Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Front Immunol . 2026 Mar 27;17:1782008. doi: 10.3389/fimmu.2026.1782008 Search in PMC Search in PubMed View in NLM Catalog Add to search Radiomics and blood biomarkers for predicting efficacy of sintilimab plus lenvatinib in advanced hepatocellular carcinoma Yong Cui Yong Cui 1 Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education), Department of Radiology, Peking University Cancer Hospital & Institute, Beijing, China 2 Peking University Cancer Hospital (Inner Mongolia Campus)/Affiliated Cancer Hospital of Inner Mongolia Medical University, Department of Radiology, Inner Mongolia Cancer Center, Hohhot, China Data curation, Methodology, Funding acquisition, Writing – original draft Find articles by Yong Cui 1, 2, † , Lijun Wang Lijun Wang 3 Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education), Department of Hepatopancreatobiliary Surgery Unit I, Peking University Cancer Hospital & Institute, Beijing, China Resources, Writing – original draft, Methodology, Data curation Find articles by Lijun Wang 3, † , Xiaoting Li Xiaoting Li 1 Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education), Department of Radiology, Peking University Cancer Hospital & Institute, Beijing, China Methodology, Formal analysis, Writing – original draft Find articles by Xiaoting Li 1 , Kun Wang Kun Wang 3 Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education), Department of Hepatopancreatobiliary Surgery Unit I, Peking University Cancer Hospital & Institute, Beijing, China Supervision, Writing – original draft, Data curation, Resources Find articles by Kun Wang 3 , Hongwei Wang Hongwei Wang 3 Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education), Department of Hepatopancreatobiliary Surgery Unit I, Peking University Cancer Hospital & Institute, Beijing, China Resources, Data curation, Writing – original draft Find articles by Hongwei Wang 3 , Quan Bao Quan Bao 3 Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education), Department of Hepatopancreatobiliary Surgery Unit I, Peking University Cancer Hospital & Institute, Beijing, China Resources, Data curation, Writing – original draft Find articles by Quan Bao 3 , Haitao Zhu Haitao Zhu 1 Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education), Department of Radiology, Peking University Cancer Hospital & Institute, Beijing, China Software, Methodology, Writing – original draft, Data curation Find articles by Haitao Zhu 1 , Xiaolei Gu Xiaolei Gu 1 Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education), Department of Radiology, Peking University Cancer Hospital & Institute, Beijing, China Resources, Formal analysis, Software, Writing – original draft Find articles by Xiaolei Gu 1 , Qian Xing Qian Xing 1 Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education), Department of Radiology, Peking University Cancer Hospital & Institute, Beijing, China Software, Resources, Writing – original draft Find articles by Qian Xing 1 , Kemin Jin Kemin Jin 3 Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education), Department of Hepatopancreatobiliary Surgery Unit I, Peking University Cancer Hospital & Institute, Beijing, China Resources, Writing – original draft, Supervision, Data curation Find articles by Kemin Jin 3 , Yingshi Sun Yingshi Sun 1 Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education), Department of Radiology, Peking University Cancer Hospital & Institute, Beijing, China Writing – review & editing, Funding acquisition, Conceptualization, Project administration Find articles by Yingshi Sun 1, * , Baocai Xing Baocai Xing 3 Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education), Department of Hepatopancreatobiliary Surgery Unit I, Peking University Cancer Hospital & Institute, Beijing, China Conceptualization, Validation, Writing – review & editing, Supervision, Funding acquisition Find articles by Baocai Xing 3, * Author information Article notes Copyright and License information 1 Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education), Department of Radiology, Peking University Cancer Hospital & Institute, Beijing, China 2 Peking University Cancer Hospital (Inner Mongolia Campus)/Affiliated Cancer Hospital of Inner Mongolia Medical University, Department of Radiology, Inner Mongolia Cancer Center, Hohhot, China 3 Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education), Department of Hepatopancreatobiliary Surgery Unit I, Peking University Cancer Hospital & Institute, Beijing, China * Correspondence: Yingshi Sun, [email protected] ; Baocai Xing, [email protected] † These authors have contributed equally to this work Roles Yong Cui : Data curation, Methodology, Funding acquisition, Writing – original draft Lijun Wang : Resources, Writing – original draft, Methodology, Data curation Xiaoting Li : Methodology, Formal analysis, Writing – original draft Kun Wang : Supervision, Writing – original draft, Data curation, Resources Hongwei Wang : Resources, Data curation, Writing – original draft Quan Bao : Resources, Data curation, Writing – original draft Haitao Zhu : Software, Methodology, Writing – original draft, Data curation Xiaolei Gu : Resources, Formal analysis, Software, Writing – original draft Qian Xing : Software, Resources, Writing – original draft Kemin Jin : Resources, Writing – original draft, Supervision, Data curation Yingshi Sun : Writing – review & editing, Funding acquisition, Conceptualization, Project administration Baocai Xing : Conceptualization, Validation, Writing – review & editing, Supervision, Funding acquisition Received 2026 Jan 9; Accepted 2026 Mar 11; Revised 2026 Mar 7; Collection date 2026. Copyright © 2026 Cui, Wang, Li, Wang, Wang, Bao, Zhu, Gu, Xing, Jin, Sun and Xing. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY) . The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms. PMC Copyright notice PMCID: PMC13066827 PMID: 41972186 Abstract Background Immune checkpoint inhibitor–based combination therapy has emerged as an important treatment option for advanced hepatocellular carcinoma (HCC), yet therapeutic response remains highly heterogeneous. Biomarkers that jointly reflect tumor-intrinsic heterogeneity and host immune status are needed to improve response stratification and better understand variability in immunotherapy outcomes. Methods We evaluated the predictive value of Computed Tomography (CT) -derived radiomic heterogeneity and systemic inflammatory biomarkers in patients with advanced HCC treated with sintilimab plus lenvatinib. A total of 62 patients were included and divided into a training cohort (n = 36) and an independent real-world validation cohort (n = 26). Radiomic features were extracted from multiphase contrast-enhanced CT and summarized as a radiomics score. Hematological indices reflecting systemic immune-inflammatory status were assessed in parallel. An integrated model combining imaging-derived heterogeneity and immune-inflammatory markers was constructed for response stratification. Results The radiomics score discriminated responders from non-responders with area under the curve (AUC) values of 0.840 in the training cohort and 0.759 in the validation cohort. Among systemic biomarkers, the systemic immune-inflammation index (SII) was independently associated with treatment response. The integrated model combining radiomic heterogeneity and SII demonstrated improved discriminatory performance (AUC 0.938 and 0.819 in training and validation cohorts, respectively). Stratification based on the combined model was associated with differences in event-free survival, supporting the biological relevance of multimodal immune–tumor characterization. Conclusions Integrating imaging-derived tumor heterogeneity with systemic immune-inflammatory status may provide a biologically informed and non-invasive strategy for exploratory response stratification in advanced HCC undergoing immune checkpoint inhibitor–based therapy. Larger multicenter studies incorporating prospective validation and immune profiling are warranted to confirm clinical applicability. Keywords: CT imaging, hepatocellular carcinoma, lenvatinib, predictive model, radiomics, sintilimab, systemic immune-inflammation index Introduction Hepatocellular carcinoma remains one of the leading causes of cancer-related mortality worldwide and is frequently diagnosed at an advanced stage, precluding curative interventions ( 1 – 3 ). In recent years, immune checkpoint inhibitors targeting programmed cell death protein 1 (PD-1), particularly in combination with anti-angiogenic agents such as lenvatinib, have expanded therapeutic options for advanced HCC ( 4 ). Clinical trials and real-world studies have reported objective response rates of approximately 30–40%; however, durable benefit is observed only in a subset of patients, underscoring marked interindividual variability in immunotherapy response ( 5 , 6 ). From an immunological standpoint, response to immune checkpoint inhibitor–based therapy is shaped by the dynamic interplay between tumor-intrinsic characteristics and host immune context. Intratumoral heterogeneity, vascular architecture, and stromal composition influence immune cell infiltration and the establishment of an immunologically permissive or suppressive microenvironment. Concurrently, systemic immune and inflammatory status modulates antitumor immunity and may determine the capacity of immune checkpoint blockade to restore effective cytotoxic T-cell function ( 7 , 8 ). Despite increasing recognition of these dual determinants, clinically accessible biomarkers capable of capturing both tumor-local and host-systemic immune factors remain limited. Radiomics ( 9 ) has emerged as a non-invasive approach to quantify tumor phenotypic heterogeneity through high-dimensional feature extraction from medical imaging. In HCC, CT-derived radiomic features have been associated with tumor vascular patterns, hypoxia-related characteristics, and microenvironmental complexity—biological processes closely linked to immune responsiveness ( 10 – 13 ). Nevertheless, most radiomics-based predictive models focus primarily on tumor-intrinsic imaging signatures without incorporating host immune-related information, potentially limiting their biological interpretability in the context of immunotherapy. In parallel, hematological biomarkers reflecting systemic immune-inflammatory status have gained attention as accessible predictors of immunotherapy outcomes ( 14 – 16 ). Indices such as the SII ( 17 ), which integrates neutrophil, platelet, and lymphocyte counts, are considered surrogate markers of the balance between pro-tumor inflammation and antitumor immune surveillance. Elevated neutrophil and platelet levels may contribute to immunosuppressive signaling and tumor progression, whereas reduced lymphocyte counts may reflect impaired adaptive immune responses. However, when used alone, these systemic markers lack tumor-specific spatial context and may insufficiently characterize intratumoral immune heterogeneity. Given these complementary strengths and limitations, integrating imaging-derived tumor heterogeneity with systemic immune-inflammatory indicators may provide a more comprehensive framework for understanding variability in immunotherapy response. Conceptually, such