Share：
Share this content in WeChat
X
[Chinese] [PDF] 164 7
• Clinical Article •
A study on the prediction of preoperative risk stratification of hepatocellular carcinoma based on multi-phase MRI radiomics combined with different machine learning models
HAN Xiaobing  ZHANG Chunyu  PENG Weisheng  CAI Huiliang  WANG Chengli  YANG Cuiting  DENG Na  LIU Xuhong  DING Bijiao  WANG Xinda  ZHANG Sizhu  ZHENG Yufeng  ZHANG Yalan  ZENG Yaping  ZHANG Qianying 

DOI:10.12015/issn.1674-8034.2025.08.012.


[Abstract] Objective To explore the value of multiphase magnetic resonance imaging radiomics combined with different machine learning models in predicting risk stratification of hepatocellular carcinoma (HCC).Materials and Methods We retrospectively analyzed clinical and imaging data from ​​a cohort of 120 patients​​ with pathologically confirmed HCC who underwent surgery between January 2020 and December 2024, ​​all meeting predefined inclusion/exclusion criteria. Based on the Edmondson-Steiner (ES) grading system, patients were stratified into two groups: the low-grade group (ES grade Ⅰ and Ⅰ/Ⅱ; n=29) and the high-grade group (ES grade Ⅱ, Ⅱ/Ⅲ, Ⅲ, Ⅲ/Ⅳ, and Ⅳ; n=91). The cohort was subsequently randomly divided in a 7∶3 ratio into a training set (84 cases: 60 high-grade and 24 low-grade) and a validation set (36 cases: 31 high-grade and 5 low-grade). Arterial-phase MRI images were used to delineate the whole-tumor region of interest (ROI) using ITK-SNAP software. ROIs were propagated to portal venous and delayed phases via registration. A total of 3396 radiomic features were extracted using PyRadiomics. Feature selection was performed using Spearman correlation analysis, maximum relevance-minimum redundancy (mRMR), and least absolute shrinkage and selection operator (LASSO) regression. Radiomics models were constructed using five machine learning algorithms: logistic regression (LR), support vector machine (SVM), random forest (RF), naive Bayes (NB), and multilayer perceptron (MLP). The optimal radiomics model was combined with clinical imaging features to develop a combined model. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), calibration curves, and decision curve analysis (DCA).Results A total of 1132 radiomics features were extracted from three contrast-enhanced phases (arterial, portal venous, and delayed). Following dimensionality reduction and feature selection, 8 radiomics features (2 from arterial phase, 3 from portal venous phase, and 3 from delayed phase) were selected to construct radiomics models. Five machine learning algorithms LR, SVM, RF, NB, and MLP demonstrated training and validation sets AUC values of 0.899, 0.897, 0.893, 0.814, 0.876 and 0.865, 0.845, 0.590, 0.723, 0.735, respectively, for predicting HCC pathological grades, indicating that the LR model exhibited the best performance and stability. Univariate and multivariate logistic regression analyses of clinical-radiological features identified age (P = 0.046) and alpha-fetoprotein (AFP) (P = 0.031) as independent predictors of HCC pathological grading. These predictors were subsequently integrated with the radiomics model to develop a combined model, achieving AUC of 0.929 (training set) and 0.884 (validation set). DeLong test revealed significant differences between the clinical model versus the radiomics model and combined model in the training set (P < 0.05), while no statistical distinction was observed between the radiomics and combined models (P > 0.05). In the validation set, no significant differences were found among the three models (P > 0.05). Calibration curves demonstrated closer alignment between predicted and actual probabilities for the combined model in both sets. DCA indicated enhanced net clinical benefit within clinically relevant threshold probabilities. Ultimately, the combined model integrating clinical and radiomics features provided a more accurate prediction of HCC pathological grading.Conclusions The integration of multiphase dynamic contrast-enhanced MRI radiomics with clinical imaging features enables accurate prediction of HCC risk stratification.
[Keywords] hepatocellular carcinoma；magnetic resonance imaging；radiomics；machine learning；pathological grading

HAN Xiaobing1   ZHANG Chunyu2   PENG Weisheng1   CAI Huiliang1   WANG Chengli1   YANG Cuiting1   DENG Na1   LIU Xuhong1   DING Bijiao1   WANG Xinda3   ZHANG Sizhu1   ZHENG Yufeng1   ZHANG Yalan1   ZENG Yaping4   ZHANG Qianying1*  

1 Department of Radiology, the 910th Hospital of the Joint Logistics Support Force of the Chinese People's Liberation Army, Quanzhou 362000, China

2 The Liver Disease Central Laboratory of the 910th Hospital of the Joint Logistics Support Force of the Chinese People's Liberation Army, Quanzhou 362000, China

3 Department of Medical Imaging, Quanzhou First Hospital, Quanzhou 362000, China

4 Department of Medical Imaging, Jinjiang Municipal Hospital, Quanzhou 362200, China

Corresponding author: ZHANG Q Y, E-mail: 836477654@qq.com

Conflicts of interest   None.

