MRI-based clinical-radiomics-habitat model for predicting prognosis of hepatocellular carcinoma patients treated with HAIC

Background: Hepatocellular carcinoma (HCC) is a highly heterogeneous malignant tumor with generally poor prognosis. Hepatic arterial infusion chemotherapy (HAIC) serves as a crucial treatment modality for intermediate to advanced HCC, but its efficacy is significantly influenced by tumor heterogeneity and individual variability. This study aimed to develop an MRI-based Clinical-Radiomics-Habitat model for non-invasive prediction of early response to HAIC treatment.

Methods: 105 HCC patients who received HAIC treatment across two institutions were retrospectively analyzed. Tumor subregions were segmented on four preoperative MRI sequences, including T1-weighted imaging (T1WI) and contrast-enhanced T1WI (arterial late phase, portal venous phase, and delayed phase), from which image features were extracted. Clinical data, habitat analysis features, and radiomics features were collected to construct three distinct predictive models. Each model was internally validated using the bootstrap method, evaluated using multiple performance metrics, and employed to explore prognostic information.

Results: Among the 105 patients, treatment responses included complete response (n=9), partial response (n=48), stable disease (n=34), and progressive disease (n=14), yielding responder and non-responder rates of 54.3% and 45.7%, respectively. With an AUC of 0.771 (95% CI 0.682–0.860), the Clinical–Radiomics–Habitat model performed better than both the Clinical model (0.633) and the Clinical–Radiomics model (0.747). Habitat imaging exhibits significant potential in analyzing tumor heterogeneity and predicting treatment early response in HCC patients.

Conclusion: We established a multiparametric MRI-based Clinical-Radiomics-Habitat model for preoperative early response prediction in HAIC-treated HCC patients. This model may assist clinicians in optimizing personalized treatment decisions.

1 Introduction

Hepatocellular carcinoma (HCC) is a highly lethal but common malignancy, and its incidence continues to rise globally (1, 2). HCC not only reduces the patients’ life expectancy and life quality, but also places a heavy financial burden on them owing to the high cost of treatment. Most patients with HCC are not candidates for curative resection at the time of diagnosis because the tumor often presents asymptomatically (3); consequently, therapy options including radiotherapy, chemotherapy, and molecular targeted therapy should be considered more comprehensively for the management of such patients (4).

By direct administration of chemotherapeutic agents into the hepatic artery at high concentrations, hepatic arterial infusion chemotherapy (HAIC) is considered effective in tumor reduction, vascular thrombus control, and improvement in patient survival with a high treatment response rate (5), and has become an important therapy for HCC at an intermediate-to-advanced stage (6, 7). However, as responses to HAICs vary among individual patients due to tumor heterogeneity and tumor treatment-related factors, identifying optimal treatment strategies and using objective criteria for estimating the treatment response are considered key clinical challenges.

In recent years, habitat imaging analysis has gained attention as a novel tumor-assessment method. By partitioning tumors into biologically meaningful subregions, this technique quantifies intratumoral composition and spatial interactions and has been applied across various malignancies (8, 9). Wu et al. (10) demonstrated that a Computed Tomography (CT) -based habitat radiomics model outperformed the model based solely on clinical variables. However, few studies have explored habitat analysis for predicting outcomes following HAIC in HCC. Moreover, Magnetic Resonance Imaging (MRI) provides superior sensitivity for HCC detection and characterization compared with CT, avoids ionizing radiation, and is widely used in pretreatment evaluation (2, 11). Therefore, this study aimed to extract features from whole tumors and their internal habitats on MRI to predict HAIC treatment response in patients with HCC.

2 Materials and methods

2.1 Patients

Our study received approval from the Ethics Committee of Ruijin Hospital, Shanghai Jiao Tong University School of Medicine. We retrospectively reviewed all HCC patients who received HAIC from January 2021 to December 2024 at Ruijin Hospital and Ruijin Hospital North.

Patients were included if (1) HCC diagnosed according to the American Association for the Study of Liver Diseases (AASLD) guidelines; (2) age ≥18 years; (3) at least one measurable unresectable lesion; (4) ECOG performance status of 0–1; (5) Child–Pugh class A or B; (6) available contrast-enhanced MRI within 1 month before HAIC and within 1 month after the final treatment cycle. Exclusion criteria included (1) receipt of additional treatments during follow-up; (2) severe cardiac, pulmonary, or renal insufficiency; (3) history of other malignancies; (4) poor-quality imaging or incomplete clinical data.

We included 105 patients who met the eligibility criteria. A flowchart of our patient selection process is provided in Figure 1A.

Figure 1

Figure 1. (A) Flowchart of patient enrollment and selection. (B) Workflow for predicting early response to HAIC in HCC.

