Indeterminate nodule-like arterial phase hyperenhancement in the liver on CT: differentiation of hepatocellular carcinomas from vascular pseudolesions using a CT radiomics approach

Introduction

Non-invasive imaging diagnosis of hepatocellular carcinoma (HCC) is based on the radiological hallmark of non-rim arterial phase hyperenhancement (APHE) followed by washout appearance. However, HCCs with atypical enhancement patterns are not uncommon (1). Some small HCCs may exhibit nodular APHE yet appear iso-enhancing on subsequent dynamic phases—a feature defined as nodule-like APHE (NAPH) in Liver Imaging Reporting and Data System (LI-RADS) v2018 (2). Importantly, NAPH may also represent benign vascular pseudolesions arising from arterioportal shunt (APS) in cirrhotic livers (3). Although APS-related pseudolesions typically show wedge-shaped APHE, a subset can present as nodular lesions (i.e., NAPH) (4). Therefore, when NAPH is identified in high-risk patients, both HCC and benign vascular pseudolesions should be considered. Differentiating between these entities is crucial because they require markedly different management. Nevertheless, this distinction can be challenging when they present as NAPHs on computed tomography (CT) (5). LI-RADS v2018 classifies NAPHs into LI-RADS category (LR)-2 (probably benign), LR-3 (indeterminate), and LR-4 (probably HCC) based on HCC likelihood (2). However, the diagnostic accuracy of this classification remains unproven. Although the hepatobiliary phase (HBP) of gadoxetic acid-enhanced magnetic resonance imaging (MRI) can aid differentiation (6,7), its cost and limited availability underscore the importance of CT-based risk stratification.

Radiomics is an emerging technique that extracts quantitative data from medical imaging, capturing features often imperceptible to the human eye (8). These features include shape metrics, first-order statistics based on gray-scale values, and second-order texture features reflecting gray-level inter-relationships and spatial orientation (9). Such metrics can indicate tissue heterogeneity and subtle biological variations. As a novel imaging biomarker, radiomics has been actively investigated in oncology. In HCC, studies have shown promising results in distinguishing HCC from other tumors (10,11), predicting pathologic features such as histologic grade and microvascular invasion (12,13), and forecasting patient outcomes (14). In this context, given the different vascular mechanisms underlying APHE—neovascularization in HCC vs. APS in vascular pseudolesions—radiomics analysis could potentially reveal differences between NAPHs.

Therefore, this study aimed to investigate the feasibility of CT radiomics for distinguishing HCCs from vascular pseudolesions in indeterminate NAPH observations detected in HCC high-risk patients. We present this article in accordance with the TRIPOD reporting checklist (available at https://jgo.amegroups.com/article/view/10.21037/jgo-2025-aw-919/rc).

Methods

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Review Board of Seoul National University Hospital (No. 2010-026-1161) and individual consent for this retrospective analysis was waived.

Study population

Between December 2019 and August 2020, we retrospectively screened patients at high risk for HCC who underwent four-phase liver CT with arterial phase acquisition at 100 kVp and gadoxetic acid-enhanced liver MRI within 4 months to identify CT-detected NAPHs with MRI-based final diagnoses. Filtered back projection reconstructions were also available for these CT scans. These criteria for CT tube voltage and reconstruction were applied because this study performed arterial phase radiomics analysis, and consistent acquisition and reconstruction are essential to minimize variability in radiomics features (15,16). Detailed CT scan parameters are summarized in Appendix 1.

NAPH identification was performed in a two-step review process: an initial unblinded preliminary review by the coordinating radiologist (I.J.), who incorporated CT and MRI findings to identify candidate lesions on CT and assign an MRI-based diagnosis for each lesion, followed by a second step in which three radiologists (J.H.K., S.K., and J.S.C.), blinded to the final diagnoses, independently reviewed all CT phases and confirmed NAPH by consensus. Conventional imaging features, including lesion size (mm) and lesion location (categorized as subcapsular or non-subcapsular), were also assessed. Although NAPH is commonly defined as lesions smaller than 10 mm (2), we conservatively used a 20 mm cutoff to account for measurement variability and to include larger lesions of potential clinical significance.

