demonstrated that partial VOI segmentation achieves comparable efficacy to whole VOI segmentation while substantially reducing manual workload. Interpretability of Radiomics Features The interpretability of the machine learning model was assessed using SHAP (SHapley Additive exPlanations) values (Supplementary Fig. 3–6). Pearson correlation coefficients were calculated between radiomics features and individual regeneration rate, with p-values adjusted using the false discovery rate (FDR) method (Supplementary Table S7). Among the 22 PHLF-related features, 2 features were positively correlated with liver regeneration rate, and the remaining features exhibited negative correlations . The mediation analysis was performed using the wVOIs (Fig. 3 ). Radiomic scores derived from models based on wVOIs were confirmed as independent risk factors for PHLF (OR 1.171, P < 0.001). The mediation analysis revealed that 86.2% of the total effect was mediated by the 10-day liver regeneration rate (OR 1.146, P < 0.001), whereas the direct effect of the radiomic scores was not significant (OR 1.022, P = 0.853). Discussion This retrospective study analyzed clinical data and preoperative CT images from HBV-infected patients undergoing ALPPS Stage I. Liver regeneration dynamics were quantified using Log-LME modeling, and a comprehensive predictive model was developed to predict the occurrence of PHLF. The comprehensive model demonstrated strong predictive performance in our cohorts, offering a novel quantitative tool for early risk assessment. Collectively, the correlation coefficients and mediation analysis highlighted the central role of liver regeneration in linking radiomic features to PHLF in ALPPS. These findings provide mechanistic insights into liver regeneration. Predictive modeling in ALPPS has evolved from relying solely on traditional clinical parameters to the integration of multi-omics data. Early studies relied on preoperative liver function and FLR volume for risk assessment; however, this approach did not account for interindividual regenerative variability and failed to leverage the rich information in medical imaging( 18 , 19 ). Advances in imaging have enabled the incorporation of CT- or MRI-based morphological evaluations to predict FLR growth, with machine learning enhancing model accuracy( 20 , 21 ). Our clinical-radiomics model is based on radiomics and machine learning, but innovatively associates the FLR dynamic regeneration with radiomics features. Moreover, SHAP was incorporated for interpretable machine learning analysis, mitigating the black-box nature of traditional models to some extent and providing a more comprehensive perspective. Notably, no statistically significant differences were observed between partial and whole segmentation strategies. This may be attributed to the parenchymal nature of the liver or the diffuse pathology associated with HBV, although partial marking risks the loss of shape features. Mechanistically, ALPPS-driven rapid regeneration involves hemodynamic shifts, inflammatory cytokine release, and cell proliferation pathway activation( 22 – 24 ). Portal vein ligation markedly increases portal flow to the remnant liver, activating endothelial nitric oxide synthase (eNOS) via shear stress to drive hepatocyte cell cycle entry( 25 ). Additionally, parenchymal transection releases IL-6 and TNF-α, further promoting proliferation( 26 , 27 ). Previous studies have suggested that HBV infection may impair regeneration by inducing fibrosis and altering the microenvironment( 4 , 5 ). Zhang et al. reported significantly higher 90-day mortality in severe fibrosis patients post-ALPPS (P = 0.014), consistent with our finding that preoperative MELD score predicts stage-I PHLF risk( 4 ). Furthermore, our study identified multiple radiomics features associated with ALPPS-related liver regeneration. Notably, mediation analysis indicated that the effect of radiomics feature on PHLF risk was primarily mediated by the 10-day liver regeneration rate. Therefore, radiomics may partially reflect early regenerative capacity, providing a novel perspective for liver regeneration. Recent advances in conversion therapy, such as interventional hepatoma therapy, targeted therapy, and immunotherapy, have shifted focus away from ALPPS, but the latter still offers unique benefits in enhancing FLR regeneration and resectability. A multicenter study by Lv et al. reported a 5-year survival rate of 31.7% following ALPPS in patients with HBV-related cirrhosis, which was significantly higher than that observed with TACE (P < 0.001). These results highlight the importance of patient selection, particularly for those with localized tumors and well-compensated liver function( 28 ). Future directions include integrating multi-omics (e.g., transcriptomics, proteomics) to elucidate radiomic mechanisms, shifting from correlation to causation( 29 , 30 ). Combining ALPPS with interventions, targeted therapy, or immunotherapy—such as the AITI conversion regimen—shows promising potential( 31 ). With advancements in artificial intelligence (AI) technology, deep learning models based on time-series data hold potential to further enhance the accuracy of PHLF prediction( 32 ). The innovative aspects of this study are summarized below. First, a nonlinear mixed-effects model was employed, overcoming the limitations of conventional static volume measurements and capturing key parameter changes during the regenerative process. Secondly, deep learning segmentation technology (TotalSegmentator) was integrated with standardized radiomics processes, thereby reducing subjective biases due to manual operations; in addition, the partial segmentation strategy further decreases the workload for labeling. Thirdly, the “black-box” nature of radiomics features was mitigated through multiple interpretative approaches. Nevertheless, the limitations of the present study should be acknowledged. Firstly, the single-center study and relatively small sample size remain a major limitation in this study. Although various methods, such as learning curves, nested CV, and four machine learning algorithms suitable for small datasets, have been employed to maintain model stability and generalization, larger samples are required for further validation and training. Moreover, the study participants involved primarily patients with HBV infection, so the applicability of the findings to patients with other etiologies should be carefully verified. In addition, preoperative FLR volume was estimated based on short-term postoperative CT registration, which may lead to measurement errors. Finally, although the model demonstrated good predictive performance and radiomics features were statistically associated with liver regeneration, the underlying biological mechanisms still require validation through further basic research. In conclusion, the clinical-radiomics model established in this study can effectively predict the risk of PHLF after ALPPS Stage I. The partial segmentation strategy significantly improves efficiency while maintaining the predictive performance, providing a valuable tool for clinical practice. Future prospective multicenter studies are warranted to validate the clinical applicability of this model and to explore optimized strategies combining ALPPS with systemic therapies to further improve patient outcomes. Declarations Disclosure Statement: The authors declare no commercial or financial conflicts of interest related to this study. No financial or material support was received. Author Contributions Jixu Guo: statistical analysis and manuscript writing; Shiqin Pang, Deyang Huang, Shengjie Xie: data acquisition; Sichen Feng, Li Li: image segmentation; Shuiping Yu: study conception and design, critical revision, and supervision. Ethics approval This retrospective study involving human participants was approved by the Medical Ethics Committee of The First Affiliated Hospital of Guangxi Medical University (Approval No. 2025-E0957). All procedures performed in this study were in accordance with the ethical standards of the institutional and/or national research committee and with the Declaration of Helsinki and its