2.2 HAIC therapy procedure

All patients underwent HAIC using a standardized FOLFOX regimen, repeated every 3 weeks. Under local anesthesia, femoral arterial access was obtained, and a conventional catheter with a microcatheter was selectively advanced into the tumor-feeding hepatic artery. The FOLFOX regimen consisted of oxaliplatin (85 mg/m²) administered via arterial infusion on day 1, followed by leucovorin (400 mg/m²) and fluorouracil (5-FU, 400 mg/m²) delivered intra-arterially, and continuous arterial infusion of 5-FU (2400 mg/m²) over 46 hours (days 2–3). No embolic agents were used.

2.3 Follow-up and response evaluation

Follow-up was conducted until the last recorded date (July 18, 2025). Treatment response was evaluated according to the modified Response Evaluation Criteria in Solid Tumors (mRECIST) criteria (12) based on MRI obtained within 1 month after completion of HAIC. Responses were categorized as complete response (CR), partial response (PR), stable disease (SD), or progressive disease (PD). Patients were subsequently grouped as having objective response (OR = CR + PR) or no objective response (nOR = SD + PD). Two radiologists specializing in abdominal interventional radiology (8 and 3 years of experience) independently evaluated all cases while blinded to clinical data; disagreements were resolved through consensus with a highly experienced radiologist (25 years of experience).

2.4 MRI acquisition

The workflow of our study is showed in Figure 1B. Four T1-weighted imaging (T1WI) sequences were included: pre-contrast (PRE), arterial late phase (LAP), portal venous phase (PVP), and delayed phase (DP). MRI scanners consisted of the uMR 790 3.0T, Philips Ingenia 3.0T, and GE SIGNA Architect AIR 3.0T. Detailed imaging parameters are provided in the Supplementary Material S1. Raw DICOM files were obtained from the Picture Archiving and Communication System (PACS) system for analysis.

2.5 Image segmentation and preprocessing

PVP images obtained prior to HAIC were imported into a software named 3D Slicer (version 5.2.2; www.slicer.org), and tumors were manually segmented by two radiologists (3 and 7 years of experience). Regions of Interest (ROIs) were delineated slice-by-slice.

To minimize variability across scanners and sequences, all images underwent N4 bias-field correction and were resampled to 1×1×1 mm. Registration was then conducted using ANTs (https://antsx.github.io/ANTs/), followed by min–max intensity normalization. All image processing was performed using Python (version 3.8.16; https://www.python.org). Following inter-sequence registration, all MRI sequences were spatially aligned; therefore, the tumor ROIs delineated on the portal venous phase were automatically mapped and correspondingly synchronized to the same anatomical locations on the remaining sequences.

2.6 Habitat subregion partitioning

Habitat subregion partitioning was performed using the open-source Python-based HABIT toolkit (https://github.com/lichao312214129/HABIT.git). For each tumor, unsupervised K-means clustering was applied to the voxel-wise intensity features within the ROI. The optimal number of clusters was then determined using the Elbow method by computing a value called the inertia that represents the sum of squared distances from each sample to its nearest cluster center. The inertia was calculated for k-means with k ranging from 2 to 10; we identified k as the point where the reduction of inertia became steady in the inertia–k curve. In our analyses, we selected k = 3. (Figure 2F). The intratumoral heterogeneity of HCC, as characterized by distinct habitat subregions, is shown in Figures 2A–E.

Figure 2

Figure 2. Habitat subregion segmentation across multiphase MRI and clustering validation. (A) Pre-contrast. (B) Arterial phase. (C) Portal venous phase. (D) Delayed phase. (E) 3D view of the ROI divided into three habitat subregions. (F) Determination of the optimal number of clusters using the Elbow method.

2.7 Feature extraction

Three types of features were considered: clinical features, radiomic features, and habitat features. Clinical features included demographic information, laboratory test data, and imaging features of tumors and were screened by univariate logistic regression analyses followed by multivariable logistic regression analyses. Radiomic features were extracted from whole-tumor ROIs whereas habitat features were extracted from tumor subregions, using PyRadiomics (https://pypi.org/project/pyradiomics/).

The stability of the extracted features was tested in 30 randomly selected patients by performing three repetitive extractions to calculate its coefficient of variation (CV), which was less than 15%. Features showing CV < 15% were considered stable. To further assess the inter-observer reproducibility of feature extraction, intraclass correlation coefficient (ICC) was used, and radiomic features with ICC < 0.8 were excluded. After normalization process was completed, highly correlated features (|r| > 0.9) were considered redundant and removed. Feature selection was done using recursive feature elimination (RFE) process followed by LASSO regression analysis (13) with 10-fold cross-validation technique after which feature selection was done based on iterative model building approach.

2.8 Model construction

The three types of features were incorporated sequentially to build three models: a clinical model, a clinical-radiomics (CR) model, and a clinical-radiomics-habitat (CRH) model with corresponding risk scores. We internally validated the models using 1,000 times bootstrap resampling. Given the small sample size in our study, bootstrap resampling was implemented to make sure that the evaluation of the models was robust (14, 15).