an approach aims to approximate tumor–host immune interactions by combining spatially resolved tumor phenotypes with systemic immune context. Accordingly, this study sought to evaluate the predictive value of CT-based radiomics features and pre-treatment hematological biomarkers in patients with advanced HCC treated with sintilimab plus lenvatinib. By developing and validating an integrated multimodal model, we aimed to explore whether combining tumor-intrinsic imaging heterogeneity with systemic immune status could enhance exploratory response stratification in the immunotherapy setting. Materials and methods Study design and patient population This study employed a retrospective–prospective design to evaluate imaging-derived tumor heterogeneity and systemic immune-inflammatory biomarkers associated with treatment response in patients with advanced HCC receiving immune checkpoint inhibitor–based combination therapy. The training cohort consisted of patients enrolled in a prospective clinical trial conducted at Peking University Cancer Hospital (ClinicalTrials.gov identifier: NCT04042805 ) evaluating sintilimab combined with lenvatinib in locally advanced HCC. The independent validation cohort included consecutive patients treated at the same institution under routine clinical practice during or after the trial period who received the same therapeutic regimen. The sample size reflects the number of eligible patients treated during the study period, and all analyses were conducted as exploratory investigations. Inclusion criteria were: (1) Histologically or radiologically confirmed HCC; (2) Availability of baseline multiphase contrast-enhanced CT imaging; (3) Availability of pre-treatment hematological laboratory data;(4) Receipt of at least three cycles of sintilimab plus lenvatinib with evaluable response assessment. Exclusion criteria included:(1) Prior systemic therapy or locoregional treatment;(2) Non-standardized CT protocols unsuitable for radiomics analysis;(3) Missing imaging or laboratory data;(4) Loss to follow-up. All patients received sintilimab (200 mg every 3 weeks) combined with lenvatinib (12 mg daily for body weight ≥60 kg or 8 mg daily for <60 kg). The study was approved by the institutional ethics committee and conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants. CT acquisition and response assessment Baseline multiphase contrast-enhanced CT scans were performed prior to treatment initiation using multi-detector CT scanners (Philips and GE Healthcare platforms). Imaging protocols included non-contrast, arterial phase (20–30 seconds), portal venous phase (70–80 seconds), and delayed phase acquisitions. Treatment response was evaluated every 9 weeks during the first year and every 12 weeks thereafter, according to the modified Response Evaluation Criteria in Solid Tumors (mRECIST). Patients achieving complete response or partial response were classified as responders, while those with stable or progressive disease were classified as non-responders. Imaging indicator analysis Prior to analysis, all images were reviewed by two independent radiologists (XL Gu and Y Cui) to ensure technical adequacy. Imaging features were categorized into two groups: subjective and objective. Subjective imaging features included qualitative tumor characteristics assessed by the radiologists, such as the proximity of the tumor to the liver capsule, the presence of large tumor vessels within the lesion, arterial phase enhancement, visibility of the peripheral tumor outline, capsule integrity, regularity of tumor margins during the portal venous phase, homogeneity of enhancement in the portal venous phase, surrounding enhancement during the arterial phase, and evidence of necrosis, cystic degeneration, or intratumoral hemorrhage. Objective imaging features included radiomics features, which were quantitatively extracted using specialized software. Tumor regions of interest were manually delineated on baseline CT images using 3D Slicer software (version 4.9.0, https://www.slicer.org ). Segmentation was performed on non-contrast, arterial, and portal venous phase images by an experienced radiologist blinded to clinical outcomes. To assess intra-observer reproducibility, repeated segmentation was performed in a subset of patients after a one-month interval. Intraclass correlation coefficients (ICC) were calculated, and features with ICC < 0.75 were excluded from further analysis to ensure robustness. Radiomics features were extracted using the PyRadiomics package (version 3.1.0) implemented in Python. A total of 107 features were extracted for each CT phase, including 18 first-order features, 24 gray-level co-occurrence matrix (GLCM) features, 14 gray-level dependence matrix (GLDM) features, 16 gray-level run length matrix (GLRLM) features, 16 gray-level size zone matrix (GLSZM) features, 5 neighboring gray-tone difference matrix (NGTDM) features, and 14 shape features. In total, 321 features were extracted across the three CT phases. No ComBat harmonization was applied, which is acknowledged as a methodological limitation. Feature selection was conducted exclusively within the training cohort to avoid data leakage. Initially, features significantly associated with treatment response were identified using univariable analysis. Redundant features were subsequently removed based on correlation analysis to reduce multicollinearity. Following feature selection, three machine learning classifiers—support vector machine (SVM), extreme gradient boosting (XGBoost) and least absolute shrinkage and selection operator (LASSO) logistic regression—were used to construct