later amendments or comparable ethical standards. The requirement for informed consent was waived by the Medical Ethics Committee due to the retrospective nature of the study. Human Ethics declaration This study involved human participants and was approved by the Medical Ethics Committee of The First Affiliated Hospital of Guangxi Medical University (Approval No. 2025-E0957). Funding This study received no funding. No competing interests exist. Conflict of Interest The authors declare no conflicts of interest. Data statement The datasets are not publicly available to protect patient privacy, but they can be obtained from the corresponding author upon reasonable request via email. Acknowledgments We would like to thank the developers of ITK-SNAP for providing the open-source software for medical image segmentation, and the authors of the TotalSegmenter for facilitating automated liver segmentation. We thank Home for Researchers editorial team (www.home-for-researchers.com) for language editing service. References Schnitzbauer AA, Lang SA, Goessmann H, Nadalin S, Baumgart J, Farkas SA, Fichtner-Feigl S, et al. Right portal vein ligation combined with in situ splitting induces rapid left lateral liver lobe hypertrophy enabling 2-staged extended right hepatic resection in small-for-size settings. Ann Surg. 2012;255:405–14. Chen H, Wang X, Zhu W, Li Y, Yu Z, Li H, Yang Y et al. Application of associating liver partition and portal vein ligation for staged hepatectomy for initially unresectable hepatocellular carcinoma. BMC Surg 2022;22. Zhang J, Xu Y, Yang H, Huang H, Bian J, Jiang S, Sang X, et al. Application of associating liver partition and portal vein ligation for staged hepatectomy for hepatocellular carcinoma related to hepatitis B virus: comparison with traditional one-stage right hepatectomy. Translational cancer Res. 2020;9:5371–9. Zhang J, Zhang L, Yang X, Zheng Y, Xu H, Du S, Mao Y, et al. Liver fibrosis as a predictor of liver failure and outcome following ALPPS among patients with primary liver cancer. Sci Rep. 2024;14:15827. Hung K-C, Wang H-P, Li W-F, Lin Y-C, Wang C-C. Single center experience with ALPPS and timing with stage 2 in patients with fibrotic/cirrhotic liver. Updates Surg. 2024;76:1213–21. Qin Y, Li C, Ge X, Zhang Q, Wei X, Liu R. MMP2/9 downregulation is responsible for hepatic function recovery in cirrhotic rats following associating liver partition and portal vein ligation for staged hepatectomy. Ann Transl Med. 2022;10:468. Chan A, Zhang WY, Chok K, Dai J, Ji R, Kwan C, Man N et al. ALPPS Versus Portal Vein Embolization for Hepatitis-related Hepatocellular Carcinoma: A Changing Paradigm in Modulation of Future Liver Remnant Before Major Hepatectomy. Annals of Surgery 2021;273. Liu W, Liu X, Peng M, Chen GQ, Liu PH, Cui XW, Jiang F, et al. Artificial intelligence for hepatitis evaluation. World J Gastroenterol. 2021;27:5715–26. Xiang F, Liang X, Yang L, Liu X, Yan S. CT radiomics nomogram for the preoperative prediction of severe post-hepatectomy liver failure in patients with huge ([greater than or equai to] 10 cm) hepatocellular carcinoma. World J Surg Oncol. 2021;19:NA–NA. Gerwing M, Schindler P, Katou S, Köhler M, Stamm AC, Schmidt VF, Heindel W, Gerwing, et al. Ann Surg Oncol. 2023;30:7976–85. Wang Q, Brismar TB, Björk D, Baubeta E, Lindell G, Björnsson B, Sparrelid E. Development and External Validation of a Combined Clinical–Radiomic Model for Predicting Insufficient Hypertrophy of the Future Liver Remnant following Portal Vein Embolization. Ann Surg Oncol. 2025;32:1795–807. Guidetti C, Müller PC, Magistri P, Jonas JP, Odorizzi R, Kron P, Guerrini G, et al. Full robotic versus open ALPPS: a bi-institutional comparison of perioperative outcomes. Surg Endosc. 2024;38:3448–54. Caringi S, Delvecchio A, Dezio M, Casella A, Ferraro V, Filippo R, Stasi M et al. Feasibility and safety of ALPPS procedure: our experience. Surg Endosc 2025:1–9. Rahbari NN, Garden OJ, Padbury R, Brooke-Smith M, Crawford M, Adam R, Koch M, et al. Posthepatectomy liver failure: a definition and grading by the International Study Group of Liver Surgery (ISGLS). Surgery. 2011;149:713–24. Urata K, Kawasaki S, Matsunami H, Hashikura Y, Ikegami T, Ishizone S, Momose Y, et al. Calculation of child and adult standard liver volume for liver transplantation. Hepatology. 1995;21:1317–21. Wasserthal J, Breit H-C, Meyer MT, Pradella M, Hinck D, Sauter AW, Heye T et al. TotalSegmentator: robust segmentation of 104 anatomic structures in CT images. Radiology: Artificial Intelligence. 2023;5:e230024. Yushkevich PA, Piven J, Hazlett HC, Smith RG, Ho S, Gee JC, Gerig G. User-guided 3D active contour segmentation of anatomical structures: significantly improved efficiency and reliability. NeuroImage. 2006;31:1116–28. Khan AS, Garcia-Aroz S, Ansari MA, Atiq SM, Senter-Zapata M, Fowler K, Doyle M, et al. Assessment and optimization of liver volume before major hepatic resection: current guidelines and a narrative review. Int J Surg. 2018;52:74–81. Lopez-Lopez V, Linecker M, Cruz J, Brusadin R, Lopez‐Conesa A, Machado MA, Hernandez‐Alejandro R, et al. Liver growth prediction in ALPPS–A multicenter analysis from the international ALPPS registry. Liver Int. 2022;42:2815–29. Than VS, Le TD, Cao MT, Pham MT. Simultaneous Portal and Hepatic Vein Embolization versus Portal Vein Embolization Only in Patients with Hepatocellular Carcinoma: A Retrospective Review of Safety and Effectiveness. Journal of Vascular and Interventional Radiology; 2025. Wang X, Qiu C, Du X, Qin J, Zhang Y, Hu Z, Luo Y et al. Fatty liver regeneration after partial hepatectomy: an experimental study based on intravoxel incoherent motion and T2* mapping MRI. Magnetic Resonance Materials in Physics, Biology and Medicine 2025:1–11. Xu H, Qiu X, Wang Z, Wang K, Tan Y, Gao F, Perini MV, et al. Role of the portal system in liver regeneration: From molecular mechanisms to clinical management. Liver Res. 2024;8:1–10. Zhang S, Ma Y, Chen X, Wu S, Chen G. Circulating proliferative factors versus portal inflow redistribution: mechanistic insights of ALPPS-derived rapid liver regeneration. Front Oncol. 2025;14:1429564. Chen X, Zhou B, Wu S, Li G, Ma Y, Chen P, Chen G. Impact of the extent and location of liver split on future liver remnant hypertrophy after portal vein ligation in a rat model. Surgery. 2024;175:1321–8. Masuo H, Shimizu A, Motoyama H, Kubota K, Notake T, Yoshizawa T, Hosoda K, et al. Impact of endothelial nitric oxide synthase activation on accelerated liver regeneration in a rat ALPPS model. World J Gastroenterol. 2023;29:867. Tihanyi DK, Szijarto A, Fulop A, Jiang D, Ernst L, Meister FA, Bleilevens C, et al. Exploring the Enhanced Liver Regeneration Patterns Following ALPPS Versus Selective Portal Vein Ligation in an Experimental Model. Cancer Rep. 2025;8:e70221. Du Y, Yang Y, Zhang Y, Zhang F, Wu J, Yin J. Unraveling enhanced liver regeneration in ALPPS: Integrating multi-omics profiling and in vivo CRISPR in mouse models. Hepatol Commun. 2025;9:e0630. Lv J-H, Chen W-Z, Li Y-N, Wang J-X, Fu Y-K, Zeng Z-X, Wu J-Y, et al. Should associating liver partition and portal vein ligation for staged hepatectomy be applied to hepatitis B virus-related hepatocellular carcinoma patients with cirrhosis? A multi-center study. HPB. 2022;24:2175–84. Lefebvre ATN, Ghosh S, Baciu C, Hasjim BJ, Naimimohasses S, Oldani G, Pasini E et al. Modelling the Liver’s Regenerative Capacity across Different Clinical Conditions. JHEP Rep 2025:101465. Guan Y, Cui L, Inchai J, Fang Z, Law J, Brito AAG, Pawlosky A et al. AI-Assisted Drug Re‐Purposing for Human Liver Fibrosis. Adv Sci 2025:e08751. Chen Z, Chen X, Hu H, Chen K, Xiao H, Du C, Lan X. The combination of associating liver partition and portal vein ligation for staged hepatectomy (ALPPS), interventional hepatoma therapy, targeted therapy, and immunotherapy: a case series of a novel AITI conversion therapy model. J Gastrointest Oncol. 