Received  2025-04-29
Accepted  2025-08-05
DOI: 10.12015/issn.1674-8034.2025.08.012
DOI:10.12015/issn.1674-8034.2025.08.012.

[1]
Department of Medical Administration of National Health Commission of the People's Republic of China. Guideline for diagnosis and treatment of primary liver cancer (2024 edition) [J]. Chin J Magn Reson Imaging, 2024, 15(6): 1-18. DOI: 10.12015/issn.1674-8034.2024.06.001.
[2]
BRAY F, LAVERSANNE M, SUNG H, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries[J]. CA A Cancer J Clin, 2024, 74(3): 229-263. DOI: 10.3322/caac.21834.
[3]
KAMIYAMA T, NAKANISHI K, YOKOO H, et al. Analysis of the risk factors for early death due to disease recurrence or progression within 1 year after hepatectomy in patients with hepatocellular carcinoma[J/OL]. World J Surg Oncol, 2012, 10: 107 [2025-04-28]. https://pubmed.ncbi.nlm.nih.gov/22697061/. DOI: 10.1186/1477-7819-10-107.
[4]
LIAN S M, CHENG H J, LI H J, et al. Construction of nomogram model based on contrast-enhanced ultrasound parameters to predict the degree of pathological differentiation of hepatocellular carcinoma[J/OL]. Front Oncol, 2025, 15: 1519703 [2025-04-28]. https://pubmed.ncbi.nlm.nih.gov/39931079/. DOI: 10.3389/fonc.2025.1519703.
[5]
EDMONDSON H A, STEINER P E. Primary carcinoma of the liver: a study of 100 cases among 48, 900 necropsies[J]. Cancer, 1954, 7(3): 462-503. DOI: 10.1002/1097-0142(195405)7:3<462::aid-cncr2820070308>3.0.co;2-e.
[6]
MIR I H, JYOTHI K C, THIRUNAVUKKARASU C. The prominence of potential biomarkers in the diagnosis and management of hepatocellular carcinoma: Current scenario and future anticipation[J]. J Cell Biochem, 2022, 123(10): 1607-1623. DOI: 10.1002/jcb.30190.
[7]
CAO S, TANG Y, NIE W. The spiral CT manifestations of the blood supply of small hepatocellular carcinoma: correlation with pathologic findings[J]. Guide China Med, 2012, 10(19): 10-12. DOI: 10.15912/j.cnki.gocm.2012.19.266.
[8]
HUANG K, DONG Z, CAI H S, et al. Imaging biomarkers for well and moderate hepatocellular carcinoma: preoperative magnetic resonance image and histopathological correlation[J/OL]. BMC Cancer, 2019, 19(1): 364 [2025-04-28]. https://pubmed.ncbi.nlm.nih.gov/30999947/. DOI: 10.1186/s12885-019-5574-8.
[9]
MO Z Y, LIAO J Y. The value of signal on hepatobiliary phase of Gd-EOB-DTPA-enhanced MRI in the degree of differentiation of hepatocellular carcinoma[J]. J Med Imag, 2022, 32(8): 1301-1305.
[10]
GUIOT J, VAIDYANATHAN A, DEPREZ L, et al. A review in radiomics: Making personalized medicine a reality via routine imaging[J]. Med Res Rev, 2022, 42(1): 426-440. DOI: 10.1002/med.21846.
[11]
ZHAO B, CAO B Y, XIA T Y, et al. Multiparametric MRI for assessment of the biological invasiveness and prognosis of pancreatic ductal adenocarcinoma in the era of artificial intelligence[J]. J Magn Reson Imag, 2025, 62(1): 9-19. DOI: 10.1002/jmri.29708.
[12]
YAMADA A, KAMAGATA K, HIRATA K, et al. Clinical applications of artificial intelligence in liver imaging[J]. La Radiol Med, 2023, 128(6): 655-667. DOI: 10.1007/s11547-023-01638-1.
[13]
ZHOU W, ZHANG L, WANG K, et al. Malignancy characterization of hepatocellular carcinomas based on texture analysis of contrast-enhanced MR images[J]. J Magn Reson Imaging, 2017, 45(5): 1476-1484. DOI: 10.1002/jmri.25454.
[14]