predictive models. Each classifier generated a Radiomics Score, which estimated the probability of treatment response for individual patients. To mitigate class imbalance within the training set, the Synthetic Minority Over-sampling Technique (SMOTE) was applied during model training only. The independent validation cohort was evaluated without oversampling. Hematological indicator analysis Peripheral blood samples were collected within one week prior to treatment initiation. Hematological indices reflecting systemic immune and inflammatory status were calculated, including neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), lymphocyte-to-monocyte ratio (LMR), SII and systemic inflammation response index (SIRI). The NLR was calculated by dividing the neutrophil count by the lymphocyte count, while the PLR was calculated by dividing the platelet count by the lymphocyte count. The LMR was calculated by dividing the lymphocyte count by the monocyte count. The SII was computed by multiplying the platelet count and neutrophil count, then dividing by the lymphocyte count. Similarly, the SIRI was calculated by multiplying the neutrophil count and monocyte count, then dividing by the lymphocyte count. These markers were evaluated as potential indicators of host immune-inflammatory status relevant to immunotherapy response. The combination model construction, evaluation and validation Subjective imaging features, hematological indicators, and other potential clinical factors were assessed using the Chi-square test or Fisher’s exact test. Features with a significance level of less than 0.05, along with the radiomics score, were incorporated into a logistic regression model to construct the combination model using a machine learning approach. The diagnostic performance of the model was assessed using the receiver operating characteristic (ROC) curve. To evaluate the clinical relevance of the model, survival analysis was conducted on the model-predicted response groups using Kaplan–Meier curves and log-rank testing. Statistical analysis All analyses were exploratory, and no adjustment for multiple testing was performed given the hypothesis-generating nature of the study. A sample of 13 from the responder group and 13 from the non-responder group achieves 80% power to detect a difference of 0.25 between the area under the ROC curve under the null hypothesis of 0.7 and an AUC under the alternative hypothesis of 0.95 using a two-sided z-test at a significance level of 0.05. Thus at least 26 patients were needed for the validation group. Continuous variables with a normal distribution were presented as means with standard deviations, while categorical variables were represented as counts. Comparisons of continuous variables between response groups were made using either Student’s t-test or the Mann–Whitney U test as appropriate. Categorical variables were compared using the chi-square test or Fisher’s exact test. Kaplan–Meier analysis and log-rank testing were performed to explore event-free survival (EFS) differences between model-stratified groups. A two-sided P value < 0.05 was considered statistically significant. Statistical analyses were conducted using R4.4.2 (R Core Team, 2024) and SPSS 25.0 (SPSS, Chicago, IL) software. Results Patient characteristics Between August 1, 2019, and December 10, 2021, a total of 62 HCC patients were included, with 36 assigned to the training cohort and 26 to the independent validation cohort. As of the data cutoff date (September 30, 2024), 12 patients in the training set had undergone surgery with curative intent, and 1 patient had received a combination of radiofrequency ablation and stereotactic radiotherapy. In the validation set, 6 patients had undergone surgery. The remaining patients in both cohorts continued to receive the combination therapy until tumor progression, adverse reactions became intolerable, or they withdrew informed consent. Based on the mRECIST, 24 patients (66.7%) in the training set and 16 patients (61.5%) in the validation set were classified as responders. Table 1 presents the baseline characteristics of the patients. All variables were comparable between the training and validation cohorts, with the exception of three subjective radiomics features: surrounding enhancement in the arterial phase, necrosis, and cystic changes. Moderate to excellent ICC (0.481 to 1.000) was achieved for the CT variables, as detailed in Supplementary Table 1 . Table 1. Comparison of Baseline Characteristics Between the Training and Validation Cohorts. Variables Category Training cohort Validation cohort P’ Responder (n=24) Non-responder (n=12) P Responder (n=16) Non-responder (n=10) P General clinical data: Gender Female 1 4 0.034 2 2 0.625 >0.999 Male 23 8 14 8 Age (years) 58 ± 12 60 ± 10 0.685 61 ± 11 63 ± 11 0.642 0.325 ECOG, PS 0 17 7 0.479 11 9 0.352 0.821 1 7 5 5 1 BCLC Stage B 11 6 0.813 3 5 0.189 0.193 C 13 6 13 5 Etiology of HCC HBV 22 11 1.000 15 10 1.000 1.000 HCV/Other 2 1 1 0 Largest tumor diameter (mm) 87 ± 36 101 ± 49 0.338 78 ± 48 64 ± 33 0.42 0.088 Number of tumors Single 16 7 0.72 6 4 >0.999 0.048 Multiple 8 5 10 6 AFP (ng/ml) <400 15 6 0.473 7 7 0.248 0.725 ≥400 9 9 9 3 Subjective radiomics features of tumors: Near the liver capsule No 3 0 >0.999 3 1 >0.999 0.439 Yes 21 12 13 9 Large tumor vessels inside the tumor No 11 5 0.813 8 4 0.701 0.894 yes 13 7 8 6 Arterial phase enhancement No 1 0 >0.999 2 1 >0.999 0.296 Yes 23 12 14 9 Peripheral outline Unclear 3 0 0.536 2 1 >0.999 0.689 Clear 21 12 14 9 Capsule integrity Complete 22 11 >0.999 16 10 NA 0.258 Incomplete 2 1 0 0 Irregular protrusions at the margin in the