A flow diagram of the overall study workflow is provided in Figure 1, and further details of the imaging review process are described in Appendix 2.

Figure 1 Flowchart presenting the overall study workflow. CT, computed tomography; FBP, filtered back projection; HCC, hepatocellular carcinoma; MRI, magnetic resonance imaging; NAPH, nodule-like arterial phase hyperenhancement.

MRI diagnosis of NAPHs

Gadoxetic acid-enhanced liver MRI served as the reference standard for CT-detected NAPHs. Full MRI sequences were reviewed to evaluate dynamic enhancement patterns and ancillary features. HCC was diagnosed based on the Korean Liver Cancer Association and National Cancer Center (KLCA-NCC) v2022 (17), with the exception that sub-centimeter lesions showing hallmark imaging features were also classified as HCC in this study. Vascular pseudolesions were defined as lesions with APHE that exhibited iso-intensity on HBP, T2-weighted imaging, and high b-value diffusion-weighted imaging (6,7).

Radiomics feature extraction

Digital Imaging and Communications in Medicine (DICOM) files of CT arterial phase images reconstructed by filtered back projection from the final study population were loaded into a radiomics analysis software program (syngo.via Frontier, RADIOMICS prototype, Siemens Healthineers, Forchheim, Germany). NAPH lesions were then selected within these images for feature extraction using algorithms implemented in the PyRadiomics library (v3.0.1; https://github.com/Radiomics/pyradiomics). Each NAPH lesion was semi-automatically segmented by an abdominal radiologist (S.P.), blinded to the MRI diagnosis (Figure 2). The reproducibility of three-dimensional (3D) radiomics features using this software and segmentation method has been reported in prior research on intrahepatic cholangiocarcinoma (18). In the present study, given the small size of the NAPH lesions, two-dimensional (2D) segmentation was performed on the axial slice depicting the largest lesion diameter. Radiomics analysis was performed on CT images resampled to isotropic 1×1×1 mm³ voxels via linear interpolation. For each segmented lesion, 1,226 radiomics features—including shape-based, first-order, and second-order features—were extracted using a Laplacian of Gaussian filter at five sigma levels, one-level wavelet decomposition (yielding eight derived images), and mathematical transforms (square, square root, logarithm, exponential).

Figure 2 Illustration of the semi-automated 2D segmentation of NAPHs. For NAPH in the liver, the radiologist manually draws a line [red line in (A)] visually estimating the longest dimension of the tumor. Subsequently, the software automatically segments the entire tumor [yellow circle in (B)], with manual adjustments performed if necessary. 2D, two-dimensional; NAPH, nodule-like arterial phase hyperenhancement.

CT radiomics model

A CT radiomics model was developed to differentiate HCC from vascular pseudolesions among NAPHs. The lesions were divided into training and validation cohorts based on CT examination dates (December 2019–May 2020 vs. June–August 2020). In the training set, radiomics features were reduced to ten using the classic minimum redundancy maximum relevance (mRMR) algorithm, based on R² difference. The mRMR algorithm selects features that are highly relevant to the target classes while minimizing redundancy. From those ten, four features were then selected via maximization of R²—to build the model. Because some patients contributed multiple lesions, observations were clustered within patients. We therefore refitted the logistic model using a generalized estimating equation (GEE) with patient ID as the clustering variable and an exchangeable working correlation structure. Robust (sandwich) standard errors were used for inference.

For each NAPH, the model generated a probability score indicating the likelihood of HCC vs. vascular pseudolesion. In the training set, the optimal threshold for these probability scores was determined using the Youden index to maximize the sum of sensitivity and specificity. The model was then applied to the independent validation set to assess its diagnostic performance.