To assure feature independence and complementary information, we performed Spearman correlation and variance inflation factor (VIF) analyses with the combined features of the CR and CRH models. Severe redundancy indicated by |r|>0.8, and significant multicollinearity was indicated by VIF >10. Here, |r| is the absolute value of the Spearman correlation coefficient measuring the linear association between variables and VIF reflects how much a parameter estimate is increased due to multicollinearity.

2.9 Statistical analysis

Python (version 3.10.19; https://www.python.org) was used for all statistical analyses, with key functions from the scikit-learn, lifelines, matplotlib, and scipy libraries. For continuous variables, we reported those with a normal distribution as mean ± standard deviation (SD), while non-normally distributed ones were summarized as median (interquartile range). Categorical variables were presented as counts and percentages. For inter-group comparisons, independent t-test or Mann–Whitney U test according to distributions was used for continuous variables, and chi-squared test was used for categorical variables.

We assessed model discrimination with receiver operating characteristic (ROC) curves, area under the curve (AUC), and confusion matrices. Calibration and reliability were checked using calibration plots, and decision curve analysis was used to explore clinical utility. When comparing AUCs across models, we applied DeLong’s test. A two-sided p-value < 0.05 was considered statistically significant.

3 Results

3.1 Patient characteristics

Of 218 HCC patients who received HAIC, 47 were excluded due to concomitant malignancies, 21 for poor image quality or missing clinical data, and 45 for incomplete follow-up (Figure 1). Ultimately, 105 patients were included (mean age 57.3 ± 11.7 years; 97 male [92.4%], 8 female [7.6%]). According to treatment response, 9 achieved CR, 48 PR, 34 SD, and 14 PD. Overall, 57 patients (54.3%) were classified as OR, and 48 (45.7%) as nOR. Baseline characteristics are summarized in Table 1. Compared with the nOR group, the OR group showed notable differences in vascular invasion and tumor diameter; other baseline variables were similar (Table 2). Here, vascular invasion was defined as macrovascular invasion, specifically referring to the involvement of the portal vein and hepatic vein.

Table 1

Table 1. Patients’ characteristics.

Table 2

Table 2. Baseline characteristics of effective and ineffective groups in hepatocellular carcinoma patients treated with HAIC.

3.2 Radiomics and habitat feature extraction

A total of 428 radiomics features and 1,140 habitat features were initially extracted per patient. For each sequence, 93 features were extracted from both the whole-tumor ROI and each habitat subregion, including 18 first-order features, 24 GLCM features, 14 GLDM features, 16 GLRLM features, 16 GLSZM features, and 5 NGTDM features. Additionally, 14 shape features were calculated for each ROI, and habitat volume and proportion were recorded. Details of the feature extraction process are provided in Supplementary Material S2. After feature selection, six radiomics features (DP_Original_firstorder_10Percentile, DP_Original_glrlm_ShortRunEmphasis, DP_Original_firstorder_90Percentile, DP_Original_firstorder_Maximum, LAP_Original_firstorder_10Percentile, PRE_Original_firstorder_Maximum) and six habitat features (DP_Habitat3_firstorder_Kurtosis, PVP_Habitat3_firstorder_RootMeanSquared, PVP_Habitat1_firstorder_Minimum, PRE_Habitat1_firstorder_Range, PRE_Habitat2_firstorder_Skewness, DP_Habitat2_firstorder_Minimum) were retained as the most predictive (Figure 3).

Figure 3

Figure 3. Feature selection of radiomics and habitat features. (A, E) Bubble plots showing feature importance for radiomics (A) and habitat (E) features. (B, F) LASSO coefficient paths for radiomics (B) and habitat (F) features. (C, G) Cross-validation error curves for radiomics (C) and habitat (G) feature selection. (D, H) Spearman correlation heatmaps for radiomics (D) and habitat (H) features.

3.3 Construction of predictive models

Univariate and multivariate logistic regression identified Vascular Invasion (p=0.015) and d_tumor (p=0.037) as independent clinical predictors (Table 3). These two factors were used to construct the clinical model, while the CR model was built by combining clinical features with the selected radiomics features. The CRH model was further developed by incorporating habitat analysis features. Risk scores were calculated using binary logistic regression as follows:

Table 3

Table 3. Univariate and multivariate logistic regression analyses of the risk factors for treatment response to HAIC in patients with hepatocellular carcinoma (HCC).

