Next Article in Journal
The LINC02381/let-7g-5p/THBS1 Signaling Axis Modulates Cellular Proliferative Activity in Osteosarcoma
Previous Article in Journal
Functional Heterogeneity and Context-Dependent Roles of LncRNAs in Breast Cancer
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Cancers
Volume 17
Issue 19
10.3390/cancers17193192
cancers-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Use of a Pathomics Signature to Predict the Prognosis of Hepatocellular Carcinoma with Cirrhosis: A Multicentre Retrospective Study

by
Ting Wang
1,†,
Jixiang Zheng
2,†,
Lingling Guo
3,
Jiawen Fan
4,
Yubin Lu
5,
Zhen Peng
1,
Yanfeng Zhong
6,
Zhengjun Zhou
7,* and
Erbao Chen
1,*
1
Department of Hepatobiliary and Pancreatic Surgery, Peking University Shenzhen Hospital, Shenzhen 518036, China
2
Department of General Surgery, Guangzhou Digestive Disease Center, Guangzhou First People’s Hospital, Guangzhou Medical University, Guangzhou 510180, China
3
Department of Radiation Oncology, Peking University Shenzhen Hospital, Shenzhen 518036, China
4
Department of Gastrointestinal Surgery, Shenzhen Hospital of Southern Medical University, Shenzhen 518100, China
5
Minimally Invasive Interventional Department, Peking University Shenzhen Hospital, Shenzhen 518036, China
6
Central Laboratory, Peking University Shenzhen Hospital, Shenzhen 518036, China
7
Liver Cancer Institute, Zhongshan Hospital, Fudan University, Shanghai 200032, China
*
Authors to whom correspondence should be addressed.
†
These authors contributed equally to this article.
Cancers 2025, 17(19), 3192; https://doi.org/10.3390/cancers17193192
Submission received: 13 August 2025 / Revised: 19 September 2025 / Accepted: 23 September 2025 / Published: 30 September 2025
(This article belongs to the Section Cancer Biomarkers)
Browse Figures
Versions Notes

Simple Summary

Prognostic models are lacking for cirrhotic hepatocellular carcinoma (HCC). We constructed a pathomics signature based on pathomics features extracted from digital H-E images and LASSO Cox regression, which was easy to reproduce and may be conveniently applied in clinical practice. The pathomics signature is the independent predictor of OS and DFS, and the nomograms incorporating the pathomics signature and clinicopathological factors outperformed the traditional model consisting of clinicopathological factors alone. These findings may contribute to the use of the pathomics signature in clinical practice and facilitate personalized treatment strategies for HCC with cirrhosis.

Abstract

Background: Hepatocellular carcinoma (HCC) is a highly aggressive and heterogeneous malignancy which predominantly arises in the setting of cirrhosis, and there is lack of models to predict prognosis in cirrhotic HCC. This study aims to develop and validate a prediction model based on the pathomics signature and clinicopathological characteristics to predict the prognosis of HCC with cirrhosis. Methods: In this multicenter, retrospective study, 389 patients were enrolled (training cohort: 268; independent validation cohort: 121). A total of 351 pathomics features were extracted from digital H-E–stained images, and a pathomics signature (PSHCC) was constructed using a least absolute shrinkage and selection operator Cox regression model. Then two nomograms were established by combining the PSHCC and clinicopathological characteristics. Further validation was performed in the validation cohort. Results: This study included 389 patients. A 24 feature-based PSHCC was constructed. A higher PSHCC was significantly associated with worse OS and DFS in both the training (OS: hazard ratio [HR], 4.341 [95% CI, 3.109–6.062]; DFS: HR, 3.058 [95% CI, 2.223–4.207]) and validation (OS: HR, 4.145 [95% CI, 2.357–7.291]; DFS: HR, 3.395 [95% CI, 2.104–5.479]) cohorts (p < 0.001 for all comparisons). Multivariable analysis revealed that the PSHCC was an independent factor associated with OS and DFS. Integrating the PSHCC into pathomics nomograms resulted in better performance for prognosis prediction than the traditional model in both cohorts. Conclusions: The PSHCC may serve as a reliable surrogate for prognosis, and the nomograms offer promising tools to predict individual outcomes, facilitating personalized management of HCC with cirrhosis.

1. Introduction

Hepatocellular carcinoma (HCC), the most prevalent liver malignancy, ranks as the third leading cause of cancer-related mortality globally [1]. While hepatectomy remains a curative option for specific patients with HCC, recurrence occurs in 50–70% of cases within five years, and the overall 5-year survival rate ranges from 25% to 55% [2,3,4,5]. Notably, HCC is distinguished by its association with cirrhosis in 90% of cases, with primary causes including hepatitis B or C virus, nonalcoholic fatty liver disease, and alcohol-related liver disease [6]. Patients with cirrhosis face a heightened risk of postoperative complications and a poorer prognosis than those without [7]. Accurate prognostic assessment in cirrhotic HCC remains a critical challenge for personalized treatment approaches.
Current HCC staging systems, such as the Barcelona Clinic Liver Cancer (BCLC), Hong Kong Liver Cancer (HKLC), and American Joint Committee on Cancer (AJCC) TNM systems, have been employed to perform prognostic stratification and inform treatment decisions in HCC [1,2,3,4]. However, these models fail to account for the immunosuppressive environment induced by cirrhosis, limiting their efficacy in precise risk stratification [8]. Consequently, there is an urgent need for predictive models to optimize individual outcome prediction and treatment strategies, particularly for patients with cirrhosis.
Hematoxylin and eosin (H&E)-stained slides provide critical diagnostic information and contain vast morphometric data that are potentially indicative of survival outcomes—details often beyond the scope of routine pathological evaluation. Advances in digital pathology and artificial intelligence have enabled deeper extraction of these features, giving rise to the field of “pathomics,” which has garnered increasing interest [9,10]. Pathomics encompasses the quantitative extraction of data from digital histopathology images, capturing cellular architecture and histological texture in both tumor cells and the surrounding extracellular matrix [11,12,13]. Unlike traditional staging systems, pathomics offers a comprehensive characterization of the HCC tumor microenvironment. Previous studies have demonstrated its utility in predicting outcomes for lung, bladder, renal, gastric, and colon cancers [14,15,16,17,18]. Building on this, we hypothesized that pathomics analysis of digital H&E-stained images could enhance prognostic predictions for cirrhotic HCC.
Integrating multiple features into a unified signature, as opposed to analyzing them individually, has the potential to significantly enhance predictive accuracy [19,20]. The least absolute shrinkage and selection operator (LASSO) Cox regression, a powerful machine learning algorithm for high-dimensional data, is widely applied in prognostic modeling [21,22,23]. In this study, we employed LASSO Cox regression to derive a pathomics signature for HCC (PSHCC) based on quantitative features extracted from digital H&E images. Subsequently, we evaluated the prognostic efficacy of the PSHCC. To facilitate clinical application, we established and validated two pathomics nomograms, incorporating both PSHCC and clinicopathological factors, to individually predict overall survival (OS) and disease-free survival (DFS) in cirrhotic patients with HCC.

