Radiomics and Machine Learning in Hepatocellular Carcinoma

Radiomics harnesses high-throughput extraction of quantitative image features—such as texture, shape and intensity—from standard clinical scans to characterise tumour phenotype. When combined with machine-learning algorithms, these data permit the development of predictive models for diagnosis, prognosis and therapeutic response in hepatocellular carcinoma (HCC). By analysing large cohorts of imaging data alongside clinical and molecular parameters, radiomic-machine-learning pipelines can identify patterns imperceptible to the human eye and stratify patients according to individual risk. Applications range from noninvasive grading of tumour heterogeneity and vascular invasion to forecasting early recurrence after surgery or ablation, and predicting benefit from systemic therapies, including immune checkpoint inhibitors. The global burden of HCC, particularly in regions with high hepatitis prevalence, underlines the urgency of robust, generalisable tools that can guide personalised management. Advances in feature-selection methods, such as LASSO regression, and ensemble classifiers, such as random forests, have enhanced model stability. Integration of radiomic signatures with clinicopathological factors continues to improve clinical utility and offers a promising avenue for precision hepatology.

Research from Nature Portfolio

Quantitative multiparametric MRI has been employed to characterise intratumoural heterogeneity in HCC and to link imaging metrics with detailed histopathology and gene-expression profiles. Analyses of diffusion-weighted, oxygenation-sensitive and dynamic contrast-enhanced MRI revealed that histogram-based heterogeneity parameters provide complementary insights into hypoxic markers and pathways governing stemness and immune evasion, supporting noninvasive molecular phenotyping.

Investigation of non-contrast CT texture features demonstrated that simple statistical measures—such as entropy, standard deviation and skewness—predict both overall and disease-free survival independently of established clinical and pathological factors. These CT-derived heterogeneity metrics improved prognostic stratification and underline the value of routinely acquired scans for risk assessment.

Radiomics and Machine Learning in Hepatocellular Carcinoma publication trend

The graph below shows the total number of articles in radiomics and machine learning in hepatocellular carcinoma across all publications each year (not limited to Nature Index journals).

Technical terms

Radiomics: Extraction of large-scale quantitative features from medical images for subsequent analysis.

Machine learning: Computational methods that enable algorithms to learn patterns from data and make predictions or decisions.

Texture feature: Quantitative descriptor of spatial variation in pixel intensities reflecting tissue heterogeneity.

LASSO regression: Regularisation technique that selects a subset of predictive features by imposing a penalty on absolute coefficients.

Random forest: Ensemble learning method that constructs multiple decision trees to improve predictive accuracy and control overfitting.

Radiomic signature: Combination of selected radiomic features that yields a prognostic or predictive score.

Area under the curve (AUC): Measure of a model’s ability to distinguish between outcome classes, with 1.0 being perfect discrimination.

References

1. Radiomics-based biomarker for PD-1 status and prognosis analysis in patients with HCC. Frontiers in Immunology (2025).
2. MRI-based radiomics model for preoperative prediction of 5-year survival in patients with hepatocellular carcinoma. British Journal of Cancer (2020).
3. Quantification of hepatocellular carcinoma heterogeneity with multiparametric magnetic resonance imaging. Scientific Reports (2017).
4. Impact of hepatocellular carcinoma heterogeneity on computed tomography as a prognostic indicator. Scientific Reports (2017).