$$\begin{array}{c}Clinical\text{ Model score}=0.6212+0.4103\hfill \\ \times \text{ Vascular Invasion }-0.0652\times \text{ d}\_\text{tumor}\hfill \end{array}$$

$$\begin{array}{c}CR\text{ Model score}\hfill \\ =0.7242+0.3415\times \text{ Vascular Invasion }-0.0694\times \text{ d}\_\text{tumor }\hfill \\ +0.1967\times \text{ DP}\_\text{Original}\_\text{original}\_\text{firstorder}\_10\text{Percentile }\hfill \\ -0.2450\times \text{ DP}\_\text{Original}\_\text{original}\_\text{glrlm}\_\text{ShortRunEmphasis }\hfill \\ +0.1234\times \text{ DP}\_\text{Original}\_\text{original}\_\text{firstorder}\_90\text{Percentile }\hfill \\ +0.2326\times \text{ DP}\_\text{Original}\_\text{original}\_\text{firstorder}\_\text{Maximum }\hfill \\ +0.1769\times \text{ LAP}\_\text{Original}\_\text{original}\_\text{firstorder}\_10\text{Percentile }\hfill \\ -0.2804\times \text{PRE}\_\text{Original}\_\text{original}\_\text{firstorder}\_\text{Maximum}\hfill \end{array}$$

$$\begin{array}{c}CRH \text{Model score}\hfill \\ =0.3814+0.1261\times \text{ Vascular Invasion }-0.0255\times \text{ d}\_\text{tumor }\hfill \\ +0.0953\times \text{ DP}\_\text{Original}\_\text{original}\_\text{firstorder}\_10\text{Percentile }\hfill \\ -0.1305\times \text{ DP}\_\text{Original}\_\text{original}\_\text{glrlm}\_\text{ShortRunEmphasis }\hfill \\ +0.0726\times \text{ DP}\_\text{Original}\_\text{original}\_\text{firstorder}\_90\text{Percentile }\hfill \\ +0.1217\times \text{ DP}\_\text{Original}\_\text{original}\_\text{firstorder}\_\text{Maximum }\hfill \\ +0.0904\times \text{ LAP}\_\text{Original}\_\text{original}\_\text{firstorder}\_10\text{Percentile }\hfill \\ -0.0658\times \text{ PRE}\_\text{Original}\_\text{original}\_\text{firstorder}\_\text{Maximum }\hfill \\ -0.1812\times \text{ DP}\_\text{Habitat}\_3\_\text{original}\_\text{firstorder}\_\text{Kurtosis }\hfill \\ +0.1756\hfill \end{array}$$

$$\begin{array}{c}\times \text{ PVP}\_\text{Habitat}\_3\_\text{original}\_\text{firstorder}\_\text{RootMeanSquared}\hfill \\ +0.1000\times \text{ PVP}\_\text{Habitat}\_1\_\text{original}\_\text{firstorder}\_\text{Minimum }\hfill \\ -0.0866\times \text{ PRE}\_\text{Habitat}\_1\_\text{original}\_\text{firstorder}\_\text{Range }\hfill \\ -0.0824\times \text{ PRE}\_\text{Habitat}\_2\_\text{original}\_\text{firstorder}\_\text{Skewness }\hfill \\ +0.0523\times \text{ DP}\_\text{Habitat}\_2\_\text{original}\_\text{firstorder}\_\text{Minimum}\hfill \end{array}$$

No significant redundancy or multicollinearity was observed (CR: 8 features, mean VIF = 2.445, max=4.930; CRH: 14 features, mean VIF = 2.802, max=5.990), supporting feature independence and model stability (Figure 4).

Figure 4

Figure 4. Heatmaps showing feature correlations in the two combined models. (A) CR model. (B) CRH model.

3.4 Model evaluation

ROC analysis showed that the clinical, CR, and CRH models achieved AUCs of 0.633 (95% CI: 0.525–0.742), 0.747 (95% CI: 0.653–0.841), and 0.771 (95% CI: 0.682–0.860), respectively. DeLong tests confirmed that the CR model showed significantly better performance than the clinical model (Z=–1.972, p=0.05), and the CRH model performed even better (Z=–2.386, p=0.02). Although the difference between the CR and CRH models did not reach statistical significance (Z = –1.071, p = 0.28), bootstrap validation indicated stable model performance (SD of AUC: CR = 0.048, CRH = 0.046). By incorporating six additional habitat features, the CRH model enhanced the representation of tumor microenvironment heterogeneity, increasing feature dimensionality and providing more biologically meaningful prognostic information (Figure 5).

Figure 5

Figure 5. Performance evaluation of the three models. (A) ROC curves of each model. (B) Calibration curves. (C) Decision curves. (D) AUC Comparison Bar-Scatter Plot.

We validated the performance of the CRH model in classifying patients. Figure 6A depicts a confusion matrix with satisfactory overall accuracy. The difference in predicted risk scores between OR and nOR groups is statistically significant (p<0.001), with a median risk score of 0.626 in the OR group compared with that of 0.486 in the nOR group (Figure 6B). This indicated that our model can distinguish between patients with different clinical outcomes. Figure 6C shows the Sankey diagram analysis, which revealed good agreement between our model’s prediction of early response and patients’ actual outcomes.