2. Materials and Methods

2.1. Study Design

The study design is illustrated in Figure 1. In the training cohort, a total of 351 pathomics features were obtained, followed by the construction of a pathomics signature (PSHCC) based on OS-associated features identified via LASSO Cox regression. Two pathomics nomograms, incorporating both the PSHCC and clinical predictors, were then constructed to predict prognosis in HCC with cirrhosis. Their predictive performances were validated in the external validation cohort. Additionally, two clinicopathological models were established to evaluate the incremental prognostic value provided by the PSHCC.

2.2. Participants

This retrospective study included 389 HCC participants with cirrhosis, sourced from two medical centers. The training cohort comprised 268 consecutive patients from Zhongshan Hospital, Fudan University, received curative resection between 1 July 2010 and 31 March 2013. Inclusion criteria were (1) histologically confirmed HCC, treated with curative resection; (2) Child–Pugh class A or B with liver cirrhosis; (3) no prior history of other malignancies; and (4) availability of comprehensive clinicopathological and follow-up data. Patients who underwent repeat hepatectomy, local ablation, or transcatheter arterial chemoembolization prior to surgery were excluded. The validation cohort, consisting of 121 patients, was selected from Peking University Shenzhen Hospital between 1 July 2008 and 31 March 2011, using identical inclusion and exclusion criteria.
Baseline patient data, including age, sex, HBV infection, Child–Pugh grade, AFP level, tumor number, tumor size, differentiation grade, encapsulation, vascular invasion, TNM stage, and follow-up information, were collected. Follow-up occurred every three months during the first two years post-surgery, every six months for the subsequent three years, and annually thereafter. Follow-up duration was measured from the surgery date to the last follow-up. In this study, curative resection was defined as complete macroscopic removal of tumor, as confirmed by intraoperative inspection and postoperative imaging (CT/MRI within four weeks after surgery) and surgical margins were histologically negative (R0 resection) for cancer cells. This study included HCC patients with cirrhosis who have worse survival than those without cirrhosis [7]. Thus, a high rate of early recurrence is reported despite enrolled patients receiving curative resection.
This study was in accordance with the Declaration of Helsinki and approved by the Institutional Review Boards of Fudan University Zhongshan Hospital (approval number: B2024-199, date: 31 May 2024) and Peking University Shenzhen Hospital (approval number: 2022-164, date: 2 December 2022). Written informed consent was waived by the institutional review boards for this retrospective study. The study followed the TRIPOD (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis) statement guidelines.

2.3. Sample Preparation and Region of Interest Selection

Initially, two experienced pathologists, blinded to the prognostic data, reviewed the H&E-stained slides and selected the most representative samples. Paraffin-embedded tissue sections were then cut at a thickness of 5 μm and stained with H&E. All slides were subsequently scanned using the Aperio ScanScope system at ×20 magnification, digitized into svs format files. Under the supervision of an experienced pathologist, the tumor regions in each section were identified. To enhance efficiency, a pathologist selected ten non-overlapping representative tiles per core, each containing the most tumor cells, with a field of view of 1000 × 1000 pixels (0.504 μm/pixel). The selected tiles were confirmed by a second pathologist, and disagreements were resolved through consultation with the pathology director. All pathologists remained blinded to clinical and prognostic data. The selected tiles were saved in tif format.

2.4. Pathomics Feature Extraction

Pathomics features were extracted from digital H&E-stained images using CellProfiler, an open-source software (version 4.2.1; Broad Institute; Cambridge, MA, USA, https://cellprofiler.org/, accessed on 5 January 2023) [24]. Initially, the selected H&E-stained tiles were split into hematoxylin-stained and eosin-stained grayscale images via the “UnmixColors” module. Additionally, the H&E-stained tiles were converted to grayscale using the “ColorToGray” module with the “Combine” method for subsequent processing. Image quality metrics for the grayscale, hematoxylin, and eosin channels were evaluated through the “MeasureImageQuality”, “MeasureImageIntensity”, and “MeasureObjectIntensity” modules. Pixel-level colocalization and intensity correlation features across an entire image were obtained using the “MeasureColocalization” module. Granularity features, reflecting the spectra of size measurements of textures within the image, were quantified via the “MeasureGranularity” module. Finally, texture features, which describe the roughness and smoothness of textures within the images, were derived using the “MeasureTexture” module. A total of 351 features were obtained and presented in Table S1.

2.5. PSHCC Construction

LASSO Cox regression, a robust method for high-dimensional predictor analysis in survival studies, was used [16,17]. It employs an L1 penalty to shrink coefficients toward zero, with the tuning constant λ controlling the penalty strength. A smaller λ allows more predictors into the model. The optimal value of λ is determined by 10-fold cross-validation. In the training cohort, we used LASSO Cox regression to select the most prognostic pathomics features and constructed a formula based on a linear combination of the selected features to calculate the pathomics signature (PSHCC). The PSHCC for the validation cohort was calculated using the derived formula. The PSHCC calculation formula is presented in the Supplementary File S1.

2.6. Prognostic Value of the PSHCC

The optimal PSHCC cutoff value was established using maximally selected rank statistics, dividing patients in the training cohort into high- and low-PSHCC groups. The same cutoff was utilized to stratify patients in the validation cohort. The Kaplan–Meier method and log-rank test were used to assess OS and DFS between the high- and low-PSHCC groups.

2.7. Development and Validation of the Pathomics Nomogram for Prognosis

In the training cohort, univariate Cox regression analyses were performed for OS and DFS, incorporating both PSHCC and clinicopathological features, with variables showing p < 0.05 selected for subsequent multivariable analysis. Backward stepwise Cox regression identified the independent predictors. Based on these predictors, two pathomics nomograms were constructed to estimate OS and DFS.
To assess the discrimination of the pathomics nomograms, area under the receiver operating characteristic curve (AUROC) and concordance index (C-index) were calculated. Calibration curves were generated to evaluate the alignment between predicted and actual survival probabilities, and decision curve analysis (DCA) was applied to assess the nomograms’ clinical utility. These nomograms were further validated in the external cohort to confirm their discrimination, calibration, and clinical applicability.

2.8. Incremental Value of the PSHCC for Prognosis Prediction

The incremental prognostic value of the PSHCC, when added to clinicopathological models including clinicopathological characteristics only, was evaluated using C-index and AUROC. Comparisons of C-indexes and AUROCs at 5 years between models were conducted using the z-score and DeLong tests.

2.9. Statistical Analysis

Continuous variables were analyzed using the t-test, while categorical variables were examined using the χ2 test. The Kaplan–Meier method, along with the log-rank test, was employed to assess OS and DFS. Univariate and multivariate Cox regression analyses were performed, with HR and 95% CI calculated for each variable. All statistical analyses were carried out using R (version 4.1.3) software. All p-values were two-sided, with statistical significance set at p < 0.05.