portal venous phase No 16 2 0.005 9 3 0.191 0.829 Yes 8 10 7 7 Uneven enhancement in the portal venous phase No 3 1 >0.999 3 2 >0.999 0.473 Yes 21 11 13 8 Surrounding enhancement in the arterial phase No 23 11 >0.999 12 7 >0.999 0.028 Yes 1 1 4 3 Necrosis and cystic changes No 9 6 0.473 11 8 0.668 0.014 Yes 15 6 5 2 Intratumoral hemorrhage No 24 10 0.105 16 9 0.385 >0.999 Yes 0 2 0 1 Hematological indicators: NLR (Neutrophil-to-lymphocyte ratio) <5 24 11 0.333 13 6 0.369 0.007 ≥5 0 1 3 4 PLR (Platelet-to-lymphocyte ratio) <200 22 9 0.307 12 8 >0.999 0.502 ≥200 2 3 4 2 LMR (Lymphocyte-to-monocyte ratio) <340 8 3 0.715 7 6 0.42 0.121 ≥340 16 9 9 4 SII (Systemic Immune-Inflammation Index) <340 16 3 0.018 8 2 0.126 0.549 ≥340 8 9 8 8 SIRI (Systemic Inflammation Response Index) <1 16 6 0.471 10 3 0.107 0.384 ≥1 8 6 6 7 Open in a new tab P, comparison between responder and non-responder groups; P’, comparison between training and validation groups; BCLC, Barcelona Clinic Liver Cancer; ECOG PS, Eastern Cooperative Oncology Group Performance Status; HBV, hepatitis B virus; HCC, hepatocellular carcinoma; HCV, hepatitis C virus. Radiomics model performance Initially, a radiomics model based solely on objective imaging features was developed to predict treatment efficacy. Specifically, three CT-based radiomics models were constructed using the LASSO, SVM, and XGBoost algorithms. The diagnostic performance of these models in both the training and validation cohorts is presented in Table 2 . Among the evaluated classification approaches, the LASSO-based radiomics model demonstrated stable performance and was selected for further analysis. Performance in the validation cohort was evaluated without oversampling. The radiomics workflow is illustrated in Figure 1 . Table 2. Diagnostic performance of three radiomics models in the training and validation cohorts. Methods Training Validation LASSO 0.840 (95%CI,0.701 to 0.980) 0.750 (95%CI,0.544 to 0.956) VM 0.854 (95%CI,0.7109 to 0.989) 0.600 (95%CI,0.373 to 0.827) XGBoost 0.898 (95%CI,0.797 to 0.998) 0.672 (95%CI,0.458 to 0.887) Open in a new tab Figure 1. Open in a new tab Radiomics workflow for model development and validation (A) Imaging Segmentation: Multiphase contrast-enhanced CT images (plain, arterial, and portal venous phases) were acquired, and regions of interest (ROIs) were manually delineated. (B) Feature Extraction: A total of 107 radiomic features, including shape, first-order, and texture features (GLCM, GLRLM, GLSZM, GLDM, NGTDM), were extracted from both original and wavelet-transformed images using PyRadiomics. (C) Feature Selection: The LASSO regression with cross-validation was employed to select the most predictive features. (D) Model Evaluation: ROC curves assessed discrimination performance, while calibration curves evaluated model accuracy in both training and validation cohorts. Associations of hematological biomarkers with treatment response Among the evaluated hematological markers, the SII was significantly associated with treatment response (P = 0.018). Patients with elevated SII were more frequently classified as non-responders. Other inflammatory indices did not show statistically significant associations. The combination model construction, evaluation, and validation To enhance predictive accuracy, we integrated the radiomics score with significant clinical and hematological variables into a combination model. A multivariable logistic regression analysis using a stepwise selection method identified the radiomics score (OR = 4408.23, P = 0.028) and the SII-based classification (OR = 10.30, P = 0.048) as independent predictors of treatment non-response. Additionally, the presence of tumor margin protrusions during the portal venous phase showed a trend toward statistical significance (OR = 5.94, P = 0.089). The combined model achieved AUCs of 0.938 in the training cohort and 0.819 in the validation cohort, demonstrating improved discrimination compared with single-modality models as detailed in Table 3 and Figure 2 . Calibration analysis indicated good agreement between predicted and observed probabilities. Decision curve analysis suggested potential clinical net benefit across a range of threshold probabilities. Table 3. Diagnostic performance of the radiomics score, subjective CT imaging features, hematological biomarkers, and the combined model in the training and validation cohorts. Model/Feature Training Validation Model/Feature AUC Cutoff Sen Spe Acu AUC Sen Spe Acu Radiomics score 0.840 >0.37 0.92 0.67 0.78 0.759 0.4 1 0.77 Irregular protrusions at the tumor edge 0.750 Yes 0.83 0.63 0.69 0.650 0.80 0.56 0.65 SII classification 0.708 >344 0.75 0.63 0.67 0.667 0.70 0.63 0.65 Combination model 0.938 >0.45 0.92 0.96 0.94 0.819 0.8 0.75 0.77 Open in a new tab AUC, area under curve; Sen, sensitivity; Spe, specificity; Acu, accuracy; SII, Systemic immune-inflammation index. Figure 2. Open in a new tab Combination model development and validation (A) Nomogram Construction: A model combining the radiomics score (RS) and SII was developed to predict treatment non-response. (B) ROC Analysis: The model achieved an AUC of 0.938 in the training cohort and maintained good performance in the validation cohort with an AUC of 0.819. (C, D) Calibration Curves: Calibration curves showed strong agreement between predicted and actual probabilities, with mean absolute errors of 0.04 and 0.067. (E) Decision Curve Analysis: The model demonstrated superior clinical net benefit across threshold probabilities. Exploratory survival analysis The median EFS was 17 months (range: 2–61 months) in the training cohort, and 18 months (range: 2–51 months) in the validation cohort. Patients were categorized into model-predicted non-responder and responder groups based on the cutoff value determined by the combination model. In the training cohort, 24 patients were assigned to the model-predicted non-responder group, and 12 patients were assigned to the responder group. In the validation cohort, 14 patients were classified as non-responders and 12 patients as responders. The model-predicted non-responder group exhibited significantly inferior EFS compared to the responder group in both cohorts, with P-values of 0.045 in the training cohort and 0.044 in the validation cohort, as shown in Figure 3 . These findings support the clinical relevance of multimodal response stratification, although the survival analysis was exploratory. Figure 3. Open in a new tab Kaplan–meier survival analysis based on the combination model patients were classified into responder and non-responder groups based on the model, confirming its prognostic value. (A) Training Cohort: Non-responders had significantly worse event-free survival (EFS) compared to responders (P = 0.045). (B) Validation Cohort: The non-responder group had inferior EFS relative to responders (P = 0.044). Discussion This study suggests that tumor-intrinsic imaging heterogeneity and systemic immune-inflammatory status jointly contribute to variability in response to immune checkpoint inhibitor–based therapy in advanced HCC. The principal finding is that tumor imaging heterogeneity and systemic immune-inflammatory status jointly contribute to treatment response stratification. Specifically, a CT-derived radiomics score and the SII independently correlated with therapeutic response, and their integration improved predictive performance compared with either modality alone. Although radiomics is fundamentally an imaging-based technique, increasing evidence suggests that quantitative texture features may indirectly reflect tumor immune microenvironment characteristics. Prior studies ( 10 – 13 ) have linked radiomic heterogeneity patterns to immune cell infiltration, stromal composition, vascular remodeling, and hypoxic gradients—all of which influence responsiveness to immune checkpoint blockade. In this context, the radiomics score derived in our study may capture spatial variations in tumor architecture that parallel immune-related biological heterogeneity. Interestingly, irregular protrusions at the tumor margin in the portal venous phase did not reach statistical significance, although a trend toward association with poor response was observed. Such morphological irregularities have been previously associated with microvascular invasion (MVI) ( 18 ) and aggressive tumor behavior ( 19 , 20 ). The lack of statistical significance in our cohort may be attributable to limited sample size or insufficient power. Nevertheless, this imaging phenotype warrants further evaluation in larger studies to clarify its potential predictive value. Systemic immune-inflammatory status is increasingly recognized as a determinant of immunotherapy outcomes ( 21 – 23 ). The SII integrates neutrophil, platelet, and lymphocyte counts and reflects the dynamic balance between pro-tumor inflammatory signaling and anti-tumor adaptive immunity ( 17 , 24 ). Elevated neutrophils and platelets may promote tumor progression through cytokine secretion, angiogenic factor release, and suppression of cytotoxic lymphocyte function, whereas lymphocyte depletion reflects impaired immune surveillance. Therefore, elevated SII may represent a systemic immune milieu unfavorable for effective immune checkpoint blockade, which is consistent with the poorer response observed in our cohort. Importantly, these hematological indices serve as indirect surrogates of immune dynamics rather than direct measurements of the tumor immune microenvironment. Importantly, tumor-intrinsic heterogeneity and host systemic immune status represent two biologically interconnected dimensions of cancer progression. While radiomics-derived features may reflect localized immune–tumor interactions within the tumor microenvironment, hematological biomarkers such as SII capture systemic immune dynamics. The integration of these dimensions may provide a more comprehensive approximation of tumor–host immune interplay, which is particularly relevant in the context of immune checkpoint inhibitor–based therapies. From a clinical standpoint, early identification of patients unlikely to respond to immune checkpoint inhibitor–based therapy remains a critical challenge. A non-invasive model integrating imaging-derived heterogeneity and systemic immune markers may support risk stratification and closer monitoring. However, the present findings remain exploratory and hypothesis-generating, and prospective validation is required before implementation in treatment decision-making. Several limitations should be acknowledged. First, the sample size was modest, particularly for non-responder cases, which increases the risk of model instability and overfitting. Although an independent validation cohort was included, both cohorts originated from the same institution, and therefore the validation represents a real-world split-sample validation rather than true external validation. Second, the radiomics feature selection strategy relied on univariate filtering and correlation-based reduction, which may introduce selection bias in small datasets. The absence of nested cross-validation or feature stability assessment further limits robustness. Third, synthetic oversampling was applied to address class imbalance. While this approach may improve classifier