Statistical analysis

Categorical variables were compared between NAPHs diagnosed as HCC vs. vascular pseudolesion using the χ² test or Fisher’s exact test, and continuous variables using the Student t-test. Model performance was evaluated using predicted probabilities from the GEE model, and 95% confidence intervals (CIs) were estimated using patient-level (cluster) bootstrap resampling to account for within-patient correlation. Radiomics model performance was assessed by the area under the receiver operating characteristic (ROC) curve (AUC) of the predicted probability, with sensitivity, specificity, and accuracy calculated at the Youden index-derived threshold. In addition, model performance under class imbalance was evaluated using the precision-recall AUC (PR-AUC). Positive and negative likelihood ratios (LR⁺ and LR⁻) were also calculated at the same operating threshold to facilitate clinical interpretation. A comparison of AUCs between the radiomics model and conventional imaging features—lesion size and location—was performed, and the added value of integrating significant conventional features with radiomics was evaluated using DeLong’s test. A two-sided P value of less than 0.05 was considered statistically significant. Model calibration was assessed by regressing the observed outcome on the logit of the predicted probabilities to estimate the calibration intercept and slope, and by visual inspection of calibration plots. All statistical analyses were performed using SPSS version 23 (IBM software), MedCalc (version 23.0.2), R version 4.5.2 (R Foundation for Statistical Computing, Vienna, Austria), including the geepack package for GEEs, and Python for additional performance evaluation and visualization.

Results

Patient and lesion characteristics

A total of 259 NAPHs from 181 patients were finally included; MRI-based evaluation classified 51 as HCC and 208 as vascular pseudolesions. Patient and lesion characteristics are summarized in Table 1. In the training cohort (204 NAPHs), 35 were classified as HCCs and 169 as pseudolesions, while in the validation cohort (55 NAPHs), 16 were HCCs and 39 were pseudolesions. Mean lesion size was 9.6 mm in the training cohort and 9.5 mm in the validation cohort, with no significant size differences between HCCs and pseudolesions in either set (P=0.55 and P=0.95). Pseudolesions occurred significantly more often in subcapsular locations than HCCs in the training set (P=0.004) and showed a similar trend in the validation set (P=0.06). Chronic hepatitis B was the predominant underlying liver disease in both cohorts.

Performance of the radiomics model

Four 2D arterial phase radiomics features were finally selected to develop the model for differentiating HCC from vascular pseudolesions among NAPHs: Square_glcm_SumEntropy, Wavelet-HLL_ngtdm_Busyness, Exponential_glrlm_LowGrayLevelRunEmphasis, and Logarithm_glszm_SizeZoneNonUniformityNormalized. The developed radiomics model demonstrated an AUC of 0.794 (95% CI: 0.710–0.874) in the training cohort and 0.688 (95% CI: 0.528–0.842) in the validation cohort (Table 2), both higher than the AUCs of lesion size [0.547 (95% CI: 0.476–0.617), P<0.001; and 0.502 (95% CI: 0.364–0.640), P=0.13, respectively] or AUCs of lesion location [0.633 (95% CI: 0.563–0.699), P=0.008; and 0.639 (95% CI: 0.498–0.764), P=0.29, respectively]. The PR-AUC was 0.504 (95% CI: 0.341–0.682) in the training set and 0.439 (95% CI: 0.255–0.711) in the validation set (Figure S2). Using the Youden index-derived probability threshold, sensitivity, specificity, and accuracy of the radiomics model were 60.0% (21/35), 87.0% (147/169), 82.4% (168/204), LR⁺ 4.61, and LR⁻ 0.46 in the training cohort, and 43.8% (7/16), 76.9% (30/39), 67.3% (37/55), LR⁺ 1.90, and LR⁻ 0.73 in the validation cohort, respectively (Table 2). In the validation cohort, the radiomics-only model had a calibration intercept of 0.853 and a slope of 0.502. The full model specification, including all regression coefficients and the intercept, is provided in Table S1. In addition, a calibration plot was presented in Figure S1.