3. Results

3.1. Patient Characteristics

The clinicopathologic characteristics of patients in the training (n = 268) and validation (n = 121) cohorts are detailed in Table 1. Among the 389 participants, 343 (87.3%) were male. No significant differences in clinicopathological features were observed between the training and validation cohorts. In the training cohort, the median follow-up duration [interquartile range (IQR)] was 57 months (26–79), with 5-year OS and DFS rates of 51.6% and 44.2%, respectively (Figure S1A,B). The validation cohort had a median (IQR) follow-up of 66 months (33–80), with corresponding 5-year OS and DFS rates of 60.9% and 46.7% (Figure S1C,D).

3.2. Construction of the PSHCC

The PSHCC construction framework is depicted in Figure 2. In the training cohort, a 24-feature-based PSHCC was developed using the LASSO Cox regression model (Figure S2). The formula used to compute the PSHCC is provided in the Supplementary Information, from which the PSHCC of the validation cohort was directly derived.

3.3. Association of the PSHCC with Prognosis

The optimal PSHCC cutoff value, determined as 14.469 via the “survminer” package in the training cohort (Figure S3), was used to categorize all patients into high- and low-PSHCC groups. High-PSHCC patients in the training cohort had significantly shorter OS and DFS than their low-PSHCC counterparts, with hazard ratios (HRs) of 4.341 (95% CI: 3.109–6.062; log-rank p < 0.001) and 3.058 (95% CI: 2.223–4.207; log-rank p < 0.001), respectively (Figure 3A).
Similar analyses were performed in the validation cohort, where high-PSHCC patients also exhibited significantly reduced OS and DFS compared to low-PSHCC patients, with HRs of 4.145 (95% CI: 2.357–7.291; log-rank p < 0.001) and 3.395 (95% CI: 2.104–5.479; log-rank p < 0.001), respectively (Figure 3B). The PSHCC demonstrated robust performance in estimating 2-, 3-, and 5-year OS and DFS across both cohorts (Figure S4). Furthermore, it remained an independent prognostic factor when stratified by clinical and pathological variables, underscoring its prognostic value (Figures S5–S8).

3.4. Construction and Validation of the Pathomics Nomograms

In the univariate Cox regression analysis, PSHCC, AFP, tumor size, tumor number, differentiation grade, vascular invasion, and TNM stage were significantly linked to OS, while PSHCC, AFP, tumor size, tumor number, vascular invasion, and TNM stage were significantly associated with DFS (Table 2). Multivariate Cox regression further identified PSHCC, tumor size, tumor number, and differentiation grade as independent predictors for OS, with PSHCC, tumor size, and tumor differentiation being independent risk factors for DFS. Consequently, two pathomics nomograms were developed by incorporating these independent factors to predict OS and DFS (Figure 4).
In the training cohort, the pathomics nomogram achieved a concordance index (C-index) of 0.761 (95% CI, 0.727–0.795) for OS and 0.703 (95% CI, 0.667–0.739) for DFS. Time-dependent receiver operating characteristic (ROC) curves showed area under the curve (AUROC) values of 0.827 (95% CI, 0.771–0.883), 0.823 (95% CI, 0.770–0.877), and 0.852 (95% CI, 0.807–0.898) at 2, 3, and 5 years for OS, and 0.791 (95% CI, 0.735–0.846), 0.799 (95% CI, 0.747–0.852), and 0.777 (95% CI, 0.720–0.834) for DFS at the same intervals (Figure S9A,B).
In the validation cohort, the pathomics nomogram produced a C-index of 0.774 (95% CI, 0.714–0.834) for OS and 0.720 (95% CI, 0.668–0.772) for DFS. The AUCs for OS were 0.830 (95% CI, 0.740–0.919), 0.841 (95% CI, 0.757–0.928), and 0.813 (95% CI, 0.731–0.895) at 2, 3, and 5 years, while DFS AUCs were 0.784 (95% CI, 0.702–0.866), 0.797 (95% CI, 0.71–0.876), and 0.752 (95% CI, 0.664–0.840) at corresponding time points (Figure S9C,D).
Calibration curves showed strong concordance between the predicted and actual survival rates at 2, 3, and 5 years in both cohorts (Figure S10). Decision curve analysis further confirmed that pathomics nomograms provided greater net benefit across a wider range of reasonable threshold probabilities compared to clinicopathological models (Figure S11).

3.5. Comparison Between the Pathomics Nomograms and Traditional Models

To investigate the incremental value of the PSHCC added to traditional models for predicting prognosis, we constructed two traditional models including tumor size, tumor number, and differentiation grade to estimate OS and DFS. In the training cohort, the traditional models achieved a C-index of 0.681 (95% CI, 0.641–0.721) for OS and 0.615 (95% CI, 0.573–0.657) for DFS. Pathomics nomograms, integrating PSHCC with clinicopathological risk factors, significantly improved the C-index to 0.761 (95% CI, 0.727–0.795, p < 0.001) for OS and 0.703 (95% CI, 0.667–0.739, p < 0.001) for DFS. These results could also be found in the validation cohort (Table S2). Similarly, the AUROCs of the traditional models were 0.734 (95% CI, 0.673–0.795) for OS and 0.654 (95% CI, 0.591–0.718) for DFS at 5 years. In contrast, the pathomics nomograms yielded significantly higher AUROCs of 0.852 (95% CI, 0.807–0.898, p < 0.001) for OS and 0.777 (95% CI, 0.720–0.834, p < 0.001) for DFS at 5 years (Figure S12), and these results could also be found in the validation cohort.