2025;16:1736–48. Wang K, Yang Q, Li K, Tang S, Zhang B, Liao X, Du S et al. Learning-based early detection of post-hepatectomy liver failure using temporal perioperative data: a nationwide multicenter retrospective study in China. eClinicalMedicine 2025;83. Tables Table 1 to 4 are available in the Supplementary Files section. Additional Declarations No competing interests reported. Supplementary Files Table1.xlsx Table 1. The Clinical characteristics of all patients Table2.xlsx Table 2. Goodness-of-Fit of Different Models for Liver Regeneration after Stage I Table3.xlsx Table 3: Univariate and Multivariate Logistic regression for clinical characteristic Table4.xlsx Table 4. Prediction performance of all models in different types SupplementaryTables.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviewers agreed at journal 11 Apr, 2026 Reviewers invited by journal 09 Apr, 2026 Editor assigned by journal 21 Mar, 2026 Submission checks completed at journal 19 Mar, 2026 First submitted to journal 16 Mar, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9143445","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":621584847,"identity":"2a606f5d-03c2-45b4-8193-cfb10d664f6b","order_by":0,"name":"Ji Xu Guo","email":"","orcid":"","institution":"First Affiliated Hospital of GuangXi Medical University","correspondingAuthor":false,"prefix":"","firstName":"Ji","middleName":"Xu","lastName":"Guo","suffix":""},{"id":621584848,"identity":"04a46132-ced8-45fc-b9e0-cb683feed087","order_by":1,"name":"Shiqin pang","email":"","orcid":"","institution":"First Affiliated Hospital of GuangXi Medical University","correspondingAuthor":false,"prefix":"","firstName":"Shiqin","middleName":"","lastName":"pang","suffix":""},{"id":621584849,"identity":"753a43ff-ca04-4f37-bdee-1fa0a215cca9","order_by":2,"name":"Deyang Huang","email":"","orcid":"","institution":"First Affiliated Hospital of GuangXi Medical University","correspondingAuthor":false,"prefix":"","firstName":"Deyang","middleName":"","lastName":"Huang","suffix":""},{"id":621584850,"identity":"8d8d47fb-9c7f-444a-ba23-cd60c6691dd7","order_by":3,"name":"Tao Liu","email":"","orcid":"","institution":"First Affiliated Hospital of GuangXi Medical University","correspondingAuthor":false,"prefix":"","firstName":"Tao","middleName":"","lastName":"Liu","suffix":""},{"id":621584851,"identity":"313baf70-ada4-4094-89ab-9b6f734e94a2","order_by":4,"name":"Li Li","email":"","orcid":"","institution":"first affiliated hospital of Guilin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Li","middleName":"","lastName":"Li","suffix":""},{"id":621584852,"identity":"f1453a9d-893b-422d-9df5-0a8f82617b38","order_by":5,"name":"Sichen Feng","email":"","orcid":"","institution":"The People's Hospital of Guangxi Zhuang Autonomous Region","correspondingAuthor":false,"prefix":"","firstName":"Sichen","middleName":"","lastName":"Feng","suffix":""},{"id":621584853,"identity":"70b081c4-f613-4266-b857-acf1400a6ffe","order_by":6,"name":"Shenjie Xie","email":"","orcid":"","institution":"First Affiliated Hospital of GuangXi Medical University","correspondingAuthor":false,"prefix":"","firstName":"Shenjie","middleName":"","lastName":"Xie","suffix":""},{"id":621584854,"identity":"772a78fa-b6ca-4ba1-a80c-e87855e21a59","order_by":7,"name":"Shuiping Yu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAy0lEQVRIiWNgGAWjYBACxvmPj39I/Pevfn97A5FamBvS0hgesB1g3MBzgEgt7A05ZoxgLRIJRGrhbTiW9iCB5w6zueTjjTcYamyiCWqRbGw+bpAg8YzNcnZasQXDsbTcBkJaDJvZEiQSDJh5GG7nmEkwNhwmrMX+GI+BREICswTDzTNEamHs4TGTSDhw2MDgBg+xWmawJRskNqQlSPYA/ZJAjF8YZzAffPizwSaBn/3wxhsfamwIa0EGBkRHDZIWUnWMglEwCkbByAAAbElDL0IeKwcAAAAASUVORK5CYII=","orcid":"","institution":"First Affiliated Hospital of GuangXi Medical University","correspondingAuthor":true,"prefix":"","firstName":"Shuiping","middleName":"","lastName":"Yu","suffix":""},{"id":621584855,"identity":"b06fb4e3-47c9-416d-9809-b5f8318ed3d5","order_by":8,"name":"Songqing He","email":"","orcid":"","institution":"First Affiliated Hospital of GuangXi Medical University","correspondingAuthor":false,"prefix":"","firstName":"Songqing","middleName":"","lastName":"He","suffix":""}],"badges":[],"createdAt":"2026-03-17 03:24:00","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9143445/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9143445/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":107257628,"identity":"c8818c42-0edf-43d2-a69f-009d7707b4f4","added_by":"auto","created_at":"2026-04-19 12:32:21","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":4847531,"visible":true,"origin":"","legend":"\u003cp\u003eFlow chart of the study of radiomics-based prediction of post-hepatectomy liver failure (PHLF) in patients undergoing associating liver partition and portal vein ligation for staged hepatectomy (ALPPS).\u003c/p\u003e","description":"","filename":"Figure1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9143445/v1/86c4f2ebd0817aa31b2bb824.jpg"},{"id":107257630,"identity":"a4a4662a-9403-4f8e-a4c7-f7ad4e1da457","added_by":"auto","created_at":"2026-04-19 12:32:21","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":2396518,"visible":true,"origin":"","legend":"\u003cp\u003eThe dynamic changes in liver volume at different time points after the ALPPS stage I. Note: The day of Stage I ALPPS surgery was designated as Day 0.\u003c/p\u003e","description":"","filename":"Figure2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9143445/v1/fbd948fe87fd54226acda08e.jpg"},{"id":107257633,"identity":"d9a12e0e-4d16-42c4-b687-82ccdbe4d3f9","added_by":"auto","created_at":"2026-04-19 12:32:21","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":860140,"visible":true,"origin":"","legend":"\u003cp\u003eMediation of the effect of preoperative radiomic scores on PHLF by the 10-day liver regeneration rate. Abbreviation: ACME: Average Causal Mediation Effect; ADE: Average Direct Effect.\u003c/p\u003e","description":"","filename":"Figure3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9143445/v1/1d8159e0914adc9a70db82c9.jpg"},{"id":107485916,"identity":"39858876-83ff-4bc4-8d22-cdc34647fde6","added_by":"auto","created_at":"2026-04-22 02:36:46","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":10810533,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9143445/v1/e61a58a3-b31d-4ac9-bd94-5ab769869b38.pdf"},{"id":107483370,"identity":"f50b5c94-926a-40bc-a9cb-25f7c5e7fbad","added_by":"auto","created_at":"2026-04-22 02:27:28","extension":"xlsx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":12791,"visible":true,"origin":"","legend":"\u003cp\u003eTable 1. The Clinical characteristics of all patients\u003c/p\u003e","description":"","filename":"Table1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-9143445/v1/a4ce726576dded734673d63b.xlsx"},{"id":107257631,"identity":"b507b042-bd05-4338-857f-3623c5392a5b","added_by":"auto","created_at":"2026-04-19 12:32:21","extension":"xlsx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":10098,"visible":true,"origin":"","legend":"\u003cp\u003eTable 2. Goodness-of-Fit of Different Models for Liver Regeneration after Stage I\u003c/p\u003e","description":"","filename":"Table2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-9143445/v1/258db70c5ef237ecfdf7afce.xlsx"},{"id":107483369,"identity":"20122ce8-cbe7-4b62-bfba-a29157351a3b","added_by":"auto","created_at":"2026-04-22 02:27:28","extension":"xlsx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":10454,"visible":true,"origin":"","legend":"\u003cp\u003eTable 3: Univariate and Multivariate Logistic regression for clinical characteristic\u003c/p\u003e","description":"","filename":"Table3.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-9143445/v1/f96998c20df8a02ccc594bcb.xlsx"},{"id":107257635,"identity":"af4e0135-632f-4208-a47b-25d58afe4f37","added_by":"auto","created_at":"2026-04-19 12:32:21","extension":"xlsx","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":10179,"visible":true,"origin":"","legend":"\u003cp\u003eTable 4. Prediction performance of all models in different types\u003c/p\u003e","description":"","filename":"Table4.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-9143445/v1/70eafc60c965973fe8827ce9.xlsx"},{"id":107257634,"identity":"e4c1a628-f3c3-4737-b1a4-c1c99a4f323d","added_by":"auto","created_at":"2026-04-19 12:32:21","extension":"docx","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":3924727,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTables.docx","url":"https://assets-eu.researchsquare.com/files/rs-9143445/v1/39b24250055eecc37fe1aa1b.