Figure 6

Figure 6. Performance evaluation of the CRH model for predicting the therapeutic response of HAIC. (A) Confusion matrix. (B) Comparison of model scores between OR and nOR groups. (C) Sankey diagram showing correspondence between predicted and actual treatment responses.

4 Discussion

HCC is known for significant intratumoral heterogeneity and a complex tumor microenvironment, which contribute to the marked interpatient variability in response to HAIC (16, 17). Therefore, preoperative individualized treatment planning and prediction of response are crucial. To date, existing prognostic models have included laboratory-based markers (eg, AFP, DCP, and VEGF) (18–22), imaging parameters (e.g., ADC, intratumoral fat content) (23–25), and genomic features (26, 27). However, these models have been imprecise and limited in their predictive ability. Traditional radiomics-based models have shown superior performance over prior approaches (7, 28), but such analyses generally assess the entire tumor and fail to capture spatial heterogeneity within lesions. Our model includes habitat features combined with traditional radiomics features; as a result, AUC increased without impairing the stability of the model. The enriched feature set provides a more flexible and biologically informed basis for individualized prediction, supporting pre-treatment clinical decision-making.

Prior radiomics studies have frequently investigated wavelet and Gaussian filtering features (7, 29). These filtered features indicate latent image information extracted after being transformed by the original image using certain type of transformation. In this study, we only used the original image characteristics. Given our small sample size, including filtered features would have greatly increased feature dimensionality and increased overfitting risk while reducing generalizability. Furthermore, these high-dimensional abstract features often lack clear biological or clinical interpretability, and substantial redundancy exists across different filters and habitats (30–32). Compared with filtered features, habitat features derived from original images can effectively capture intratumoral heterogeneity. Meanwhile, the CT-based habitat model developed by Wu et al. (10) used arterial phase and venous phase enhancement map (enhancement maps are generated by subtracting the non-contrast image from a contrast-enhanced arterial/venous phase image). This model emphasized intratumoral perfusion characteristics and yielded biologically interpretable subregions, providing valuable insights for related research. However, it may also overlook global tumor information.

Our study used multiparametric MRI for feature extraction. This is another approach suitable for establishing HAIC response prediction models. MRI provides much richer information than CT without irradiating patients with ionizing radiation and is now recognized as an international standard method of evaluating HCC prior to treatment (2, 11). The diversity of MRI sequences enables more detailed and multidimensional characterization of tumor subregions. Although no statistically significant performance differences were observed between the CRH and CR models, the CRH model exhibited a numerically higher AUC and showed a trend toward distinguishing patients with differing actual early response outcomes, which hints at its potential clinical utility. Moreover, as suggested by several previous studies (33), peritumoral regions represent a promising new target for habitat analysis in HCC. Diversified habitat segmentation and feature extraction remain important avenues for future research in HAIC response prediction.

Several of the selected habitat features were derived from first-order statistics across different contrast-enhanced MRI phases, particularly portal venous and delayed phases, which may reflect intratumoral heterogeneity in perfusion and tissue composition. From an imaging and biological perspective, these habitat subregions may correspond to areas with variable vascular supply, necrosis, or fibrosis, leading to heterogeneous drug delivery and treatment sensitivity. This is particularly relevant for HAIC, a locoregional therapy that strongly depends on arterial blood flow and microvascular integrity, where spatial heterogeneity within the tumor may directly influence therapeutic efficacy. In addition, the present study focuses on early treatment response rather than long-term survival outcomes, as the prognosis of patients undergoing HAIC is frequently influenced by subsequent treatments, including surgery and systemic therapies; therefore, early response was chosen as the primary study endpoint to reduce potential confounding effects from post-HAIC interventions.

Our study has a few limitations worth noting. First, it’s based on retrospective data and includes only a modest number of patients, which may influence feature selection and model stability and necessitate a large-scale multi-center prospective validation. Second, only multi-phase contrast-enhanced T1WI images are used in our study. Adding other sequences, such as T2WI, may further improve model performance. Third, because the imaging data were acquired with a variety of scanners and protocols, although we have conducted preprocessing to mitigate inter-scanner variability, more standardized image acquisition and processing protocols would be helpful.

In conclusion, we developed an MRI-based Clinical–Radiomics–Habitat model for non-invasive pre-treatment prediction of early response in HCC patients undergoing HAIC treatment, which may assist clinicians in optimizing personalized therapeutic strategies.

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 authors.

Ethics statement

The studies involving humans were approved by Ethics Committee of Ruijin Hospital, Shanghai Jiao Tong University School of Medicine. The studies were conducted in accordance with the local legislation and institutional requirements. The ethics committee/institutional review board waived the requirement of written informed consent for participation from the participants or the participants’ legal guardians/next of kin because this is a retrospective study that exclusively used anonymized medical images and clinical data.