discrimination, it may also alter feature distributions and affect generalizability. Finally, the survival analysis based on model-predicted response groups was exploratory, as the model was trained to predict radiologic response rather than survival endpoints. Taken together, our findings suggest a dual-layer immune–tumor interaction framework, in which intratumoral structural heterogeneity and systemic inflammatory balance jointly modulate sensitivity to immune checkpoint inhibition. Conclusion In conclusion, this study suggests that integrating CT-derived radiomic heterogeneity with systemic immune-inflammatory status may provide a biologically informed, hypothesis-generating framework for predicting response to immune checkpoint inhibitor–based combination therapy in advanced HCC. These findings support the concept that both tumor-intrinsic architecture and host immune dynamics contribute to therapeutic sensitivity. Future multicenter and prospective studies incorporating standardized imaging and immunological profiling are warranted to validate and refine this multimodal predictive strategy. Acknowledgments We thank the patients and their families who took part in this study. Funding Statement The author(s) declared that financial support was received for this work and/or its publication. This study was supported by grants from the Clinical Research Fund for Distinguished Young Scholars of Beijing Cancer Hospital (2019088), Natural Science Foundation of Inner Mongolia Autonomous Region of China (Grant No. 2024MS08005); Science and Technology Program of the Joint Fund of Scientific Research for the Public Hospitals of Inner Mongolia Academy of Medical Sciences (Grant No. 2023GLLH0122); and Fund of Scientific Research of Public Hospital Reform and High Quality Development Demonstration Program (Digestive Tract Tumors) of Peking University Cancer Hospital (Inner Mongolia Campus) (Grant No. 2023SGGZ075). Footnotes Edited by: Shui Liu , The Second Hospital of Jilin University Department of Hepatobiliary and Pancreatic Surgery, China Reviewed by: Nanqing Liao , Guangxi University, China Yangyang Xie , Hangzhou Hospital of Traditional Chinese Medicine, China Data availability statement The original contributions presented in the study are included in the article/ Supplementary Material , further inquiries can be directed to the corresponding author/s. Ethics statement The study was approved by the Ethics Committee of Peking University Cancer Hospital (Beijing, China). The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study. Author contributions YC: Data curation, Methodology, Funding acquisition, Writing – original draft. LW: Resources, Writing – original draft, Methodology, Data curation. XL: Methodology, Formal analysis, Writing – original draft. KW: Supervision, Writing – original draft, Data curation, Resources. HW: Resources, Data curation, Writing – original draft. QB: Resources, Data curation, Writing – original draft. HZ: Software, Methodology, Writing – original draft, Data curation. XG: Resources, Formal analysis, Software, Writing – original draft. QX: Software, Resources, Writing – original draft. KJ: Resources, Writing – original draft, Supervision, Data curation. YS: Writing – review & editing, Funding acquisition, Conceptualization, Project administration. BX: Conceptualization, Validation, Writing – review & editing, Supervision, Funding acquisition. Conflict of interest The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Generative AI statement The author(s) declared that generative AI was not used in the creation of this manuscript. Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us. Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. Supplementary material The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fimmu.2026.1782008/full#supplementary-material Table1.docx (21.7KB, docx) References 1. Rumgay H, Arnold M, Ferlay J, Lesi O, Cabasag CJ, Vignat J, et al. Global burden of primary liver cancer in 2020 and predictions to 2040. J Hepatol. (2022) 77:1598–606. doi: 10.1016/j.jhep.2022.08.021, PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 2. Forner A, Reig M, Bruix J. Hepatocellular carcinoma. Lancet. (2018) 391:1301–14. doi: 10.1016/S0140-6736(18)30010-2, PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 3. Park JW, Chen M, Colombo M, Roberts LR, Schwartz M, Chen PJ, et al. Global patterns of hepatocellular carcinoma management from diagnosis to death: the BRIDGE Study. Liver Int. (2015) 35:2155–66. doi: 10.1111/liv.12818, PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 4. Ringelhan M, Pfister D, O’Connor T, Pikarsky E, Heikenwalder M. The immunology of hepatocellular carcinoma. Nat Immunol. (2018) 19:222–32. doi: 10.1038/s41590-018-0044-z, PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 5. Llovet JM, Kudo M, Merle P, Meyer T, Qin S, Ikeda M, et al. Lenvatinib plus pembrolizumab versus lenvatinib plus placebo for advanced hepatocellular carcinoma (LEAP-002): a randomised, double-blind, phase 3 trial. Lancet Oncol. (2023) 24:1399–410. doi: 10.1016/S1470-2045(23)00469-2, PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 6. Wang L, Wang H, Cui Y, Liu M, Jin K, Xu D, et al. Sintilimab plus Lenvatinib conversion therapy for intermediate/locally advanced hepatocellular carcinoma: A phase 2 study. Front Oncol. (2023) 13:1115109. doi: 10.3389/fonc.2023.1115109, PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 7. Morita M, Nishida N, Sakai K, Aoki T, Chishina H, Takita M, et al. Immunological microenvironment predicts the survival of the patients with hepatocellular carcinoma treated with anti-PD-1 antibody. Liver