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Preoperative CT-Based Radiomics for Predicting Post-Hepatectomy Liver Failure and Assessing Liver Regeneration after ALPPS Stage I in Hepatitis B Patients","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAssociating liver partition and portal vein ligation for staged hepatectomy (ALPPS) is a surgical strategy that aims to promote rapid growth of the future liver remnant (FLR)(\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e). Compared with conventional portal vein embolization (PVE), ALPPS can increase FLR by over 40% within 7\u0026ndash;14 days, offering a substantial advantage(\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e). However, its high rates of postoperative complications and mortality, particularly post-hepatectomy liver failure (PHLF), limit its widespread clinical adoption(\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e). In patients with hepatitis B virus (HBV) infection, liver fibrosis or cirrhosis significantly impairs hepatic regeneration following ALPPS(\u003cspan additionalcitationids=\"CR5\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e). Therefore, predicting PHLF risk after stage I of ALPPS holds clinical significance in this population. Current predictive models primarily rely on clinical indicators such as the Child\u0026ndash;Pugh score, MELD score, and FLR to the standard liver volume (SLV) ratio. However, they do not fully leverage the medical imaging information, leading to limited predictive accuracy (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eRecent advances in artificial intelligence have led to new approaches for preoperative assessment and risk prediction in liver surgery. Specifically, the integration of machine learning (ML) and radiomics has enabled high-throughput extraction and analysis of detailed quantitative imaging features from Computed Tomography (CT) or Magnetic Resonance Imaging (MRI) scans(\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e). This approach has been applied in multiple clinical contexts within hepatobiliary surgery. For example, combined radiomics\u0026ndash;clinical models have successfully predicted postoperative liver failure in patients with large hepatocellular carcinoma (\u0026gt;\u0026thinsp;10 cm), demonstrating strong performance(\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e). Similar to PVE, comprehensive models can accurately predict FLR growth, thereby informing optimal timing for surgery(\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e). These findings suggest that radiomics features reflect the functional status of liver tissue at cellular and molecular levels, offering novel insights into assessing regenerative potential.\u003c/p\u003e \u003cp\u003eThis study aims to develop a predictive model for PHLF after stage I ALPPS in patients with HBV, performing a radiomics analysis and machine learning on preoperative clinical, non-contrast CT (NCCT), and portal venous phase CT (PVCT) data. Additionally, the present study compares the efficiency between whole-FLR (wFLR) and partial-FLR (pFLR) to determine whether predictive accuracy can be maintained while reducing computational workload. Such a model could assist surgeons in identifying high-risk patients preoperatively and uncover potential imaging biomarkers associated with liver regeneration.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003ePatients\u003c/h2\u003e \u003cp\u003eThis retrospective study included patients with HBV infection who underwent the ALPPS Stage I in the First Affiliated Hospital of Guangxi Medical University between January 2018 and December 2024. This research was approved by the Ethics Committee of our institution (2025-E0957). Inclusion criteria were: (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e) contrast-enhanced CT was performed within 14 days preoperatively, with operative records and postoperative imaging confirming standard ALPPS Stage I; (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e) preoperative Child-Pugh class A or B liver function. Exclusion criteria were: (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e) preoperative imaging indicating unresectable lesions in FLR, main portal vein tumor thrombosis, or extrahepatic metastasis; (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e) postoperative imaging revealing ligation located at a secondary branch of the right portal vein; (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e) non-standard ALPPS variants (e.g., Tourniquet ALPPS, Haro-ALPPS, or other hybrid procedures).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eStandard ALPPS procedure\u003c/h3\u003e\n\u003cp\u003eThe ALPPS technique was performed in two stages (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e). In Stage I, the right portal vein is ligated, the liver parenchyma is partitioned in situ along the planned resection plane (typically along Cantlie\u0026rsquo;s line), and a cholecystectomy is conducted. In Stage II, right hepatectomy or extended right hepatectomy was completed to achieve R0 resection.\u003c/p\u003e\n\u003ch3\u003eDefinition of PHLF\u003c/h3\u003e\n\u003cp\u003ePHLF was defined according to the International Study Group of Liver Surgery (ISGLS) criteria: increased international normalized ratio (INR\u0026thinsp;\u0026gt;\u0026thinsp;1.5) and hyperbilirubinemia (serum total bilirubin\u0026thinsp;\u0026gt;\u0026thinsp;1.2 mg/dl) on or after postoperative day 5(\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e). Specifically, PHLF was graded as A, B, or C based on clinical management and severity. The primary outcome of this study was clinically significant PHLF (grades B and C).\u003c/p\u003e\n\u003ch3\u003eLiver Volume Measurement and Regeneration Model\u003c/h3\u003e\n\u003cp\u003eSLV was calculated according to the formula described by Urata et al.(\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e). FLR volume was measured based on radiological images. The preoperative estimated FLR was delineated using the actual in-situ partitioning plane confirmed by postoperative imaging registration. Postoperative FLR volumes were measured based on available imaging until the day before stage II or within 90 days after stage I.\u003c/p\u003e \u003cp\u003eTo assess the regeneration kinetics of FLR after Stage I, the relative volume was used as the primary metric. Liver regeneration was modeled first using linear mixed-effects (LME) models and log-transformed LME (log-LME) models fitted to longitudinal volume data collected at multiple time points. The model showing the lowest Akaike information criterion (AIC) was selected as the optimal model and used to calculate individual liver regeneration rates. All formulas are detailed in Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e.\u003c/p\u003e\n\u003ch3\u003eImage Segmentation and Radiomics Feature Extraction\u003c/h3\u003e\n\u003cp\u003eCT images were acquired from different scanner models, tube voltage, and slice thickness (Supplementary Table S2). To eliminate inter-scanner variability, all images underwent standardized preprocessing: resampling to 1\u0026times;1\u0026times;1 mm\u0026sup3; using B-spline interpolation, clipping of Hounsfield units to [-100, 400], and gray-level discretization with a fixed bin width of 25 HU.\u003c/p\u003e \u003cp\u003eTwo types of volumes of interest (VOIs) were delineated for model construction and comparison: whole-FLR (wFLR) and partial-FLR (pFLR)(Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The whole-FLR (wFLR) was first automatically segmented from preoperative portal venous phase CT (PVCT) images using TotalSegmentator v1.5.7 (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e). Subsequently, the resulting segmentation was refined manually by a surgeon and a radiologist to simulate the in-situ partition plane during surgery. The remaining liver after virtual resection served as the wFLR-VOIs. For the pFLR-VOIs, a 2-cm radius sphere was manually segmented at the portal venous level in segments S2/3. Large intrahepatic vessels and obvious non-hepatic tissues were excluded from all VOIs. Registration between non-contrast CT(NCCT) and PVCT images was performed to ensure spatial consistency of VOIs across sequences. All final VOIs were independently reviewed and corrected by two senior radiologists with more than 5 years of experience in liver imaging. All steps were performed using ITK-SNAP v4.2.0(\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e). Radiomics features were extracted using PyRadiomics v3.1.0 (Supplementary Table S3).