Author contributions

JC: Writing – original draft, Data curation, Investigation, Visualization, Conceptualization, Formal analysis, Methodology. TZ: Visualization, Writing – original draft, Investigation, Data curation, Formal analysis, Conceptualization, Methodology. QS: Formal analysis, Visualization, Data curation, Investigation, Writing – original draft. YL: Writing – original draft, Methodology, Investigation, Visualization. HY: Writing – original draft, Investigation. XG: Writing – original draft, Investigation. QL: Methodology, Writing – original draft. XD: Investigation, Data curation, Writing – original draft. JW: Data curation, Validation, Methodology, Investigation, Supervision, Writing – review & editing. ZW: Validation, Project administration, Methodology, Supervision, Data curation, Conceptualization, Funding acquisition, Investigation, Writing – review & editing. YZ: Data curation, Writing – original draft, Validation, Project administration, Conceptualization, Supervision, Writing – review & editing, Funding acquisition, Methodology, Investigation.

Funding

The author(s) declared that financial support was received for this work and/or its publication. This study was funded by the National Natural Science Foundation of China (62173223, 62472275), the Young Scientists Fund of the National Natural Science Foundation of China (82202973), and the Shanghai 2023 Innovative Medical Device Application Demonstration Project (23SHS03900).

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/fonc.2026.1764150/full#supplementary-material

References

1. Li K, Mathew B, Saldanha E, Ghosh P, Krainer AR, Dasarathy S, et al. New insights into biomarkers and risk stratification to predict hepatocellular cancer. Mol Med. (2025) 31:152. doi: 10.1186/s10020-025-01194-6, PMID: 40269686

PubMed Abstract | Crossref Full Text | Google Scholar

2. Vogel A, Chan SL, Dawson LA, Kelley RK, Llovet JM, Meyer T, et al. Hepatocellular carcinoma: ESMO Clinical Practice Guideline for diagnosis, treatment and follow-up☆. Ann Oncol. (2025) 36:491–506. doi: 10.1016/j.annonc.2025.02.006, PMID: 39986353

PubMed Abstract | Crossref Full Text | Google Scholar

3. Roayaie S, Jibara G, Tabrizian P, Park JW, Yang J, Yan L, et al. The role of hepatic resection in the treatment of hepatocellular cancer. Hepatology. (2015) 62:440. doi: 10.1002/hep.27745, PMID: 25678263

PubMed Abstract | Crossref Full Text | Google Scholar

4. Reig M, Forner A, Rimola J, Ferrer-Fàbrega J, Burrel M, Garcia-Criado Á, et al. BCLC strategy for prognosis prediction and treatment recommendation: The 2022 update. J Hepatology. (2022) 76:681–93. doi: 10.1016/j.jhep.2021.11.018, PMID: 34801630

PubMed Abstract | Crossref Full Text | Google Scholar

5. Ikeda M, Morizane C, Ueno M, Okusaka T, Ishii H, and Furuse J. Chemotherapy for hepatocellular carcinoma: current status and future perspectives. Japanese J Clin Oncol. (2018) 48:103–14. doi: 10.1093/jjco/hyx180, PMID: 29253194

PubMed Abstract | Crossref Full Text | Google Scholar

6. Li QJ, He MK, Chen HW, Fang WQ, Zhou YM, Xu L, et al. Hepatic arterial infusion of oxaliplatin, fluorouracil, and leucovorin versus transarterial chemoembolization for large hepatocellular carcinoma: A randomized phase III trial. J Clin Oncol. (2022) 40:150–60. doi: 10.1200/JCO.21.00608, PMID: 34648352

PubMed Abstract | Crossref Full Text | Google Scholar

7. Xu Z, An C, Shi F, Ren H, Li Y, Chen S, et al. Automatic prediction of hepatic arterial infusion chemotherapy response in advanced hepatocellular carcinoma with deep learning radiomic nomogram. Eur Radiol. (2023) 33:9038–51. doi: 10.1007/s00330-023-09953-x, PMID: 37498380

PubMed Abstract | Crossref Full Text | Google Scholar

8. Li J, Qiu Z, Zhang C, Chen S, Wang M, Meng Q, et al. ITHscore: comprehensive quantification of intra-tumor heterogeneity in NSCLC by multi-scale radiomic features. Eur Radiol. (2023) 33:893–903. doi: 10.1007/s00330-022-09055-0, PMID: 36001124

PubMed Abstract | Crossref Full Text | Google Scholar

9. Chen W, Hu J, Du M, Zhang T, Gu C, Wu Q, et al. Habitat radiomics and deep learning on gadoxetic acid-enhanced MRI for noninvasive assessment of CK19 expression and recurrence-free survival in hepatocellular carcinoma. Front Oncol. (2025) 15:1684264/full. doi: 10.3389/fonc.2025.1684264/full, PMID: 41293267