Performance of the combined radiomics-location model

Because lesion size did not differ between diagnoses, but lesion location did, we developed a combined model that incorporated lesion location (subcapsular vs. non-subcapsular) with radiomics features. This approach increased the AUC to 0.822 (95% CI: 0.744–0.899) in the training cohort and 0.720 (95% CI: 0.555–0.865) in the validation cohort compared to the radiomics-only model, although the improvements did not reach statistical significance (P=0.27 and P=0.51, respectively) (Figure 3 and Table 2). In the validation set, the model achieved a PR-AUC of 0.521 (95% CI: 0.312–0.783) (Figure S2). The corresponding average precision in the training set was 0.611 (95% CI: 0.468–0.745). In the validation set, sensitivity, specificity and accuracy were 68.8% (11/16), 66.7% (26/39), and 67.3% (37/55), corresponding to LR⁺ of 2.06 (95% CI: 1.23–3.64) and LR⁻ of 0.47 (95% CI: 0.16–0.84), using the prespecified operating threshold determined in the training set by the Youden index. In the validation cohort, the combined model showed a calibration intercept of 0.839 and a slope of 0.792. The full model specification, including all regression coefficients and the intercept, is provided in Table S2. In addition, a calibration plot was presented in Figure S1.

Figure 3 ROC curves of CT-based models differentiating HCC from vascular pseudolesions among NAPHs. In the training set (A), the AUC for the CT radiomics model was 0.794, while the combined model incorporating CT radiomics and lesion location achieved an AUC of 0.822. The lesion size model showed a lower AUC of 0.547. In the validation set (B), the AUCs of the CT radiomics and combined models were 0.688 and 0.720, respectively. The lesion size model demonstrated an AUC of 0.502. AUC, area under the ROC curve; CT, computed tomography; HCC, hepatocellular carcinoma; NAPH, nodule-like arterial phase hyperenhancement; ROC, receiver operating characteristic.

Discussion

This study evaluated the potential utility of CT-based radiomics in differentiating HCC from benign vascular pseudolesions among NAPH lesions, a subgroup of hepatic observations that often pose diagnostic challenges in patients at high risk for HCC. Using a GEE model built on four 2D radiomics features extracted from arterial phase images, we achieved an AUC of 0.794 in the training cohort and 0.688 in the independent validation cohort. These findings suggest the feasibility of using radiomics to provide additional information regarding HCC risk for lesions identified as NAPHs. Notably, conventional imaging features such as lesion size or location alone demonstrated limited diagnostic performance, underscoring the challenges inherent in evaluating these lesions. While the addition of lesion location to the radiomics model did not result in statistically significant improvements, it showed a trend toward increased diagnostic accuracy (AUC of 0.822 in the training cohort and 0.720 in the validation cohort), suggesting that combining radiomics with conventional imaging findings may provide incremental benefit. In addition, the combined model demonstrated a higher PR-AUC than the radiomics model (0.521 vs. 0.439) under class imbalance and improved calibration, indicating reduced overfitting compared with the radiomics model.

Our study’s findings contribute to the evolving landscape of liver lesion characterization by specifically addressing the diagnostic challenges posed by NAPHs. While the broader utility of radiomics in differentiating HCC from other liver tumors has been well-documented (10,19), our research uniquely focuses on distinguishing HCC from benign vascular pseudolesions within this indeterminate NAPH category. Interestingly, all four features selected in our radiomics model were second-order texture features, reflecting gray-level spatial relationships within the lesions (20,21). These features quantify aspects of lesion heterogeneity that are challenging to assess visually on conventional imaging. Although the specific histopathological correlations of these radiomics metrics remain incompletely understood, they are presumed to reflect underlying variations in tissue composition and vascularity. By capturing such subtle intralesional heterogeneity, radiomics analysis may help improve differentiation between HCC and vascular pseudolesions. Additionally, vascular pseudolesions were more frequently observed in subcapsular regions compared to HCCs, consistent with prior reports (22), highlighting the value of combining radiomics features with clinically relevant contextual information.