4. Discussion

In this study, the PSHCC was established through digital H&E images and LASSO Cox regression, demonstrating a significant association with prognosis in HCC individuals with cirrhosis. Additionally, two pathomics nomograms were developed by combining the PSHCC with clinicopathological predictors to assess OS and DFS. These newly constructed nomograms exhibited superior discrimination and calibration, outperforming traditional models based solely on clinicopathological factors. Validation cohort results reinforced the generalizability of the PSHCC, indicating its potential utility as a prognostic tool for individualized HCC management.
Accurate prognosis prediction is vital for effective risk stratification and improving survival in HCC, a highly heterogeneous disease often complicated by cirrhosis. Characterized by extracellular matrix remodeling and an immunosuppressive microenvironment, cirrhotic HCC presents challenges for conventional staging systems, which fail to guide precision treatment effectively. Pathomics, by systematically extracting histopathological features, holds promise for enhancing prognosis prediction across various cancers. This study is the first to utilize a pathomics signature to predict HCC prognosis within a cirrhotic liver.
Leveraging image-processing software and machine learning, an accessible and reproducible method was developed for clinicians to incorporate pathomics signatures into routine clinical workflows. CellProfiler, a free, open-source tool, was employed to automatically extract quantitative, high-dimensional pathomics features from H&E-stained images, proving effective in digital histopathology [24]. LASSO Cox, a widely used machine learning technique, identified prognostically relevant features by adjusting feature coefficients based on the lambda penalty parameter. This model was applied for dimensionality reduction, resulting in the construction of a clinically applicable pathomics classifier.
Significant efforts have been made to identify complementary prognostic biomarkers in HCC. In our previous studies, we developed a stemness-based classifier and a 6-gene signature for prognosis prediction. Gao et al. introduced a 9-immune-related lncRNA model to predict OS in HCC [23,25,26]. However, the high cost has significantly impeded the translation of these methods into clinical practice. While radiomics analyses have demonstrated favorable predictive value for HCC, the complexity of medical image processing and the lack of sufficient validation hinder their clinical application [27,28,29]. Pathomics, however, offers the potential to be integrated with other omics approaches to enhance risk stratification in various cancers. For instance, Wang et al. constructed a nomogram by incorporating pathomics, radiomics, and Immunoscore to predict lung metastasis in colorectal cancer [30]. Similarly, a multi-classifier system combining genomics and pathomics has been proposed to predict recurrence in papillary renal cell carcinoma [31]. It is anticipated that future multi-omics analyses, including genomic data, radiomics, immune infiltration status, and pathomics, will refine risk stratification and prognosis prediction for HCC.
Despite numerous studies presenting pathomics analyses of digital H&E-stained images in HCC, few have been translated into clinical practice [32,33]. One primary obstacle is that most reported methods rely on deep neural networks, with inaccessible codes, limiting the ability of researchers to develop or test deep learning-based models on their own data [34,35]. In contrast, open-source software like CellProfiler provides an accessible and user-friendly platform for pathomics analysis. Additionally, deep neural network-derived pathomics features often suffer from the “black box” issue, as they are not easily interpretable [32,33]. In contrast, the features extracted using CellProfiler are well-defined and interpretable, offering supplementary insights into cellular and histological characteristics [36,37].
Liver reserve function is a prognostic factor for HCC. Hepatectomy is recommended to resectable patients with Child–Pugh Class A and highly selected patients with Child–Pugh Class B liver function according to National Comprehensive Cancer Network (NCCN) Guidelines for Hepatocellular Carcinoma [38]. Thus, most patients in this study were Child–Pugh A. The albumin–bilirubin (ALBI) grade, calculated based on only albumin and total bilirubin, is also known to be predictive of prognosis for patients with HCC [39,40]. In univariate and multivariate analysis, we found that the Child–Pugh grade was not associated with the survival, which may be attributed to the majority of patients enrolled in this study having well-preserved liver function (Child–Pugh A). Importantly, PSHCC remained an independent predictor for OS and DFS.
To offer a practical tool that aids clinicians in decision-making for HCC with cirrhosis, two pathomics nomograms were developed by integrating PSHCC with clinicopathological risk factors. These nomograms exhibited superior prognostic performance compared to traditional clinicopathological models. In clinical practice, patients predicted to have poor prognosis could receive more tailored treatment and intensive monitoring to reduce the risk of recurrence. The pathomics nomograms we constructed outperformed prognostic models based on clinicopathological factors [41,42,43], with higher C-index and AUROC. A number of studies integrating pathomics signature and clinicopathological factors showed satisfactory performance in predicting the prognosis of HCC [44,45], yet their generalizability was limited due to the single-center designs or small sample sizes.
The two pathomics nomograms are well-suited for clinical application. First, tumor size and number, routinely assessed in clinical practice, are easily accessible post-hepatectomy. Second, the pathomics signature, derived from standard H&E-stained images, is straightforward to compute using the proposed workflow. Importantly, this process imposes no additional financial burden on patients, emphasizing the PSHCC’s translational value.
This study has certain limitations. As a retrospective study, it is subject to inherent biases despite external validation efforts to enhance reliability. Additionally, the study draws from two medical institutions in China, with the majority of patients having hepatitis B virus-related cirrhosis. Prospective randomized trials across diverse populations, including alcohol-related liver disease and nonalcoholic fatty liver disease, are needed to further assess the PSHCC’s predictive performance and clinical utility. Finally, we were unable to analyze the association between PSHCC and adjuvant therapy response due to the lack of this information. Thus, further investigations should focus on the relationship between PSHCC and adjuvant therapy response.

5. Conclusions

In summary, the PSHCC demonstrated a significant association with prognosis in HCC with cirrhosis. By incorporating the PSHCC with established clinicopathological predictors, two pathomics nomograms were developed and validated, outperforming traditional models in prognostic accuracy. These nomograms offer promising potential for guiding personalized management of HCC in patients with cirrhosis.

Supplementary Materials

The following supporting information can be downloaded at https://www.mdpi.com/article/10.3390/cancers17193192/s1, Figure S1: Kaplan–Meier survival analysis of the training and validation cohorts. Figure S2: Feature selection using a LASSO Cox regression model in training cohor; Figure S3: Selection of the optimum cutoff value for the PSHCC. Figure S4: Pathomics signature (PSHCC) measured by time-dependent ROC curves in the training and validation cohorts. Figure S5: Kaplan–Meier survival analysis of the OS according to the PSHCC level stratified by clinicopathological variables in the training cohort. Figure S6: Kaplan–Meier survival analysis of the DFS according to the PSHCC level stratified by clinicopathological variables in the training cohort. Figure S7: Kaplan–Meier survival analysis of the OS according to the PSHCC level stratified by clinicopathological variables in the validation cohort. Figure S8: Kaplan–Meier survival analysis of the DFS according to the PSHCC level stratified by clinicopathological variables in the validation cohort. Figure S9: Pathomics nomograms measured by time-dependent ROC curves in the training and validation cohorts. Figure S10: The calibration curves of the pathomics nomogram. Figure S11: Decision curve analyses of different models for OS and DFS. Figure S12: Time-dependent ROC curves of different models for OS and DFS. Table S1: Summary of the pathomics features; Table S2: Comparison of the C-indexes between the pathomics nomograms and clinicopathological models. File S1: PSHCC calculation formula.

Author Contributions

Conceptualization, T.W. and E.C.; methodology, T.W.; software, T.W.; validation, T.W., J.Z., L.G. and J.F.; formal analysis, T.W., J.Z., L.G. and J.F.; investigation, T.W., J.Z., L.G. and J.F.; resources, Y.L., Z.P., Y.Z. and E.C.; data curation, T.W. and J.Z.; writing—original draft preparation, T.W. and J.Z.; writing—review and editing, T.W. and J.Z.; visualization, T.W. and J.Z.; supervision, E.C. and Z.Z.; project administration, E.C.; funding acquisition, E.C. All authors have read and agreed to the published version of the manuscript.