PubMed Abstract | Crossref Full Text | Google Scholar

10. Wu M, Que Z, Lai S, Li G, Long J, He Y, et al. Predicting the early therapeutic response to hepatic artery infusion chemotherapy in patients with unresectable HCC using a contrast-enhanced computed tomography-based habitat radiomics model: a multi-center retrospective study. Cell Oncol. (2025) 48:709–723. doi: 10.1007/s13402-025-01041-0, PMID: 39903419

PubMed Abstract | Crossref Full Text | Google Scholar

11. Singal AG, Llovet JM, Yarchoan M, Mehta N, Heimbach JK, Dawson LA, et al. AASLD Practice Guidance on prevention, diagnosis, and treatment of hepatocellular carcinoma. Hepatology. (2023) 78:1922–65. doi: 10.1097/HEP.0000000000000466, PMID: 37199193

PubMed Abstract | Crossref Full Text | Google Scholar

12. Lencioni R and Llovet JM. Modified RECIST (mRECIST) assessment for hepatocellular carcinoma. Semin Liver Dis. (2010) 30:052–60. doi: 10.1055/s-0030-1247132, PMID: 20175033

PubMed Abstract | Crossref Full Text | Google Scholar

13. Song J, Yin Y, Wang H, Chang Z, Liu Z, and Cui L. A review of original articles published in the emerging field of radiomics. Eur J Radiol. (2020) 127. Available online at: https://www.ejradiology.com/article/S0720-048X(20)30180-7/fulltext., PMID: 32334372

PubMed Abstract | Google Scholar

14. Kostanek J, Karolczak K, Kuliczkowski W, and Watala C. Bootstrap method as a tool for analyzing data with atypical distributions deviating from parametric assumptions: critique and effectiveness evaluation. Data. (2024) 9:95. doi: 10.3390/data9080095

Crossref Full Text | Google Scholar

15. Feng M, Zhao M, Duan X, Hou J, Wang Q, and Yin F. Visualized clinical–radiomics model for predicting the efficacy of surufatinib in hepatic metastases of neuroendocrine neoplasms. Front Oncol. (2025) 15:1628054/full. doi: 10.3389/fonc.2025.1628054/full, PMID: 40951355

PubMed Abstract | Crossref Full Text | Google Scholar

16. Yang X, Yang C, Zhang S, Geng H, Zhu AX, Bernards R, et al. Precision treatment in advanced hepatocellular carcinoma. Cancer Cell. (2024) 42:180–97. doi: 10.1016/j.ccell.2024.01.007, PMID: 38350421

PubMed Abstract | Crossref Full Text | Google Scholar

17. Peng Z, Wang Y, Chen B, Zhu Z, He C, Wei Y, et al. Clinical efficacy of hepatic arterial infusion chemotherapy combined with transhepatic arterial embolization plus lenvatinib and tislelizumab or transarterial chemoembolization combined with lenvatinib plus tislelizumab in the treatment of advanced hepatocellular carcinoma. Front Oncol. (2025) 15:1665348/full. doi: 10.3389/fonc.2025.1665348/full

Crossref Full Text | Google Scholar

18. Xiao Y, Chen W, Deng W, Zhu G, Xie J, Luo L, et al. Prognostic value of alpha-fetoprotein in unresectable hepatocellular carcinoma treated with hepatic artery infusion chemotherapy combined with lenvatinib and camrelizumab. JHC. (2024) 11:1251–63. doi: 10.2147/JHC.S460922, PMID: 38979083

PubMed Abstract | Crossref Full Text | Google Scholar

19. An C, Wei R, Yao W, Han W, Li W, Shi G, et al. Association of serum AFP trajectories and hepatocellular carcinoma outcomes after hepatic arterial infusion chemotherapy: A longitudinal, multicenter study. Cancer Med. (2024) 13:e7319. doi: 10.1002/cam4.7319, PMID: 38819606

PubMed Abstract | Crossref Full Text | Google Scholar

20. Yamamoto S, Onishi H, Takaki A, Oyama A, Adachi T, Wada N, et al. The early decline of α-fetoprotein and des-γ-carboxy prothrombin predicts the response of hepatic arterial infusion chemotherapy in hepatocellular carcinoma patients. Gastrointestinal Tumors. (2020) 7:83–92. doi: 10.1159/000506941, PMID: 32903927

PubMed Abstract | Crossref Full Text | Google Scholar

21. Matsui D, Nagai H, Mukozu T, Ogino YU, and Sumino Y. VEGF in patients with advanced hepatocellular carcinoma receiving intra-arterial chemotherapy. Anticancer Res. (2015) 35:2205–10., PMID: 25862879