Cancer. (2021) 10:380–93. doi: 10.1159/000516899, PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 8. Dai S, Jia R, Zhang X, Fang Q, Huang L. The PD-1/PD-Ls pathway and autoimmune diseases. Cell Immunol. (2014) 290:72–9. doi: 10.1016/j.cellimm.2014.05.006, PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 9. Lambin P, Rios-Velazquez E, Leijenaar R, Carvalho S, van Stiphout RG, Granton P, et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer. (2012) 48:441–6. doi: 10.1016/j.ejca.2011.11.036, PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 10. Losic B, Craig AJ, Villacorta-Martin C, Martins-Filho SN, Akers N, Chen X, et al. Intratumoral heterogeneity and clonal evolution in liver cancer. Nat Commun. (2020) 11:291. doi: 10.1038/s41467-019-14050-z, PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 11. Hectors SJ, Lewis S, Besa C, King MJ, Said D, Putra J, et al. MRI radiomics features predict immuno-oncological characteristics of hepatocellular carcinoma. Eur Radiol. (2020) 30:3759–69. doi: 10.1007/s00330-020-06675-2, PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 12. Xu X, Zhang HL, Liu QP, Sun SW, Zhang J, Zhu FP, et al. Radiomic analysis of contrast-enhanced CT predicts microvascular invasion and outcome in hepatocellular carcinoma. J Hepatol. (2019) 70:1133–44. doi: 10.1016/j.jhep.2019.02.023, PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 13. Yoon JH, Kim H. CT radiomics in oncology: insights into the tumor microenvironment of hepatocellular carcinoma. Radiology. (2023) 307:e222988. doi: 10.1148/radiol.222988, PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 14. Greten TF, Villanueva A, Korangy F, Ruf B, Yarchoan M, Ma L, et al. Biomarkers for immunotherapy of hepatocellular carcinoma. Nat Rev Clin Oncol. (2023) 20:780–98. doi: 10.1038/s41571-023-00816-4, PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 15. He CB, Lin XJ. Inflammation scores predict the survival of patients with hepatocellular carcinoma who were treated with transarterial chemoembolization and recombinant human type-5 adenovirus H101. PloS One. (2017) 12:e0174769. doi: 10.1371/journal.pone.0174769, PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Liu S, Xu W, Shu H, Dai Y, Du Y, Liu Y, et al. Associations of circulating immunomarkers with the efficacy of immunotherapy for primary hepatic carcinoma. Cancer Med. (2023) 12:21830–48. doi: 10.1002/cam4.6754, PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 17. Wang ZY, Xu B, Wang LN, Zhu XD, Huang C, Shen YH, et al. Platelet-to-lymphocyte ratio predicts tumor response and survival of patients with hepatocellular carcinoma undergoing immunotherapies. Int Immunopharmacol. (2024) 131:111863. doi: 10.1016/j.intimp.2024.111863, PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 18. Rodriguez-Peralvarez M, Luong TV, Andreana L, Meyer T, Dhillon AP, Burroughs AK. A systematic review of microvascular invasion in hepatocellular carcinoma: diagnostic and prognostic variability. Ann Surg Oncol. (2013) 20:325–39. doi: 10.1245/s10434-012-2513-1, PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 19. Bakr S, Echegaray S, Shah R, Kamaya A, Louie J, Napel S, et al. Noninvasive radiomics signature based on quantitative analysis of computed tomography images as a surrogate for microvascular invasion in hepatocellular carcinoma: a pilot study. J Med Imaging (Bellingham). (2017) 4:041303. doi: 10.1117/1.JMI.4.4.041303, PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 20. Ishihara N, Komatsu S, Sofue K, Ueshima E, Yano Y, Fujishima Y, et al. Association between tumor morphology and efficacy of atezolizumab plus bevacizumab for advanced hepatocellular carcinoma. Hepatol Res. (2024) 54:773–80. doi: 10.1111/hepr.14024, PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 21. Mei J, Sun XQ, Lin WP, Li SH, Lu LH, Zou JW, et al. Comparison of the prognostic value of inflammation-based scores in patients with hepatocellular carcinoma after anti-PD-1 therapy. J Inflammation Res. (2021) 14:3879–90. doi: 10.2147/JIR.S325600, PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Dirim MG, Cifcibasi Ormeci A, Cavus B, Poyanli A, Serin KR, Akyuz F, et al. Inflammation and immune-based scores predict prognosis for patients with hepatocellular carcinoma: A pilot single-center study. Turk J Gastroenterol. (2025) 36:219–28. doi: 10.5152/tjg.2025.24407, PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 23. Pinto TAM, Saito HPA, Nourani CL, Ataide EC, Boin I, Lourenco GJ, et al. Clinicopathological aspects and inflammation-immune markers in alcohol and/or hepatitis C virus-induced hepatocellular carcinoma patients treated with sorafenib. Gastroenterol Res. (2024) 17:23–31. doi: 10.14740/gr1689, PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 24. Wang RH, Wen WX, Jiang ZP, Du ZP, Ma ZH, Lu AL, et al. The clinical value of neutrophil-to-lymphocyte ratio (NLR), systemic immune-inflammation index (SII), platelet-to-lymphocyte ratio (PLR) and systemic inflammation response index (SIRI) for predicting the occurrence and severity of pneumonia in patients with intracerebral hemorrhage. Front Immunol. (2023) 14:1115031. doi: 10.3389/fimmu.2023.1115031, PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Table1.docx (21.7KB, docx) Data Availability Statement The original contributions presented in the study are included in the article/ Supplementary Material , further inquiries can be directed to the corresponding author/s. Articles from Frontiers in Immunology are provided here courtesy of Frontiers Media SA ACTIONS View on publisher site PDF (1.9 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top