It is important to contextualize our findings within real-world clinical practice for NAPH management. When NAPHs are identified, management approaches vary considerably based on clinical context. While additional imaging may be performed in some cases, clinicians often employ a multifactorial decision-making process considering lesion size, patient risk factors, tumor markers, and clinical context, with surveillance commonly used for smaller lesions. As demonstrated in studies of LR-3 and LR-4 lesions, clinical factors such as concurrent LR-5 (definitely HCC) observations (23) and elevated alpha-fetoprotein levels (24) can influence the probability of HCC in indeterminate lesions, suggesting that similar multifactorial considerations may be relevant for NAPH evaluation. Our study population likely represents a higher-risk subset, as inclusion required both CT and MRI within 4 months, potentially explaining the relatively high proportion of HCC cases compared to broader clinical practice. This clinical reality suggests that the true value of radiomics may lie in its integration with other clinical parameters to enhance comprehensive risk assessment frameworks (25,26). In these contexts, radiomics-derived probability scores are not intended to function as a stand-alone diagnostic or rule-out tool, but may serve as additional evidence to support clinical decision-making. Specifically, such probability estimates could be used for risk stratification among patients with indeterminate imaging findings, helping identify which patients might benefit from expedited advanced imaging (prioritization for MRI referral) or closer surveillance intervals. Gadoxetic acid-enhanced liver MRI frequently serves as a diagnostic modality for the characterization of indeterminate hepatic lesions identified on CT, and is instrumental in guiding subsequent clinical management decisions. In contrast, the proposed CT-based radiomics approach is not intended to replace MRI-based diagnostic benchmarks, but rather to complement existing diagnostic pathways in clinical scenarios where MRI is not immediately available or routinely performed. CT-based radiomics may provide incremental quantitative information beyond visual assessment, enabling risk stratification among patients with indeterminate findings on CT and helping to identify those who may benefit most from subsequent MRI examination. Therefore, the clinical value of the proposed CT-based model lies not in achieving diagnostic performance comparable to MRI, but in facilitating more efficient triage and downstream imaging decision-making within existing workflows. However, further research is needed to evaluate these integrated approaches and their impact on clinical outcomes.

Several limitations must be acknowledged. This was a retrospective single-center study based on a specific CT protocol (100 kVp with filtered back projection reconstruction), which was deliberately chosen to minimize technical variability and ensure consistent radiomics feature extraction. While this protocol-specific single-center design may limit generalizability, it allowed us to demonstrate the feasibility of CT radiomics for NAPH differentiation under controlled conditions. However, to improve generalizability, external multicenter validation across heterogeneous scanners/vendors and reconstruction methods is warranted. Our study population likely included a higher proportion of high-risk cases due to the requirement for both CT and MRI within 4 months, and the sample size of HCCs was relatively small, especially in the validation set. However, this selection was necessary to establish a reference standard for model development and validation. In addition, the reference standard was based on MRI findings rather than pathologic results. Although imaging-based diagnosis is widely used in at-risk patients, misclassification remains possible, especially for very small observations. In particular, many diagnostic algorithms (2,17) require a minimum size threshold (commonly ≥10 mm) for “definite HCC” categorization, and lesions <10 mm may be difficult to classify reliably by imaging alone and are frequently managed with short-interval surveillance. However, histologic confirmation is rarely pursued for lesions suspected to be vascular pseudolesions, making our MRI-based approach clinically realistic. Finally, while other differential diagnoses, such as hemangiomas (27), may present with similar imaging characteristics, we focused on vascular pseudolesions as they represent a more common and challenging clinical scenario in distinguishing from HCC among NAPHs.