Funding

This work was supported by the National Natural Science Foundation of China (82303446), the Shenzhen High-level Hospital Construction Fund, the Peking University Shenzhen Hospital Scientific Research Fund (KYQD2023303), the Guangdong Basic and Applied Basic Research Foundation (2023A1515220200), the Medical Scientific Research Foundation of Guangdong Province of China (A2024351), and the Science and Technology Development Fund Project of Shenzhen (JCYJ20240813115900001).

Institutional Review Board Statement

The study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board (or Ethics Committee) of Fudan University Zhongshan Hospital (approval number: B2024-199, date: 31 May 2024) and Peking University Shenzhen Hospital (approval number: 2022-164, date: 2 December 2022).

Informed Consent Statement

Written informed consent was waived by the institutional review boards for this retrospective study.

Data Availability Statement

Data and codes are available from the corresponding authors on reasonable request.

Conflicts of Interest

The authors declare no conflicts of interest.

Abbreviations

The following abbreviations are used in this manuscript:
HCCHepatocellular carcinoma
PSHCCPathomics signature of HCC
OSOverall survival
DFSDisease-free survival
HRHazard ratio
CIConfidential interval
BCLCBarcelona Clinic Liver Cancer
HKLCHong Kong Liver Cancer
AJCCAmerican Joint Committee on Cancer
H&EHematoxylin and eosin
LASSOLeast absolute shrinkage and selection operator
DCADecision curve analysis
ROCReceiver operating characteristic
AUROCArea under the receiver operating characteristic curve
NCCNNational Comprehensive Cancer Network
ALBIAlbumin–bilirubin