KESHAVARZ P, NEZAMI N, YAZDANPANAH F, et al. Prediction of treatment response and outcome of transarterial chemoembolization in patients with hepatocellular carcinoma using artificial intelligence: A systematic review of efficacy[J/OL]. Eur J Radiol, 2025, 184: 111948 [2025-04-28]. https://pubmed.ncbi.nlm.nih.gov/39892373/. DOI: 10.1016/j.ejrad.2025.111948.
[15]
Chinese Association of Liver Cancer of Chinese Anti-Cancer Association, Hepatocellular Carcinoma Subcommittee of Chinese Society of Liver Diseases of Chinese Medical Association, Pathological Committee of Chinese Anti-Cancer Association, et al. Guidelines for standardized pathological diagnosis of primary liver cancer (2015 edition)[J]. Chinese Journal of Hepatology, 2015, 23(05): 321-327. DOI: 10.3760/cma.j.issn.1007-3418.2015.05.001.
[16]
SUNG Y S, PARK B, PARK H J, et al. Radiomics and deep learning in liver diseases[J]. J Gastroenterol Hepatol, 2021, 36(3): 561-568. DOI: 10.1111/jgh.15414.
[17]
SONG Y X, YANG X D, LUO Y G, et al. Comparison of logistic regression and machine learning methods for predicting postoperative delirium in elderly patients: a retrospective study[J]. CNS Neurosci Ther, 2023, 29(1): 158-167. DOI: 10.1111/cns.13991.
[18]
ZHU Y Y, WANG J, XUE C, et al. Deep learning and habitat radiomics for the prediction of glioma pathology using multiparametric MRI: A multicenter study[J]. Acad Radiol, 2025, 32(2): 963-975. DOI: 10.1016/j.acra.2024.09.021.
[19]
YU X Y, ZHU D L, GUO H J, et al. DASNet: A convolutional neural network with SE attention mechanism for ccRCC tumor grading[J/OL]. Interdiscip Sci Comput Life Sci, 2025 [2025-04-28]. https://pubmed.ncbi.nlm.nih.gov/40126867/. DOI: 10.1007/s12539-025-00693-8.
[20]
LEFEBVRE T L, et al. Development and validation of multiparametric MRI-based radiomics models for preoperative risk stratification of endometrial cancer[J]. Int J Med Radiol, 2023, 46(1): 104-105. DOI: 10.19300/j.2023.r1106.
[21]
JI S R, CAO L H, GAO J, et al. Proteogenomic characterization of non-functional pancreatic neuroendocrine tumors unravels clinically relevant subgroups[J/OL]. Cancer Cell, 2025, 43(4): 776-796 [2025-04-28]. https://pubmed.ncbi.nlm.nih.gov/40185092/. DOI: 10.1016/j.ccell.2025.03.016.
[22]
PENG Z H, WANG Y N, WU X R, et al. Identifying high gleason score prostate cancer by prostate fluid metabolic fingerprint-based multi-modal recognition[J/OL]. Small Methods, 2024, 8(10): e2301684 [2025-04-28]. https://pubmed.ncbi.nlm.nih.gov/38258603/. DOI: 10.1002/smtd.202301684.
[23]
KUROKAWA S, TANAKA T, YAMAZAKI H, et al. Comparing the CT and MRI findings for canine primary hepatocellular lesions[J/OL]. Vet Rec, 2022, 190(11): e1083 [2025-04-28]. https://pubmed.ncbi.nlm.nih.gov/34751436/. DOI: 10.1002/vetr.1083.
[24]
WU M, YU H J, PANG S W, 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[J/OL]. BMC Med Imaging, 2025, 25(1): 51 [2025-04-28]. https://pubmed.ncbi.nlm.nih.gov/39962429/. DOI: 10.1186/s12880-025-01588-2.
[25]
HAN Y E, CHO Y, KIM M J, et al. Hepatocellular carcinoma pathologic grade prediction using radiomics and machine learning models of gadoxetic acid-enhanced MRI: a two-center study[J]. Abdom Radiol (NY), 2023, 48(1): 244-256. DOI: 10.1007/s00261-022-03679-y.
[26]
