Multi-task neural network enhanced by data augmentation and ROI optimization for prognosis and MVI prediction in HCC using contrast-enhanced CT

Abstract

Aim: Microvascular invasion (MVI) and recurrence significantly impact hepatocellular carcinoma (HCC) prognosis. This study aims to develop a multi-task deep learning (DL) model using contrast-enhanced computed tomography (CT) to predict preoperative MVI, recurrence-free survival (RFS), and overall survival (OS).

Methods: Preoperative CT scans from 308 patients across five institutions were collected for training and internal validation. A multi-task 3D neural network was trained using a novel augmentation strategy. An independent external cohort (n = 80) from Institution VI was used to rigorously assess generalizability. Model performance was assessed using the concordance index (C-index) for RFS/OS and the F1-score for MVI.

Results: In the training cohort (n = 192), the model achieved an MVI F1 score of 0.739, and C-indices of 0.965 (OS) and 0.869 (RFS). In the internal validation cohort (n = 116), the corresponding values were 0.716, 0.831, and 0.741, respectively. Notably, in the independent external cohort, the model maintained robust performance, with an OS C-index of 0.767, RFS C-index of 0.733, and MVI F1 score of 0.708. Class activation maps confirmed that the model focused on clinically relevant liver regions.

Conclusion: Our interpretable multi-task DL model demonstrates robust predictive capabilities for RFS, OS, and MVI in HCC patients across multiple centers, offering a non-invasive tool to enhance clinical decision-making.

Keywords

Hepatocellular carcinoma / deep learning / neural network / computed tomography / prognosis

References

[1] Qiu S,Yang Z.Trends in hepatocellular carcinoma mortality rates in the US and projections through 2040.JAMA Netw Open2024;7:e2445525 PMCID:PMC11574689

[2] Koshy A.Evolving global etiology of hepatocellular carcinoma (HCC): insights and trends for 2024.J Clin Exp Hepatol2025;15:102406 PMCID:PMC11426038

[3] Singal AG,Llovet JM.Global trends in hepatocellular carcinoma epidemiology: implications for screening, prevention and therapy.Nat Rev Clin Oncol2023;20:864-84 PMCID:12996431

[4] Yu H,Mo X.MRI-based 2.5D deep learning and radiomics effectively predicted microvascular invasion and Ki-67 expression in hepatocellular carcinoma.PLoS One2025;20:e0336579 PMCID:PMC12617848

[5] Xia TY,Meng XP.Predicting microvascular invasion in hepatocellular carcinoma using CT-based radiomics model.Radiology2023;307:e222729

[6] Barcena-Varela M,Lujambio A.Precision models in hepatocellular carcinoma.Nat Rev Gastroenterol Hepatol2025;22:191-205

[7] Wang Y,Tian Y.Construction of an artificially intelligent model for accurate detection of HCC by integrating clinical, radiological, and peripheral immunological features.Int J Surg2025;111:2942-52 PMCID:12175786

[8] Jiang H,Chen Y.Development of a model including MRI features for predicting advanced-stage recurrence of hepatocellular carcinoma after liver resection.Radiology2023;309:e230527

[9] Xia T,Li B.MRI-based radiomics and deep learning in biological characteristics and prognosis of hepatocellular carcinoma: opportunities and challenges.J Magn Reson Imaging2024;59:767-83

[10] Su R,Yan L.Early screening, diagnosis and recurrence monitoring of hepatocellular carcinoma in patients with chronic hepatitis B based on serum N-glycomics analysis: a cohort study.Hepatology2026;83:40-56 PMCID:12700682

[11] Yu B.Biomarker discovery in hepatocellular carcinoma (HCC) for personalized treatment and enhanced prognosis.Cytokine Growth Factor Rev2024;79:29-38

[12] An C,Li W.An automated machine-learning model for prognostic risk stratification of intermediate-stage hepatocellular carcinoma after transarterial chemoembolization.Int J Surg2025;111:6200-10 PMCID:12430918

[13] van der Velden BHM, Kuijf HJ, Gilhuijs KGA, Viergever MA. Explainable artificial intelligence (XAI) in deep learning-based medical image analysis.Med Image Anal2022;79:102470 PMCID:12653247

[14] Chen X,Zhang K.Recent advances and clinical applications of deep learning in medical image analysis.Med Image Anal2022;79:102444 PMCID:PMC9156578

[15] Stahlschmidt SR,Synnergren J.Multimodal deep learning for biomedical data fusion: a review.Brief Bioinform2022;23:bbab569 PMCID:PMC8921642

[16] Davri A,Kanavos T.Deep learning for lung cancer diagnosis, prognosis and prediction using histological and cytological images: a systematic review.Cancers2023;15:3981 PMCID:PMC10417369

[17] Yan M,Zhang B.Deep learning nomogram based on Gd-EOB-DTPA MRI for predicting early recurrence in hepatocellular carcinoma after hepatectomy.Eur Radiol2023;33:4949-61 PMCID:PMC10289921

[18] Wang F,Chen QQ.Multitask deep learning for prediction of microvascular invasion and recurrence-free survival in hepatocellular carcinoma based on MRI images.Liver Int2024;44:1351-62

[19] Lu J, Yao J, Zhang J, et al. SOFT: softmax-free transformer with linear complexity. arXiv 2022;arXiv:2110.11945. Available from https://arxiv.org/abs/2110.11945. [accessed 17 April 2026].

[20] Golovneva O, Wang T, Weston J, Sukhbaatar S. Contextual position encoding: learning to count what’s important. arXiv 2024;arXiv:2405.18719. Available from https://arxiv.org/abs/2405.18719. [accessed 17 April 2026].

[21] Hwang YJ,Lee Y,Lee DH.Classification of microvascular invasion of hepatocellular carcinoma: correlation with prognosis and magnetic resonance imaging.Clin Mol Hepatol2023;29:733-46 PMCID:PMC10366800

[22] Altaf A,Dar A.Artificial intelligence-based model for the recurrence of hepatocellular carcinoma after liver transplantation.Surgery2024;176:1500-6

[23] Song D,Wang W.Using deep learning to predict microvascular invasion in hepatocellular carcinoma based on dynamic contrast-enhanced MRI combined with clinical parameters.J Cancer Res Clin Oncol2021;147:3757-67 PMCID:PMC11801944

[24] Li J,Xu X.Preoperative prediction and risk assessment of microvascular invasion in hepatocellular carcinoma.Crit Rev Oncol Hematol2023;190:104107

[25] Liu Z,Zhang W.Deep learning for prediction of hepatocellular carcinoma recurrence after resection or liver transplantation: a discovery and validation study.Hepatol Int2022;16:577-89 PMCID:PMC9174321

[26] Zhang K,Zhang L.Gadoxetic acid-enhanced MRI scoring model to predict pathologic features of hepatocellular carcinoma.Radiology2025;314:e233229