References

  1. Sung, H.; Ferlay, J.; Siegel, R.L.; Laversanne, M.; Soerjomataram, I.; Jemal, A.; Bray, F. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J. Clin. 2024, 74, 229–263. [Google Scholar] [CrossRef]
  2. Villanueva, A. Hepatocellular Carcinoma. N. Engl. J. Med. 2019, 380, 1450–1462. [Google Scholar] [CrossRef]
  3. European Association for the Study of the Liver. EASL Clinical Practice Guidelines: Management of hepatocellular carcinoma. J. Hepatol. 2018, 69, 182–236. [Google Scholar] [CrossRef]
  4. Vogel, A.; Cervantes, A.; Chau, I.; Daniele, B.; Llovet, J.M.; Meyer, T.; Nault, J.C.; Neumann, U.; Ricke, J.; Sangro, B.; et al. Hepatocellular carcinoma: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up. Ann. Oncol. 2018, 29, iv238–iv255. [Google Scholar] [CrossRef]
  5. Chen, E.; He, Y.; Jiang, J.; Yi, J.; Zou, Z.; Song, Q.; Ren, Q.; Lin, Z.; Lu, Y.; Liu, J.; et al. CDCA8 induced by NF-YA promotes hepatocellular carcinoma progression by regulating the MEK/ERK pathway. Exp. Hematol. Oncol. 2023, 12, 9. [Google Scholar] [CrossRef] [PubMed]
  6. Llovet, J.M.; Kelley, R.K.; Villanueva, A.; Singal, A.G.; Pikarsky, E.; Roayaie, S.; Lencioni, R.; Koike, K.; Zucman-Rossi, J.; Finn, R.S. Hepatocellular carcinoma. Nat. Rev. Dis. Primers 2021, 7, 6. [Google Scholar] [CrossRef] [PubMed]
  7. Sasaki, K.; Shindoh, J.; Margonis, G.A.; Nishioka, Y.; Andreatos, N.; Sekine, A.; Hashimoto, M.; Pawlik, T.M. Effect of Background Liver Cirrhosis on Outcomes of Hepatectomy for Hepatocellular Carcinoma. JAMA Surg. 2017, 152, e165059. [Google Scholar] [CrossRef] [PubMed]
  8. Shen, X.T.; Xie, S.Z.; Zheng, X.; Zou, T.T.; Hu, B.Y.; Xu, J.; Liu, L.; Xu, Y.F.; Wang, X.F.; Wang, H.; et al. Cirrhotic-extracellular matrix attenuates aPD-1 treatment response by initiating immunosuppressive neutrophil extracellular traps formation in hepatocellular carcinoma. Exp. Hematol. Oncol. 2024, 13, 20. [Google Scholar] [CrossRef]
  9. Baxi, V.; Edwards, R.; Montalto, M.; Saha, S. Digital pathology and artificial intelligence in translational medicine and clinical practice. Mod. Pathol. 2022, 35, 23–32. [Google Scholar] [CrossRef]
  10. Rashid, R.; Chen, Y.A.; Hoffer, J.; Muhlich, J.L.; Lin, J.R.; Krueger, R.; Pfister, H.; Mitchell, R.; Santagata, S.; Sorger, P.K. Narrative online guides for the interpretation of digital-pathology images and tissue-atlas data. Nat. Biomed. Eng. 2022, 6, 515–526. [Google Scholar] [CrossRef]
  11. Jiang, Y.; Yang, M.; Wang, S.; Li, X.; Sun, Y. Emerging role of deep learning-based artificial intelligence in tumor pathology. Cancer Commun. 2020, 40, 154–166. [Google Scholar] [CrossRef] [PubMed]
  12. Bera, K.; Schalper, K.A.; Rimm, D.L.; Velcheti, V.; Madabhushi, A. Artificial intelligence in digital pathology—New tools for diagnosis and precision oncology. Nat. Rev. Clin. Oncol. 2019, 16, 703–715. [Google Scholar] [CrossRef] [PubMed]
  13. Niazi, M.K.K.; Parwani, A.V.; Gurcan, M.N. Digital pathology and artificial intelligence. Lancet Oncol. 2019, 20, e253–e261. [Google Scholar] [CrossRef] [PubMed]
  14. Zhang, Y.; Yang, Z.; Chen, R.; Zhu, Y.; Liu, L.; Dong, J.; Zhang, Z.; Sun, X.; Ying, J.; Lin, D.; et al. Histopathology images-based deep learning prediction of prognosis and therapeutic response in small cell lung cancer. NPJ Digit. Med. 2024, 7, 15. [Google Scholar] [CrossRef]
  15. Chen, S.; Jiang, L.; Zheng, X.; Shao, J.; Wang, T.; Zhang, E.; Gao, F.; Wang, X.; Zheng, J. Clinical use of machine learning-based pathomics signature for diagnosis and survival prediction of bladder cancer. Cancer Sci. 2021, 112, 2905–2914. [Google Scholar] [CrossRef]
  16. Chen, S.; Jiang, L.; Gao, F.; Zhang, E.; Wang, T.; Zhang, N.; Wang, X.; Zheng, J. Machine learning-based pathomics signature could act as a novel prognostic marker for patients with clear cell renal cell carcinoma. Br. J. Cancer 2022, 126, 771–777. [Google Scholar] [CrossRef]
  17. Chen, D.; Fu, M.; Chi, L.; Lin, L.; Cheng, J.; Xue, W.; Long, C.; Jiang, W.; Dong, X.; Sui, J.; et al. Prognostic and predictive value of a pathomics signature in gastric cancer. Nat. Commun. 2022, 13, 6903. [Google Scholar] [CrossRef]
  18. Jiang, W.; Wang, H.; Dong, X.; Yu, X.; Zhao, Y.; Chen, D.; Yan, B.; Cheng, J.; Zhuo, S.; Wang, H.; et al. Pathomics Signature for Prognosis and Chemotherapy Benefits in Stage III Colon Cancer. JAMA Surg. 2024, 159, 519–528. [Google Scholar] [CrossRef]
  19. Jiang, Y.; Zhang, Q.; Hu, Y.; Li, T.; Yu, J.; Zhao, L.; Ye, G.; Deng, H.; Mou, T.; Cai, S.; et al. ImmunoScore Signature: A Prognostic and Predictive Tool in Gastric Cancer. Ann. Surg. 2018, 267, 504–513. [Google Scholar] [CrossRef]
  20. Chen, D.; Chen, H.; Chi, L.; Fu, M.; Wang, G.; Wu, Z.; Xu, S.; Sun, C.; Xu, X.; Lin, L.; et al. Association of Tumor-Associated Collagen Signature with Prognosis and Adjuvant Chemotherapy Benefits in Patients with Gastric Cancer. JAMA Netw. Open. 2021, 4, e2136388. [Google Scholar] [CrossRef]
  21. Ranstam, J.; Cook, J.A. LASSO regression. Br. J. Surg. 2018, 105, 1348. [Google Scholar] [CrossRef]
  22. Tibshirani, R. The lasso method for variable selection in the Cox model. Stat. Med. 1997, 16, 385–395. [Google Scholar] [CrossRef]
  23. Chen, E.; Zou, Z.; Wang, R.; Liu, J.; Peng, Z.; Gan, Z.; Lin, Z.; Liu, J. Predictive value of a stemness-based classifier for prognosis and immunotherapy response of hepatocellular carcinoma based on bioinformatics and machine-learning strategies. Front. Immunol. 2024, 15, 1244392. [Google Scholar] [CrossRef] [PubMed]
  24. McQuin, C.; Goodman, A.; Chernyshev, V.; Kamentsky, L.; Cimini, B.A.; Karhohs, K.W.; Doan, M.; Ding, L.; Rafelski, S.M.; Thirstrup, D.; et al. CellProfiler 3.0: Next-generation image processing for biology. PLoS Biol. 2018, 16, e2005970. [Google Scholar] [CrossRef]
  25. He, Y.; Zou, Z.; Lan, Z.; Chang, M.; Zhang, X.; Lin, R.; Zhang, W.; Zhang, G.; Wang, T.; Chen, E. Prognostic implication of six m6A-modulated genes signature in the ferroptosis for hepatocellular carcinoma patients. Clin. Exp. Med. 2025, 25, 180. [Google Scholar] [CrossRef]
  26. Gao, D.; Li, S.; Guo, H.; Liu, X.; Liu, Z.; Li, L.; Bao, L.; Dang, X. A Predictive Model Based on the FBXO Family Reveals the Significance of Cyclin F in Hepatocellular Carcinoma. Front. Biosci. 2024, 29, 202. [Google Scholar] [CrossRef]
  27. Wang, F.; Zhan, G.; Chen, Q.Q.; Xu, H.Y.; Cao, D.; Zhang, Y.Y.; Li, Y.H.; Zhang, C.J.; Jin, Y.; Ji, W.B.; et al. Multitask deep learning for prediction of microvascular invasion and recurrence-free survival in hepatocellular carcinoma based on MRI images. Liver Int. 2024, 44, 1351–1362. [Google Scholar] [CrossRef]
  28. Ji, G.W.; Zhu, F.P.; Xu, Q.; Wang, K.; Wu, M.Y.; Tang, W.W.; Li, X.C.; Wang, X.H. Radiomic Features at Contrast-enhanced CT Predict Recurrence in Early Stage Hepatocellular Carcinoma: A Multi-Institutional Study. Radiology 2020, 294, 568–579. [Google Scholar] [CrossRef]