Conclusions

In conclusion, for indeterminate NAPHs, CT radiomics analysis on arterial phase imaging may help differentiate HCC from vascular pseudolesions, which can impact the subsequent diagnostic work-up and management strategy.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://jgo.amegroups.com/article/view/10.21037/jgo-2025-aw-919/rc

Data Sharing Statement: Available at https://jgo.amegroups.com/article/view/10.21037/jgo-2025-aw-919/dss

Peer Review File: Available at https://jgo.amegroups.com/article/view/10.21037/jgo-2025-aw-919/prf

Funding: This study was supported by the New Faculty Startup Fund from Seoul National University and the Seoul National University Hospital Research Fund (No. 03-2021-2190).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jgo.amegroups.com/article/view/10.21037/jgo-2025-aw-919/coif). I.J. reports research support from the New Faculty Startup Fund from Seoul National University and the Seoul National University Hospital Research Fund (No. 03-2021-2190), with all payments made to the institution. J.M.L. reports research grants from Bayer Healthcare, Canon Healthcare, Philips Healthcare, GE Healthcare, CMS, Guerbet, Samsung Medison, Starmed, Medical IP, Clarify, Siemens Healthineers, and Bracco; consulting fees from Bayer Healthcare, Siemens Healthineers, Samsung Medison, Guerbet, and Philips Healthcare; and lecture honoraria from Samsung Medison, GE Healthcare, and Bayer Healthcare. These financial relationships are unrelated to the present study. The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Review Board of Seoul National University Hospital (No. 2010-026-1161) and individual consent for this retrospective analysis was waived.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Yoon SH, Lee JM, So YH, et al. Multiphasic MDCT enhancement pattern of hepatocellular carcinoma smaller than 3 cm in diameter: tumor size and cellular differentiation. AJR Am J Roentgenol 2009;193:W482-9. [Crossref] [PubMed]
2. American College of Radiology. Liver Imaging Reporting and Data System. Accessed April 3rd, 2026. Available online: https://www.acr.org/Clinical-Resources/Clinical-Tools-and-Reference/Reporting-and-Data-Systems/LI-RADS
3. Choi BI, Lee KH, Han JK, et al. Hepatic arterioportal shunts: dynamic CT and MR features. Korean J Radiol 2002;3:1-15. [Crossref] [PubMed]
4. Ahn JH, Yu JS, Hwang SH, et al. Nontumorous arterioportal shunts in the liver: CT and MRI findings considering mechanisms and fate. Eur Radiol 2010;20:385-94. [Crossref] [PubMed]
5. Hwang SH, Yu JS, Kim KW, et al. Small hypervascular enhancing lesions on arterial phase images of multiphase dynamic computed tomography in cirrhotic liver: fate and implications. J Comput Assist Tomogr 2008;32:39-45. [Crossref] [PubMed]
6. Sun HY, Lee JM, Shin CI, et al. Gadoxetic acid-enhanced magnetic resonance imaging for differentiating small hepatocellular carcinomas (< or =2 cm in diameter) from arterial enhancing pseudolesions: special emphasis on hepatobiliary phase imaging. Invest Radiol 2010;45:96-103. [Crossref] [PubMed]
7. Motosugi U, Ichikawa T, Sou H, et al. Distinguishing hypervascular pseudolesions of the liver from hypervascular hepatocellular carcinomas with gadoxetic acid-enhanced MR imaging. Radiology 2010;256:151-8. [Crossref] [PubMed]
8. Gillies RJ, Kinahan PE, Hricak H. Radiomics: Images Are More than Pictures, They Are Data. Radiology 2016;278:563-77. [Crossref] [PubMed]
9. Varghese BA, Cen SY, Hwang DH, et al. Texture Analysis of Imaging: What Radiologists Need to Know. AJR Am J Roentgenol 2019;212:520-8. [Crossref] [PubMed]
10. Mokrane FZ, Lu L, Vavasseur A, et al. Radiomics machine-learning signature for diagnosis of hepatocellular carcinoma in cirrhotic patients with indeterminate liver nodules. Eur Radiol 2020;30:558-70. [Crossref] [PubMed]