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eRadiomics Feature Selection and Model Construction\u003c/h2\u003e \u003cp\u003eMultiple strategies were employed to ensure feature robustness and model stability due to the limitation of a relatively small sample size, including z-score normalization, intraclass correlation coefficients, and cross-validation(CV)(Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Stable features were selected through stability selection based on 10 runs of repeated 5-fold cross-validation. Within each training fold, features were removed using the independent-samples t-test, Pearson correlation analysis and least absolute shrinkage and selection operator (LASSO) regression. The selection frequency of each feature was recorded across all iterations, with features showing a selection frequency of \u0026ge;\u0026thinsp;50%. Therefore, these features were used to establish radiomics models.\u003c/p\u003e \u003cp\u003eFour supervised machine learning classifiers were tested: logistic regression (LR), random forest (RF), support vector machine (SVM) with RBF kernel, and extreme gradient boosting (XGBoost). To reduce the risk of overfitting, a nested stratified CV framework was implemented, comprising an outer 5-fold stratified CV for performance estimation and class balance, as well as an inner 3-fold stratified CV for hyperparameter tuning. The classifier with the highest mean area under the receiver operating characteristic (ROC) curve (AUC) in the outer loop and the most stable learning curve was selected as the final radiomics model and used to calculate radiomics score for each patient.\u003c/p\u003e \u003cp\u003eUnivariate and multivariable LR analyses were performed on clinical variables to construct the clinical model and generate clinical scores. A combined clinical-radiomics model was then developed by LR. The predictive performance of all models was assessed using AUC, with differences compared by the DeLong test. Calibration was assessed with calibration curves. Clinical utility was evaluated by decision curve analysis (DCA).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eInterpretability\u003c/h3\u003e\n\u003cp\u003eTo enhance the interpretability of the radiomics models, SHAP (SHapley Additive exPlanations) was used to assess the contribution of individual radiomics features to the model\u0026rsquo;s predictions. Additionally, Pearson correlation analysis was applied to explore the relationship between liver regeneration rate and radiomics features. Moreover, causal mediation analysis was performed to assess whether the liver regeneration rate mediates the relationship between preoperative radiomic scores and PHLF.\u003c/p\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Analysis\u003c/h2\u003e \u003cp\u003ePrediction models were developed using Python (version 3.9). All other statistical analyses were performed using R software (version 4.2.1). The \u0026lsquo;lattice\u0026rsquo; package was used to plot liver volume growth curves for individual patients, and the \u0026lsquo;nlme\u0026rsquo; package was used to construct liver growth models. A significance level of p\u0026thinsp;\u0026lt;\u0026thinsp;0.05 in the two-sided tests was considered statistically significant.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eClinical characteristics\u003c/h2\u003e \u003cp\u003ePHLF grade B/C was found in 22 patients (24.4%) (Table\u0026nbsp;1). Compared with the non-LF group, the LF group exhibited significantly higher MELD scores (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001), aspartate aminotransferase (P\u0026thinsp;=\u0026thinsp;0.049), and total bilirubin levels (P\u0026thinsp;=\u0026thinsp;0.036).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eLiver Volume and Regeneration Model\u003c/h2\u003e \u003cp\u003eThe preoperative mean SLV was 1206 cm\u0026sup3;, and the mean estimated FLR was 353 cm\u0026sup3;, resulting in a mean FLR/SLV ratio of 29% (Table\u0026nbsp;1, Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). After Stage I, the patients underwent a comprehensive evaluation, with 68 patients achieved sufficient FLR to proceed to Stage II. In this group, the median interval between two stages was 15.5 days (IQR: 13\u0026ndash;26 days), with the FLR volume exhibiting a mean percentage increase of 152% from baseline (IQR: 142%-170%). Nonetheless, 7 patients developed PHLF but ultimately proceeded to Stage II after treatment and liver functional recovery. However, 22 patients did not undergo Stage II surgery due to PHLF after treatment (n\u0026thinsp;=\u0026thinsp;15), tumor progression (n\u0026thinsp;=\u0026thinsp;5), and insufficient liver regeneration (n\u0026thinsp;=\u0026thinsp;1).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe dynamic changes in liver volume are displayed in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, revealing a characteristic pattern of rapid FLR growth in the early postoperative phase, followed by a deceleration toward a plateau. Subsequently, a liver regeneration model was established between the LF and non-LF group by using LME and Log-LME, with a particular focus on the early postoperative period (within 10 and 15 days). The results demonstrated that the non-LF group had a significantly higher daily FLR growth rate than the LF group throughout the full period and during the early phase (Table\u0026nbsp;2). Among the models tested, the 10-day early Log-LME exhibited the best goodness-of-fit and was used to calculate the individual liver regeneration rates.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eRadiomics Feature Extractor and Model\u003c/h2\u003e \u003cp\u003eDuring VOI segmentation, the average time required for wFLR-VOIs was 0.5\u0026ndash;1 hour with the assistance of the TotalSegmentator plugin, whereas the process only required 1 to 2 minutes for pFLR-VOIs. Following the stable feature selection process, 5, 10, 4, and 3 stable features significantly associated with LF were identified from wFLR-NCCT, wFLR-PVCT, pFLR-NCCT, and pFLR-PVCT, respectively (Supplementary Table\u0026nbsp;4). Pearson correlation analysis revealed that the correlation coefficients between wFLR and pFLR features were all below 0.8, indicating weak linear correlations (Supplementary Fig.\u0026nbsp;1).\u003c/p\u003e \u003cp\u003eFinally, based on the outer-layer average scores of nested stratified CV and learning curves, LR was selected as the optimal machine learning model for all four models (Supplementary Table\u0026nbsp;5, Supplementary Fig.\u0026nbsp;2). Hence, a radiomics score was generated. In comparison, the other machine learning models exhibited varying degrees of underfitting or overfitting.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eClinical Prediction Model\u003c/h2\u003e \u003cp\u003eUnivariate analysis revealed that preoperative ALT (P\u0026thinsp;=\u0026thinsp;0.044) and MELD score (P\u0026thinsp;=\u0026thinsp;0.003) were associated with the occurrence of PHLF. The variance inflation factor (VIF) for both variables was less than 5. Multivariate logistic regression analysis identified MELD score (P\u0026thinsp;=\u0026thinsp;0.006) as an independent predictor of PHLF, while ALT showed no statistical significance (Supplementary Table S6). A logistic regression (LR) clinical prediction model was constructed using the MELD score, and a clinical model score was calculated.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eComprehensive Prediction Model\u003c/h2\u003e \u003cp\u003eThrough nested CV, the AUCs of the clinical model, radiologic models, and the comprehensive model were 0.791, 0.892, and 0.894 for wVOIs, and 0.791, 0.769, and 0.828 for pVOIs, respectively (Table\u0026nbsp;3). Calibration curves and decision curve analysis (DCA) further confirmed the clinical usefulness of the comprehensive model (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Furthermore, DeLong\u0026rsquo;s test indicated no significant difference in predictive performance between the integrated models constructed using wVOIs and pVOIs segmentation (P\u0026thinsp;=\u0026thinsp;0.238). The results demonstrated that partial VOI segmentation achieves comparable efficacy to whole VOI segmentation while substantially reducing manual workload.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eInterpretability of Radiomics Features\u003c/h2\u003e \u003cp\u003eThe interpretability of the machine learning model was assessed using SHAP (SHapley Additive exPlanations) values (Supplementary Fig.\u0026nbsp;3\u0026ndash;6). Pearson correlation coefficients were calculated between radiomics features and individual regeneration rate, with p-values adjusted using the false discovery rate (FDR) method (Supplementary Table S7). Among the 22 PHLF-related features, 2 features were positively correlated with liver regeneration rate, and the remaining features exhibited negative correlations .\u003c/p\u003e \u003cp\u003eThe mediation analysis was performed using the wVOIs (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Radiomic scores derived from models based on wVOIs were confirmed as independent risk factors for PHLF (OR 1.171, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). The mediation analysis revealed that 86.2% of the total effect was mediated by the 10-day liver regeneration rate (OR 1.146, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001), whereas the direct effect of the radiomic scores was not significant (OR 1.022, P\u0026thinsp;=\u0026thinsp;0.853).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis retrospective study analyzed clinical data and preoperative CT images from HBV-infected patients undergoing ALPPS Stage I. Liver regeneration dynamics were quantified using Log-LME modeling, and a comprehensive predictive model was developed to predict the occurrence of PHLF. The comprehensive model demonstrated strong predictive performance in our cohorts, offering a novel quantitative tool for early risk assessment. Collectively, the correlation coefficients and mediation analysis highlighted the central role of liver regeneration in linking radiomic features to PHLF in ALPPS. These findings provide mechanistic insights into liver regeneration.\u003c/p\u003e \u003cp\u003ePredictive modeling in ALPPS has evolved from relying solely on traditional clinical parameters to the integration of multi-omics data. Early studies relied on preoperative liver function and FLR volume for risk assessment; however, this approach did not account for interindividual regenerative variability and failed to leverage the rich information in medical imaging(\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e). Advances in imaging have enabled the incorporation of CT- or MRI-based morphological evaluations to predict FLR growth, with machine learning enhancing model accuracy(\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e). Our clinical-radiomics model is based on radiomics and machine learning, but innovatively associates the FLR dynamic regeneration with radiomics features. Moreover, SHAP was incorporated for interpretable machine learning analysis, mitigating the black-box nature of traditional models to some extent and providing a more comprehensive perspective. Notably, no statistically significant differences were observed between partial and whole segmentation strategies. This may be attributed to the parenchymal nature of the liver or the diffuse pathology associated with HBV, although partial marking risks the loss of shape features.\u003c/p\u003e \u003cp\u003eMechanistically, ALPPS-driven rapid regeneration involves hemodynamic shifts, inflammatory cytokine release, and cell proliferation pathway activation(\u003cspan additionalcitationids=\"CR23\" citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e). Portal vein ligation markedly increases portal flow to the remnant liver, activating endothelial nitric oxide synthase (eNOS) via shear stress to drive hepatocyte cell cycle entry(\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e). Additionally, parenchymal transection releases IL-6 and TNF-α, further promoting proliferation(\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e). Previous studies have suggested that HBV infection may impair regeneration by inducing fibrosis and altering the microenvironment(\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e). Zhang et al. reported significantly higher 90-day mortality in severe fibrosis patients post-ALPPS (P\u0026thinsp;=\u0026thinsp;0.014), consistent with our finding that preoperative MELD score predicts stage-I PHLF risk(\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e). Furthermore, our study identified multiple radiomics features associated with ALPPS-related liver regeneration. Notably, mediation analysis indicated that the effect of radiomics feature on PHLF risk was primarily mediated by the 10-day liver regeneration rate. Therefore, radiomics may partially reflect early regenerative capacity, providing a novel perspective for liver regeneration.\u003c/p\u003e \u003cp\u003eRecent advances in conversion therapy, such as interventional hepatoma therapy, targeted therapy, and immunotherapy, have shifted focus away from ALPPS, but the latter still offers unique benefits in enhancing FLR regeneration and resectability. A multicenter study by Lv et al. reported a 5-year survival rate of 31.7% following ALPPS in patients with HBV-related cirrhosis, which was significantly higher than that observed with TACE (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). These results highlight the importance of patient selection, particularly for those with localized tumors and well-compensated liver function(\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e). Future directions include integrating multi-omics (e.g., transcriptomics, proteomics) to elucidate radiomic mechanisms, shifting from correlation to causation(\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e). Combining ALPPS with interventions, targeted therapy, or immunotherapy\u0026mdash;such as the AITI conversion regimen\u0026mdash;shows promising potential(\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e). With advancements in artificial intelligence (AI) technology, deep learning models based on time-series data hold potential to further enhance the accuracy of PHLF prediction(\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe innovative aspects of this study are summarized below. First, a nonlinear mixed-effects model was employed, overcoming the limitations of conventional static volume measurements and capturing key parameter changes during the regenerative process. Secondly, deep learning segmentation technology (TotalSegmentator) was integrated with standardized radiomics processes, thereby reducing subjective biases due to manual operations; in addition, the partial segmentation strategy further decreases the workload for labeling. Thirdly, the \u0026ldquo;black-box\u0026rdquo; nature of radiomics features was mitigated through multiple interpretative approaches. Nevertheless, the limitations of the present study should be acknowledged. Firstly, the single-center study and relatively small sample size remain a major limitation in this study. Although various methods, such as learning curves, nested CV, and four machine learning algorithms suitable for small datasets, have been employed to maintain model stability and generalization, larger samples are required for further validation and training. Moreover, the study participants involved primarily patients with HBV infection, so the applicability of the findings to patients with other etiologies should be carefully verified. In addition, preoperative FLR volume was estimated based on short-term postoperative CT registration, which may lead to measurement errors. Finally, although the model demonstrated good predictive performance and radiomics features were statistically associated with liver regeneration, the underlying biological mechanisms still require validation through further basic research.