  29. Xu, X.; Zhang, H.L.; Liu, Q.P.; Sun, S.W.; Zhang, J.; Zhu, F.P.; Yang, G.; Yan, X.; Zhang, Y.D.; Liu, X.S. Radiomic analysis of contrast-enhanced CT predicts microvascular invasion and outcome in hepatocellular carcinoma. J. Hepatol. 2019, 70, 1133–1144. [Google Scholar] [CrossRef]
  30. Wang, R.; Dai, W.; Gong, J.; Huang, M.; Hu, T.; Li, H.; Lin, K.; Tan, C.; Hu, H.; Tong, T.; et al. Development of a novel combined nomogram model integrating deep learning-pathomics, radiomics and immunoscore to predict postoperative outcome of colorectal cancer lung metastasis patients. J. Hematol. Oncol. 2022, 15, 11. [Google Scholar] [CrossRef]
  31. Huang, K.B.; Gui, C.P.; Xu, Y.Z.; Li, X.S.; Zhao, H.W.; Cao, J.Z.; Chen, Y.H.; Pan, Y.H.; Liao, B.; Cao, Y.; et al. A multi-classifier system integrated by clinico-histology-genomic analysis for predicting recurrence of papillary renal cell carcinoma. Nat. Commun. 2024, 15, 6215. [Google Scholar] [CrossRef]
  32. Qu, W.F.; Tian, M.X.; Lu, H.W.; Zhou, Y.F.; Liu, W.R.; Tang, Z.; Yao, Z.; Huang, R.; Zhu, G.Q.; Jiang, X.F.; et al. Development of a deep pathomics score for predicting hepatocellular carcinoma recurrence after liver transplantation. Hepatol. Int. 2023, 17, 927–941. [Google Scholar] [CrossRef]
  33. Liu, Z.; Liu, Y.; Zhang, W.; Hong, Y.; Meng, J.; Wang, J.; Zheng, S.; Xu, X. Deep learning for prediction of hepatocellular carcinoma recurrence after resection or liver transplantation: A discovery and validation study. Hepatol. Int. 2022, 16, 577–589. [Google Scholar] [CrossRef]
  34. Bankhead, P. Developing image analysis methods for digital pathology. J. Pathol. 2022, 257, 391–402. [Google Scholar] [CrossRef] [PubMed]
  35. Carpenter, A.E.; Kamentsky, L.; Eliceiri, K.W. A call for bioimaging software usability. Nat. Methods 2012, 9, 666–670. [Google Scholar] [CrossRef] [PubMed]
  36. Rudin, C. Stop Explaining Black Box Machine Learning Models for High Stakes Decisions and Use Interpretable Models Instead. Nat. Mach. Intell. 2019, 1, 206–215. [Google Scholar] [CrossRef] [PubMed]
  37. Madabhushi, A.; Lee, G. Image analysis and machine learning in digital pathology: Challenges and opportunities. Med. Image Anal. 2016, 33, 170–175. [Google Scholar] [CrossRef]
  38. National Comprehensive Cancer Network. NCCN Clinical Practice Guidelines in Oncology (NCCN Guidelines®) Hepatocellular Carcinoma. Version 4.2024. Available online: https://www.nccn.org/professionals/physician_gls/pdf/hcc.pdf (accessed on 12 March 2025).
  39. Hiraoka, A.; Kumada, T.; Michitaka, K.; Kudo, M. Newly Proposed ALBI Grade and ALBI-T Score as Tools for Assessment of Hepatic Function and Prognosis in Hepatocellular Carcinoma Patients. Liver Cancer 2019, 8, 312–325. [Google Scholar] [CrossRef]
  40. Tadokoro, T.; Kobayashi, T.; Honmyo, N.; Kuroda, S.; Ohira, M.; Hashimoto, M.; Oishi, K.; Oshita, A.; Abe, T.; Onoe, T.; et al. Albumin-Butyrylcholinesterase as a Novel Prognostic Biomarker for Hepatocellular Carcinoma Post-hepatectomy: A Retrospective Cohort Study with the Hiroshima Surgical Study Group of Clinical Oncology. Ann. Surg. Oncol. 2025, 32, 1973–1984. [Google Scholar] [CrossRef]
  41. Akabane, M.; Kawashima, J.; Altaf, A.; Woldesenbet, S.; Cauchy, F.; Aucejo, F.; Popescu, I.; Kitago, M.; Martel, G.; Ratti, F.; et al. Enhancing Recurrence-Free Survival Prediction in Hepatocellular Carcinoma: A Time-Updated Model Incorporating Tumor Burden and AFP Dynamics. Ann. Surg. Oncol. 2025, 32, 5648–5656. [Google Scholar] [CrossRef]
  42. Xu, X.F.; Wu, H.; Gu, L.H.; Zhao, Y.Z.; Zhou, Y.H.; Chen, T.H.; Guo, H.W.; Chen, Z.; Lin, K.Y.; Gu, W.M.; et al. Development and Validation of an Individualized Prediction Model for Postoperative Late Recurrence After Hepatectomy for Hepatocellular Carcinoma (POLAR-HCC): A Multicenter Study. Ann. Surg. Oncol. 2025; Epub ahead of print. [Google Scholar] [CrossRef]
  43. Berhane, S.; Fox, R.; García-Fiñana, M.; Cucchetti, A.; Johnson, P. Using prognostic and predictive clinical features to make personalised survival prediction in advanced hepatocellular carcinoma patients undergoing sorafenib treatment. Br. J. Cancer 2019, 121, 117–124. [Google Scholar] [CrossRef]
  44. Xie, Q.; Zhao, Z.; Yang, Y.; Wang, X.; Wu, W.; Jiang, H.; Hao, W.; Peng, R.; Luo, C. A clinical-radiomic-pathomic model for prognosis prediction in patients with hepatocellular carcinoma after radical resection. Cancer Med. 2024, 13, e7374. [Google Scholar] [CrossRef]
  45. Feng, L.; Huang, W.; Pan, X.; Ruan, F.; Li, X.; Tan, S.; Long, L. Predicting overall survival in hepatocellular carcinoma patients via a combined MRI radiomics and pathomics signature. Transl. Oncol. 2025, 51, 102174. [Google Scholar] [CrossRef]
Cancers 17 03192 g001
Figure 1. Flowchart of the study. * Patients from Fudan University Zhongshan Hospital. † Patients from Peking University Shenzhen Hospital.
Figure 1. Flowchart of the study. * Patients from Fudan University Zhongshan Hospital. † Patients from Peking University Shenzhen Hospital.
Cancers 17 03192 g001
Cancers 17 03192 g002
Figure 2. Construction framework of the PSHCC. (A) Ten regions of interest (ROI), each with a field of view of 1000 × 1000 μm and containing the highest density of tumor cells, were randomly selected from the digital H&E-stained images for analysis. Scale bars: 3000 μm and 50 μm. (B) A total of 351 pathomics features were extracted from these representative H&E tiles. (C) The PSHCC was constructed using a least absolute shrinkage and selection operator Cox regression model in the training cohort. PSHCC: pathomics signature of hepatocellular carcinoma; H&E: hematoxylin and eosin.
Figure 2. Construction framework of the PSHCC. (A) Ten regions of interest (ROI), each with a field of view of 1000 × 1000 μm and containing the highest density of tumor cells, were randomly selected from the digital H&E-stained images for analysis. Scale bars: 3000 μm and 50 μm. (B) A total of 351 pathomics features were extracted from these representative H&E tiles. (C) The PSHCC was constructed using a least absolute shrinkage and selection operator Cox regression model in the training cohort. PSHCC: pathomics signature of hepatocellular carcinoma; H&E: hematoxylin and eosin.
Cancers 17 03192 g002
Cancers 17 03192 g003
Figure 3. Kaplan–Meier survival curves based on PSHCC levels. (A,B) Kaplan–Meier curves depicting differences in OS and DFS between high- and low-PSHCC groups in the training cohort. (C,D) Kaplan–Meier curves showing OS and DFS differences between high- and low-PSHCC groups in the validation cohort. Left panel: OS; right panel: DFS. OS and DFS comparisons between the two groups were performed using a two-sided log-rank test. OS: overall survival; DFS: disease-free survival; PSHCC: pathomics signature of hepatocellular carcinoma.