HU X J. Study on preoperative prediction of Ck19-positive hepatocellular carcinoma and pathological grading using machine learning radiomics models based on gadoxetic acid disodium-enhanced MRI images [D]. Southern Medical University, 2024. DOI: 10.27003/d.cnki.gojyu.2023.000156.
[27]
ZHANG K, LI W C, XIE S S, et al. Preoperative determination of pathological grades of primary single HCC: development and validation of a scoring model[J]. Abdom Radiol (NY), 2022, 47(10): 3468-3477. DOI: 10.1007/s00261-022-03606-1.
[28]
WU M H, TAN H N, GAO F, et al. Predicting the grade of hepatocellular carcinoma based on non-contrast-enhanced MRI radiomics signature[J]. Eur Radiol, 2019, 29(6): 2802-2811. DOI: 10.1007/s00330-018-5787-2.
[29]
YANG Y, ZHANG S, CUI C, et al. Multiphase MRI-based radiomics for predicting histological grade of hepatocellular carcinoma[J]. J Magn Reson Imaging, 2024, 60(5): 2117-2127. DOI: 10.1002/jmri.29289.
[30]
LIU H F, WANG M, WANG Q, et al. Multiparametric MRI-based intratumoral and peritumoral radiomics for predicting the pathological differentiation of hepatocellular carcinoma[J]. Insights Imaging, 2024, 15(1): 97. DOI: 10.1186/s13244-024-01623-w.
[31]
MAO Y F, WANG J C, ZHU Y, et al. Gd-EOB-DTPA-enhanced MRI radiomic features for predicting histological grade of hepatocellular carcinoma[J]. Hepatobiliary Surg Nutr, 2022, 11(1): 13-24. DOI: 10.21037/hbsn-19-870.
[32]
YANG Y. Prediction of pathological grade, microvascular invasion, and Ki-67 expression in hepatocellular carcinoma based on MR radiomics and machine learning[D]. Army Medical University, 2023. DOI: 10.27001/d.cnki.gtjyu.2022.000024.
[33]
YAKOVCHENKO V, CHANG M F, HERNAEZ R, et al. Access to evaluation for liver transplantation in the veterans health administration[J]. Dig Dis Sci, 2025, 70(2): 552-565. DOI: 10.1007/s10620-024-08717-x.
[34]
NAKAZAWA S, FUKAI K, SANO K, et al. Association of occupational physical activity and sedentary behaviour with the risk of hepatocellular carcinoma: a case-control study based on the Inpatient Clinico-Occupational Database of Rosai Hospital Group[J/OL]. BMJ Open, 2025, 15(3): e092020 [2025-04-28]. https://pubmed.ncbi.nlm.nih.gov/40074261//. DOI: 10.1136/bmjopen-2024-092020.
[35]
ZHAO Y, LIU A L, WU J J, et al. Preoperative prediction of pathologic grade of hepatocellular carcinoma based on radiomics derived from contrast enhanced MRI[J]. Chin J Med Imag, 2021, 29(6): 570-576.
[36]
HU M J, YU Y X, FAN Y F, et al. The predictive value of gadolinium-ethoxybenzyl-diethylenetriamine pentaacetic acid enhanced MRI imaging features combined with quantitative parameters for the pathologic grading of hepatocellular carcinoma[J]. Natl Med J China, 2020, 100(17): 1299-1304. DOI: 10.3760/cma.j.cn112137-20191021-02281.
[37]
MEHRI-KAKAVAND G, MDLETSHE S, WANG A L. A comprehensive review on the application of artificial intelligence for predicting postsurgical recurrence risk in early-stage non-small cell lung cancer using computed tomography, positron emission tomography, and clinical data[J/OL]. J Med Radiat Sci, 2025 [2025-04-28]. https://pubmed.ncbi.nlm.nih.gov/39844750/. DOI: 10.1002/jmrs.860.

PREV Differentiating non-mass breast cancer and non-lactational mastitis based on multi-parameter MRI radiomics
NEXT Study on the value of MRI multiple b-value DWI quantitative parameters in predicting lymphovascular invasion of gastric cancer
  


LINK


Tel & Fax: +8610-67113815    E-mail: editor@cjmri.cn