11. Liu X, Khalvati F, Namdar K, et al. Can machine learning radiomics provide pre-operative differentiation of combined hepatocellular cholangiocarcinoma from hepatocellular carcinoma and cholangiocarcinoma to inform optimal treatment planning? Eur Radiol 2021;31:244-55. [Crossref] [PubMed]
12. Oh J, Lee JM, Park J, et al. Hepatocellular Carcinoma: Texture Analysis of Preoperative Computed Tomography Images Can Provide Markers of Tumor Grade and Disease-Free Survival. Korean J Radiol 2019;20:569-79. [Crossref] [PubMed]
13. Xu X, Zhang HL, Liu QP, et al. Radiomic analysis of contrast-enhanced CT predicts microvascular invasion and outcome in hepatocellular carcinoma. J Hepatol 2019;70:1133-44. [Crossref] [PubMed]
14. Chen M, Cao J, Hu J, et al. Clinical-Radiomic Analysis for Pretreatment Prediction of Objective Response to First Transarterial Chemoembolization in Hepatocellular Carcinoma. Liver Cancer 2021;10:38-51. [Crossref] [PubMed]
15. Mackin D, Ger R, Dodge C, et al. Effect of tube current on computed tomography radiomic features. Sci Rep 2018;8:2354. [Crossref] [PubMed]
16. Sung P, Lee JM, Joo I, et al. Evaluation of the Impact of Iterative Reconstruction Algorithms on Computed Tomography Texture Features of the Liver Parenchyma Using the Filtration-Histogram Method. Korean J Radiol 2019;20:558-68. [Crossref] [PubMed]
17. 2022 KLCA-NCC Korea Practice Guidelines for the Management of Hepatocellular Carcinoma. Korean J Radiol 2022;23:1126-240. [Crossref] [PubMed]
18. Park S, Lee JM, Park J, et al. Volumetric CT Texture Analysis of Intrahepatic Mass-Forming Cholangiocarcinoma for the Prediction of Postoperative Outcomes: Fully Automatic Tumor Segmentation Versus Semi-Automatic Segmentation. Korean J Radiol 2021;22:1797-808. [Crossref] [PubMed]
19. Nie P, Yang G, Guo J, et al. A CT-based radiomics nomogram for differentiation of focal nodular hyperplasia from hepatocellular carcinoma in the non-cirrhotic liver. Cancer Imaging 2020;20:20. [Crossref] [PubMed]
20. Lubner MG, Smith AD, Sandrasegaran K, et al. CT Texture Analysis: Definitions, Applications, Biologic Correlates, and Challenges. Radiographics 2017;37:1483-503. [Crossref] [PubMed]
21. Feng M, Zhang M, Liu Y, et al. Texture analysis of MR images to identify the differentiated degree in hepatocellular carcinoma: a retrospective study. BMC Cancer 2020;20:611. [Crossref] [PubMed]
22. Wang Q, Koniaris LG, Milgrom DP, et al. CT and MRI imaging and interpretation of hepatic arterioportal shunts. Transl Gastroenterol Hepatol 2019;4:34. [Crossref] [PubMed]
23. Abedrabbo N, Lerner E, Lam E, et al. Is concurrent LR-5 associated with a higher rate of hepatocellular carcinoma in LR-3 or LR-4 observations? An individual participant data meta-analysis. Abdom Radiol (NY) 2025;50:1533-46. [Crossref] [PubMed]
24. Chen J, Chen H, Zheng D, et al. LI-RADS category 3, 4, and M observations: a multiple parameters diagnostic model for hepatocellular carcinoma. Acta Radiol 2023;64:2977-86. [Crossref] [PubMed]
25. Le MHN, Kha HQ, Tran NM, et al. Radiomics in liver research: A paradigm shift in disease detection and staging. Eur J Radiol Artif Intell 2025;2:100016. [Crossref]
26. Wu M, Yu H, Pang S, et al. Application of CT-based radiomics combined with laboratory tests such as AFP and PIVKA-II in preoperative prediction of pathologic grade of hepatocellular carcinoma. BMC Med Imaging 2025;25:51. [Crossref] [PubMed]
27. Jha RC, Mitchell DG, Weinreb JC, et al. LI-RADS categorization of benign and likely benign findings in patients at risk of hepatocellular carcinoma: a pictorial atlas. AJR Am J Roentgenol 2014;203:W48-69. [Crossref] [PubMed]