Figure 3. Kaplan–Meier survival curves based on PSHCC levels. (A,B) Kaplan–Meier curves depicting differences in OS and DFS between high- and low-PSHCC groups in the training cohort. (C,D) Kaplan–Meier curves showing OS and DFS differences between high- and low-PSHCC groups in the validation cohort. Left panel: OS; right panel: DFS. OS and DFS comparisons between the two groups were performed using a two-sided log-rank test. OS: overall survival; DFS: disease-free survival; PSHCC: pathomics signature of hepatocellular carcinoma.
Cancers 17 03192 g003
Cancers 17 03192 g004
Figure 4. Pathomics nomograms for predicting OS and DFS. (A) Pathomics nomogram for overall survival (OS). (B) Pathomics nomogram for disease-free survival (DFS). The patient’s position on the PSHCC axis is first identified, followed by drawing a line straight upward to the Points axis to determine the corresponding score. This process is repeated for each variable, and the points from all factors are summed. The total score is then located on the Total Points axis, and a line is drawn straight down to determine the patient’s survival probability. OS: overall survival; DFS: disease-free survival; PSHCC: pathomics signature of hepatocellular carcinoma.
Figure 4. Pathomics nomograms for predicting OS and DFS. (A) Pathomics nomogram for overall survival (OS). (B) Pathomics nomogram for disease-free survival (DFS). The patient’s position on the PSHCC axis is first identified, followed by drawing a line straight upward to the Points axis to determine the corresponding score. This process is repeated for each variable, and the points from all factors are summed. The total score is then located on the Total Points axis, and a line is drawn straight down to determine the patient’s survival probability. OS: overall survival; DFS: disease-free survival; PSHCC: pathomics signature of hepatocellular carcinoma.
Cancers 17 03192 g004
Table 1. Characteristics of the patients in the training and validation cohorts.
Table 1. Characteristics of the patients in the training and validation cohorts.
CharacteristicsPatients, No. (%)Patients, No. (%)p
Training Cohort (n = 268)Validation Cohort (n = 121)
Age, year 0.282
≤50111 (41.4)44 (36.4)
>50157 (58.6)77 (63.6)
Sex 0.406
Female30 (11.2)19 (15.7)
Male238 (88.8)102 (84.3)
Etiology of liver diseases 0.546
HBV234 (87.3)104 (86.0)
HCV21 (7.8)10 (8.3)
None or other13 (4.9)7 (5.7)
HBV 0.836
Absent34 (12.7)17 (14.0)
Present234 (87.3)104 (86.0)
Child–Pugh grade
A258 (96.3)115 (95.0) 0.587
B10 (3.7)6 (5.0)
AFP 1 0.567
Normal105 (39.2)44 (36.4)
Elevated163 (60.8)77 (63.6)
Tumor size 0.174
≤5 cm163 (60.8)83 (68.6)
>5 cm105 (39.2)38 (31.4)
Tumor number 0.050
Single216 (80.6)108 (89.3)
Multiple52 (19.4)13 (10.7)
Tumor encapsulation 0.563
Yes154 (57.5)65 (53.7)
No114 (42.5)56 (46.3)
Vascular invasion 0.266
No162 (60.4)81 (66.9)
Yes106 (39.6)40 (33.1)
Tumor differentiation
Grage I193 (72.0)92 (76.0) 0.481
Grade II-III75 (28.0)29 (24.0)
TNM stage 0.053
Stage I134 (50.0)74 (61.2)
Stage II-III134 (50.0)47 (38.8)
Categorical variables are compared using a two-sided χ2 test. Abbreviations: AFP, Alpha-fetoprotein; TNM, tumor-node-metastasis. 1 For AFP, elevated indicates 13.4 μg/L or greater.
Table 2. Univariate and multivariate Cox regression analyses of the PSHCC and clinicopathological characteristics for overall survival and disease-free survival in training cohort.
Table 2. Univariate and multivariate Cox regression analyses of the PSHCC and clinicopathological characteristics for overall survival and disease-free survival in training cohort.
VariablesUnivariate AnalysispMultivariable Analysisp
HR (95%CI)HR (95%CI)
Overall survival
Age0.897 (0.647–1.244)0.516
Sex (Female vs. Male)0.987 (0.587–1.658)0.961
HBV (Absent vs. Present)1.227 (0.730–2.063)0.439
Child–Pugh grade (A vs. B)1.267 (0.560–2.87)0.571
Preoperative AFP level (Normal vs. Elevated)1.529 (1.089–2.148)0.014
Tumor size (≤ 5 cm vs. > 5 cm)1.815 (1.314–2.508)<0.0011.649 (1.170–2.323)0.004
Tumor number (Single vs. Multiple)2.495 (1.743–3.570)<0.001
Tumor differentiation (Grade I vs. Grade II–III)1.972 (1.411–2.756)<0.0011.769 (1.250–2.502)0.001
Tumor encapsulation (Yes vs. No)1.091 (0.788–1.510)0.601
Vascular invasion (No vs. Yes)1.785 (1.291–2.467)<0.001
AJCC TNM stage (Stage I vs. Stage II-III)2.773 (1.977–3.889)<0.0012.292 (1.243–4.228)0.008
PSHCC5.351 (3.766–7.605)<0.0014.368 (3.083–6.188)<0.001
Disease-free survival
Age1.017 (0.744–1.389)0.335
Sex (Female vs. Male)1.299 (0.764–1.389)0.918
HBV (Absent vs. Present)1.188 (0.728–1.940)0.491
Child–Pugh grade (A vs. B)1.074 (0.475–2.428)0.864
Preoperative AFP level (Normal vs. Elevated)1.559 (1.129–2.154)0.007
Tumor size (≤5 cm vs. >5 cm)1.775 (1.305–2.416)<0.0011.691 (1.216–2.351)0.002
Tumor number (Single vs. Multiple)2.299 (1.618–3.268)<0.001
Tumor differentiation (Grade I vs. Grade II–III)1.716 (1.238–2.377)0.0011.548 (1.106–2.166)0.011
Tumor encapsulation (Yes vs. No)1.224 (0.899–1.664)0.198
Vascular invasion (No vs. Yes)1.667 (1.224–2.268)0.001
TNM stage (Stage I vs. Stage II–III)2.229 (1.627–3.053)<0.001
PSHCC3.565 (2.548–4.988)<0.0012.926 (2.080–4.117)<0.001
Association of all variables with prognosis is analyzed using a two-sided Cox proportional hazard regression analysis. Abbreviations: PSHCC, pathomics signature of hepatocellular carcinoma; HR, hazard ratio; CI, confidence interval; AFP, Alpha-fetoprotein. For AFP, elevated indicates 13.4 μg/L or greater.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Wang, T.; Zheng, J.; Guo, L.; Fan, J.; Lu, Y.; Peng, Z.; Zhong, Y.; Zhou, Z.; Chen, E. Use of a Pathomics Signature to Predict the Prognosis of Hepatocellular Carcinoma with Cirrhosis: A Multicentre Retrospective Study. Cancers 2025, 17, 3192. https://doi.org/10.3390/cancers17193192

AMA Style

Wang T, Zheng J, Guo L, Fan J, Lu Y, Peng Z, Zhong Y, Zhou Z, Chen E. Use of a Pathomics Signature to Predict the Prognosis of Hepatocellular Carcinoma with Cirrhosis: A Multicentre Retrospective Study. Cancers. 2025; 17(19):3192. https://doi.org/10.3390/cancers17193192

Chicago/Turabian Style

Wang, Ting, Jixiang Zheng, Lingling Guo, Jiawen Fan, Yubin Lu, Zhen Peng, Yanfeng Zhong, Zhengjun Zhou, and Erbao Chen. 2025. "Use of a Pathomics Signature to Predict the Prognosis of Hepatocellular Carcinoma with Cirrhosis: A Multicentre Retrospective Study" Cancers 17, no. 19: 3192. https://doi.org/10.3390/cancers17193192

APA Style

Wang, T., Zheng, J., Guo, L., Fan, J., Lu, Y., Peng, Z., Zhong, Y., Zhou, Z., & Chen, E. (2025). Use of a Pathomics Signature to Predict the Prognosis of Hepatocellular Carcinoma with Cirrhosis: A Multicentre Retrospective Study. Cancers, 17(19), 3192. https://doi.org/10.3390/cancers17193192

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop