Radiomics-Based Delineation of Gross Tumor Volume and Normal Liver Tissue in Hepatocellular Carcinoma using Contrast-Enhanced CT and Ensemble Machine Learning.

INTRODUCTION

Precise discrimination between gross tumor volume (GTV) and normal liver parenchyma is essential for radiotherapy target delineation in hepatocellular carcinoma (HCC). This study aimed to develop and compare machine learning models based on contrast-enhanced CT radiomic features to distinguish HCC GTV from normal liver tissue.

MATERIALS AND

METHODS

This retrospective study included 82 patients with histopathologically confirmed HCC. GTVs were delineated on contrast-enhanced CT images, and control regions of interest were defined in normal liver parenchyma distant from the tumor. A total of 1,209 radiomic features were extracted and selected using variance filtering, least absolute shrinkage and selection operator regression, and variance inflation factor analysis, followed by robust standardization. Seven machine learning classifiers were evaluated using patient-level five-fold cross-validation repeated five times. Model performance was assessed in terms of discrimination, classification performance, and probability calibration.

RESULTS

Twelve core radiomic features were identified. Among all models, the CatBoost (CB) classifier achieved the best overall performance, with a mean area under the receiver operating characteristic curve (AUC) of 0.98. Decision curve analysis demonstrated a clear net clinical benefit of the CB model across clinically relevant threshold probabilities, while Fisher discriminant analysis confirmed good separability between tumor and normal liver tissue.

DISCUSSION: CT-based radiomics combined with machine learning enables objective differentiation between HCC GTV and normal liver tissue and may reduce inter-observer variability in radiotherapy target delineation.

CONCLUSION

The CatBoost-based radiomics model shows strong potential as an auxiliary decision-support tool for radiotherapy planning in HCC.