PubMed Abstract | Google Scholar

22. Yang Y, Li G, Lu Z, Liu Y, Kong J, and Liu J. Progression of prothrombin induced by vitamin K absence-II in hepatocellular carcinoma. Front Oncol. (2025) 11:726213/full. doi: 10.3389/fonc.2021.726213/full

Crossref Full Text | Google Scholar

23. Sung PS, Choi MH, Yang H, Lee SK, Chun HJ, Jang JW, et al. Diffusion-weighted magnetic resonance imaging in hepatocellular carcinoma as a predictor of a response to cisplatin-based hepatic arterial infusion chemotherapy. Front Oncol. (2020) 10:600233/full. doi: 10.3389/fonc.2020.600233/full, PMID: 33330098

PubMed Abstract | Crossref Full Text | Google Scholar

24. Oura K, Morishita A, Tani J, Nomura T, Manabe T, Takuma K, et al. Prognostic value of skeletal muscle loss in patients with hepatocellular carcinoma treated with hepatic arterial infusion chemotherapy. Cancers. (2023) 15:1834. doi: 10.3390/cancers15061834, PMID: 36980720

PubMed Abstract | Crossref Full Text | Google Scholar

25. Yi X, Fu Y, Long Q, Zhao Y, Li S, Zhou C, et al. Myosteatosis can predict unfavorable outcomes in advanced hepatocellular carcinoma patients treated with hepatic artery infusion chemotherapy and anti-PD-1 immunotherapy. Front Oncol. (2022) 12:892192/full. doi: 10.3389/fonc.2022.892192/full, PMID: 35651812

PubMed Abstract | Crossref Full Text | Google Scholar

26. Terashima T, Honda M, Toyama T, Shimakami T, Shimizu R, Takatori H, et al. IL-28B variant as a predictor in patients with advanced hepatocellular carcinoma treated with hepatic arterial infusion chemotherapy. J Gastroenterol Hepatol. (2020) 35:1813–20. doi: 10.1111/jgh.15035, PMID: 32180251

PubMed Abstract | Crossref Full Text | Google Scholar

27. Lyu N, Wang X, Li JB, Lai JF, Chen QF, Li SL, et al. Arterial chemotherapy of oxaliplatin plus fluorouracil versus sorafenib in advanced hepatocellular carcinoma: A biomolecular exploratory, randomized, phase III trial (FOHAIC-1). JCO. (2022) 40:468–80. doi: 10.1200/JCO.21.01963, PMID: 34905388

PubMed Abstract | Crossref Full Text | Google Scholar

28. He X, Li K, Wei R, Zuo M, Yao W, Zheng Z, et al. A multitask deep learning radiomics model for predicting the macrotrabecular-massive subtype and prognosis of hepatocellular carcinoma after hepatic arterial infusion chemotherapy. Radiol Med. (2023) 128:1508–20. doi: 10.1007/s11547-023-01719-1, PMID: 37801197

PubMed Abstract | Crossref Full Text | Google Scholar

29. Zhao Y, Huang F, Liu S, Jian L, Xia X, Lin H, et al. Prediction of therapeutic response of unresectable hepatocellular carcinoma to hepatic arterial infusion chemotherapy based on pretherapeutic MRI radiomics and Albumin-Bilirubin score. J Cancer Res Clin Oncol. (2023) 149:5181–92. doi: 10.1007/s00432-022-04467-3, PMID: 36369395

PubMed Abstract | Crossref Full Text | Google Scholar

30. Zwanenburg A, Vallières M, Abdalah MA, Aerts HJWL, Andrearczyk V, Apte A, et al. The image biomarker standardization initiative: standardized quantitative radiomics for high-throughput image-based phenotyping. Radiology. (2020) 295:328–38. doi: 10.1148/radiol.2020191145, PMID: 32154773

PubMed Abstract | Crossref Full Text | Google Scholar

31. Demircioğlu A. Reproducibility and interpretability in radiomics: a critical assessment. Diagn Interv Radiol. (2024) 31:321–8. doi: 10.4274/dir.2024.242719, PMID: 39463040

PubMed Abstract | Crossref Full Text | Google Scholar

32. Riberdy V, Guida A, Rioux J, and Brewer K. Radiomics in preclinical imaging research: methods, challenges and opportunities. NPJ Imaging. (2025) 3:45. doi: 10.1038/s44303-025-00104-z, PMID: 40983685

PubMed Abstract | Crossref Full Text | Google Scholar

33. Renzulli M, Brocchi S, Cucchetti A, Mazzotti F, Mosconi C, Sportoletti C, et al. Can current preoperative imaging be used to detect microvascular invasion of hepatocellular carcinoma? Radiology. (2016) 279:432–42. doi: 10.1148/radiol.2015150998, PMID: 26653683

PubMed Abstract | Crossref Full Text | Google Scholar