\u003c/p\u003e \u003cp\u003eIn conclusion, the clinical-radiomics model established in this study can effectively predict the risk of PHLF after ALPPS Stage I. The partial segmentation strategy significantly improves efficiency while maintaining the predictive performance, providing a valuable tool for clinical practice. Future prospective multicenter studies are warranted to validate the clinical applicability of this model and to explore optimized strategies combining ALPPS with systemic therapies to further improve patient outcomes.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cb\u003eDisclosure Statement:\u003c/p\u003e\n\u003cp\u003eThe authors declare no commercial or financial conflicts of interest related to this study. No financial or material support was received.\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eJixu Guo: statistical analysis and manuscript writing; Shiqin Pang, Deyang Huang, Shengjie Xie: data acquisition; Sichen Feng, Li Li: image segmentation; Shuiping Yu: study conception and design, critical revision, and supervision.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis retrospective study involving human participants was approved by the Medical Ethics Committee of The First Affiliated Hospital of Guangxi Medical University (Approval No. 2025-E0957). All procedures performed in this study were in accordance with the ethical standards of the institutional and/or national research committee and with the Declaration of Helsinki and its later amendments or comparable ethical standards. The requirement for informed consent was waived by the Medical Ethics Committee due to the retrospective nature of the study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eHuman Ethics declaration\u003c/strong\u003e\u003cbr\u003e\u0026nbsp;This study involved human participants and was approved by the Medical Ethics Committee of The First Affiliated Hospital of Guangxi Medical University (Approval No. 2025-E0957).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study received no funding. No competing interests exist.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of Interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no conflicts of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets are not publicly available to protect patient privacy, but they can be obtained from the corresponding author upon reasonable request via email.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe would like to thank the developers of ITK-SNAP for providing the open-source software for medical image segmentation, and the authors of the TotalSegmenter for facilitating automated liver segmentation. We thank Home for Researchers editorial team (www.home-for-researchers.com) for language editing service.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eSchnitzbauer AA, Lang SA, Goessmann H, Nadalin S, Baumgart J, Farkas SA, Fichtner-Feigl S, et al. Right portal vein ligation combined with in situ splitting induces rapid left lateral liver lobe hypertrophy enabling 2-staged extended right hepatic resection in small-for-size settings. Ann Surg. 2012;255:405\u0026ndash;14.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen H, Wang X, Zhu W, Li Y, Yu Z, Li H, Yang Y et al. Application of associating liver partition and portal vein ligation for staged hepatectomy for initially unresectable hepatocellular carcinoma. BMC Surg 2022;22.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang J, Xu Y, Yang H, Huang H, Bian J, Jiang S, Sang X, et al. Application of associating liver partition and portal vein ligation for staged hepatectomy for hepatocellular carcinoma related to hepatitis B virus: comparison with traditional one-stage right hepatectomy. Translational cancer Res. 2020;9:5371\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang J, Zhang L, Yang X, Zheng Y, Xu H, Du S, Mao Y, et al. Liver fibrosis as a predictor of liver failure and outcome following ALPPS among patients with primary liver cancer. Sci Rep. 2024;14:15827.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHung K-C, Wang H-P, Li W-F, Lin Y-C, Wang C-C. Single center experience with ALPPS and timing with stage 2 in patients with fibrotic/cirrhotic liver. Updates Surg. 2024;76:1213\u0026ndash;21.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eQin Y, Li C, Ge X, Zhang Q, Wei X, Liu R. MMP2/9 downregulation is responsible for hepatic function recovery in cirrhotic rats following associating liver partition and portal vein ligation for staged hepatectomy. Ann Transl Med. 2022;10:468.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChan A, Zhang WY, Chok K, Dai J, Ji R, Kwan C, Man N et al. ALPPS Versus Portal Vein Embolization for Hepatitis-related Hepatocellular Carcinoma: A Changing Paradigm in Modulation of Future Liver Remnant Before Major Hepatectomy. Annals of Surgery 2021;273.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu W, Liu X, Peng M, Chen GQ, Liu PH, Cui XW, Jiang F, et al. Artificial intelligence for hepatitis evaluation. World J Gastroenterol. 2021;27:5715\u0026ndash;26.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXiang F, Liang X, Yang L, Liu X, Yan S. CT radiomics nomogram for the preoperative prediction of severe post-hepatectomy liver failure in patients with huge ([greater than or equai to] 10 cm) hepatocellular carcinoma. World J Surg Oncol. 2021;19:NA\u0026ndash;NA.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGerwing M, Schindler P, Katou S, K\u0026ouml;hler M, Stamm AC, Schmidt VF, Heindel W, Gerwing, et al. Ann Surg Oncol. 2023;30:7976\u0026ndash;85.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang Q, Brismar TB, Bj\u0026ouml;rk D, Baubeta E, Lindell G, Bj\u0026ouml;rnsson B, Sparrelid E. Development and External Validation of a Combined Clinical\u0026ndash;Radiomic Model for Predicting Insufficient Hypertrophy of the Future Liver Remnant following Portal Vein Embolization. Ann Surg Oncol. 2025;32:1795\u0026ndash;807.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuidetti C, M\u0026uuml;ller PC, Magistri P, Jonas JP, Odorizzi R, Kron P, Guerrini G, et al. Full robotic versus open ALPPS: a bi-institutional comparison of perioperative outcomes. Surg Endosc. 2024;38:3448\u0026ndash;54.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCaringi S, Delvecchio A, Dezio M, Casella A, Ferraro V, Filippo R, Stasi M et al. Feasibility and safety of ALPPS procedure: our experience. Surg Endosc 2025:1\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRahbari NN, Garden OJ, Padbury R, Brooke-Smith M, Crawford M, Adam R, Koch M, et al. Posthepatectomy liver failure: a definition and grading by the International Study Group of Liver Surgery (ISGLS). Surgery. 2011;149:713\u0026ndash;24.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUrata K, Kawasaki S, Matsunami H, Hashikura Y, Ikegami T, Ishizone S, Momose Y, et al. Calculation of child and adult standard liver volume for liver transplantation. Hepatology. 1995;21:1317\u0026ndash;21.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWasserthal J, Breit H-C, Meyer MT, Pradella M, Hinck D, Sauter AW, Heye T et al. TotalSegmentator: robust segmentation of 104 anatomic structures in CT images. Radiology: Artificial Intelligence. 2023;5:e230024.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYushkevich PA, Piven J, Hazlett HC, Smith RG, Ho S, Gee JC, Gerig G. User-guided 3D active contour segmentation of anatomical structures: significantly improved efficiency and reliability. NeuroImage. 2006;31:1116\u0026ndash;28.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKhan AS, Garcia-Aroz S, Ansari MA, Atiq SM, Senter-Zapata M, Fowler K, Doyle M, et al. Assessment and optimization of liver volume before major hepatic resection: current guidelines and a narrative review. Int J Surg. 2018;52:74\u0026ndash;81.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLopez-Lopez V, Linecker M, Cruz J, Brusadin R, Lopez‐Conesa A, Machado MA, Hernandez‐Alejandro R, et al. Liver growth prediction in ALPPS\u0026ndash;A multicenter analysis from the international ALPPS registry. Liver Int. 2022;42:2815\u0026ndash;29.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThan VS, Le TD, Cao MT, Pham MT. Simultaneous Portal and Hepatic Vein Embolization versus Portal Vein Embolization Only in Patients with Hepatocellular Carcinoma: A Retrospective Review of Safety and Effectiveness. Journal of Vascular and Interventional Radiology; 2025.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang X, Qiu C, Du X, Qin J, Zhang Y, Hu Z, Luo Y et al. Fatty liver regeneration after partial hepatectomy: an experimental study based on intravoxel incoherent motion and T2* mapping MRI. Magnetic Resonance Materials in Physics, Biology and Medicine 2025:1\u0026ndash;11.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu H, Qiu X, Wang Z, Wang K, Tan Y, Gao F, Perini MV, et al. Role of the portal system in liver regeneration: From molecular mechanisms to clinical management. Liver Res. 2024;8:1\u0026ndash;10.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang S, Ma Y, Chen X, Wu S, Chen G. Circulating proliferative factors versus portal inflow redistribution: mechanistic insights of ALPPS-derived rapid liver regeneration. Front Oncol. 2025;14:1429564.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen X, Zhou B, Wu S, Li G, Ma Y, Chen P, Chen G. Impact of the extent and location of liver split on future liver remnant hypertrophy after portal vein ligation in a rat model. Surgery. 2024;175:1321\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMasuo H, Shimizu A, Motoyama H, Kubota K, Notake T, Yoshizawa T, Hosoda K, et al. Impact of endothelial nitric oxide synthase activation on accelerated liver regeneration in a rat ALPPS model. World J Gastroenterol. 2023;29:867.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTihanyi DK, Szijarto A, Fulop A, Jiang D, Ernst L, Meister FA, Bleilevens C, et al. Exploring the Enhanced Liver Regeneration Patterns Following ALPPS Versus Selective Portal Vein Ligation in an Experimental Model. Cancer Rep. 2025;8:e70221.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDu Y, Yang Y, Zhang Y, Zhang F, Wu J, Yin J. Unraveling enhanced liver regeneration in ALPPS: Integrating multi-omics profiling and in vivo CRISPR in mouse models. Hepatol Commun. 2025;9:e0630.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLv J-H, Chen W-Z, Li Y-N, Wang J-X, Fu Y-K, Zeng Z-X, Wu J-Y, et al. Should associating liver partition and portal vein ligation for staged hepatectomy be applied to hepatitis B virus-related hepatocellular carcinoma patients with cirrhosis? A multi-center study. HPB. 2022;24:2175\u0026ndash;84.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLefebvre ATN, Ghosh S, Baciu C, Hasjim BJ, Naimimohasses S, Oldani G, Pasini E et al. Modelling the Liver\u0026rsquo;s Regenerative Capacity across Different Clinical Conditions. JHEP Rep 2025:101465.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuan Y, Cui L, Inchai J, Fang Z, Law J, Brito AAG, Pawlosky A et al. AI-Assisted Drug Re‐Purposing for Human Liver Fibrosis. Adv Sci 2025:e08751.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen Z, Chen X, Hu H, Chen K, Xiao H, Du C, Lan X. The combination of associating liver partition and portal vein ligation for staged hepatectomy (ALPPS), interventional hepatoma therapy, targeted therapy, and immunotherapy: a case series of a novel AITI conversion therapy model. J Gastrointest Oncol. 2025;16:1736\u0026ndash;48.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang K, Yang Q, Li K, Tang S, Zhang B, Liao X, Du S et al. Learning-based early detection of post-hepatectomy liver failure using temporal perioperative data: a nationwide multicenter retrospective study in China. eClinicalMedicine 2025;83.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTable 1 to 4 are available in the Supplementary Files section.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"world-journal-of-surgical-oncology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"wjso","sideBox":"Learn more about [World Journal of Surgical Oncology](http://wjso.biomedcentral.com)","snPcode":"12957","submissionUrl":"https://submission.nature.com/new-submission/12957/3","title":"World Journal of Surgical Oncology","twitterHandle":"@OncoBioMed","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"ALPPS, liver regeneration, radiomics, prediction, hepatitis B","lastPublishedDoi":"10.21203/rs.3.rs-9143445/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9143445/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eAiming:\u003c/h2\u003e \u003cp\u003eThis study aimed to develop a clinical\u0026ndash;radiomics model based on preoperative dual-phase computed tomography to predict post-hepatectomy liver failure (PHLF) following stage I associating liver partition and portal vein ligation for staged hepatectomy (ALPPS) in patients with hepatitis B.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eThis retrospective study included 90 patients. Mixed-effects models were employed to assess the dynamic regeneration of the future liver remnant (FLR). Radiomics features were extracted using volumes of interest defined for both whole-FLR (wFLR) and partial-FLR (pFLR). Four machine learning algorithms, combined with nested cross-validation, were used to construct stable clinical-radiomics models. Interpretability analyses, as well as mediation analyses, were also performed.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eThe incidence of grade B or C PHLF was 24.4%. The liver generation model demonstrated a significantly lower daily growth rate in the PHLF group compared to the non-PHLF group (3.67% vs. 5.61% per day, P\u0026thinsp;=\u0026thinsp;0.003). The clinical, radiologic, and combined model based on wFLR achieved AUCs of 0.791, 0.892, and 0.894, respectively; those based on pFLR achieved AUCs of 0.791, 0.769, and 0.828. No significant difference in model performance was observed between the two segmentation strategies (P\u0026thinsp;=\u0026thinsp;0.858), though pFLR segmentation substantially reduced workload. A preliminary mediation analysis suggested that 86.2% of the radiomics score\u0026rsquo;s total effect on PHLF was mediated by liver regeneration rate (OR 1.146, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eThe proposed clinical\u0026ndash;radiomics models effectively predict PHLF after ALPPS Stage I in patients with hepatitis B. The effect of the radiomics score on PHLF is mediated by impaired liver regeneration.\u003c/p\u003e","manuscriptTitle":"Preoperative CT-Based Radiomics for Predicting Post-Hepatectomy Liver Failure and Assessing Liver Regeneration after ALPPS Stage I in Hepatitis B Patients","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-19 12:32:16","doi":"10.21203/rs.3.rs-9143445/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewerAgreed","content":"309901254319899145946679931718748731605","date":"2026-04-12T03:10:40+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-10T03:09:03+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-21T12:39:25+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-19T11:40:37+00:00","index":"","fulltext":""},{"type":"submitted","content":"World Journal of Surgical Oncology","date":"2026-03-17T03:14:55+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"world-journal-of-surgical-oncology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"wjso","sideBox":"Learn more about [World Journal of Surgical Oncology](http://wjso.biomedcentral.com)","snPcode":"12957","submissionUrl":"https://submission.nature.com/new-submission/12957/3","title":"World Journal of Surgical Oncology","twitterHandle":"@OncoBioMed","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"a959681f-89b1-4147-82d2-36ee96d16bc5","owner":[],"postedDate":"April 19th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-04-19T12:32:16+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-19 12:32:16","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9143445","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9143445","identity":"rs-9143445","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}