# A novel dynamic nomogram based on contrast-enhanced computed tomography radiomics for prediction of glypican-3-positive hepatocellular carcinoma

## TL;DR

This paper introduces a new tool that uses CT scans to predict whether a liver tumor expresses a specific protein linked to cancer, which could help guide treatment decisions.

## Contribution

A novel dynamic nomogram based on CT radiomics for predicting glypican-3 (GPC3) positivity in hepatocellular carcinoma (HCC) is developed.

## Key findings

- A nomogram combining radiomics and clinical factors achieved an area under the curve of 0.794 for predicting GPC3-positive HCC.
- The nomogram outperformed clinical and radiomics-only models in predictive accuracy and net benefit.
- The tool is available as a free mobile application for clinical use.

## Abstract

The 5-year overall survival of hepatocellular carcinoma (HCC) is still poor. Since glypican-3 (GPC3) is highly expressed in most HCC but not in healthy or non-malignant livers, it may become an ideal therapeutic target for HCC. Thus, this study aimed to construct a dynamic nomogram based on contrast-enhanced computed tomography (CT) radiomics for predicting GPC3 expression.

The medical data of consecutive HCC patients from Nanjing Drum Tower Hospital (from January 2020 to August 2023) were retrospectively reviewed. Based on the immunohistochemistry analysis, GPC3-positive was defined as a positive cell rate ≥ 10% (2+ and 3+). The 3D Slicer software and PyRadiomics were used to extract radiomics features on the arterial phase (AP) and venous phase (VP). A radiomics score (Radscore) was constructed using the most predictive features identified by the least absolute shrinkage and selection operator (LASSO) regression analysis. Univariate and multivariate analyses were performed to screen clinical risk factors associated with GPC3-positive. Finally, the Radscore and clinical risk factors were incorporated using logistic regression classification to construct a nomogram.

181 HCC patients were included according to the inclusion criteria. Among them, 106 were GPC3-positive, and 75 were GPC3-negative. Five radiomics features were finally screened, including three AP and two VP features. The nomogram model combining clinical risk factors (alpha-fetoprotein [AFP] ≥ 10 ng/mL, hepatitis B virus surface antigen [HBsAg]-negative, and age) and the Radscore (area under the receiver operating characteristic curve [AUROC] = 0.794) was superior to the clinical (AUROC = 0.724) and radiomics models (AUROC = 0.722), with good consistency in the calibration curve. The decision curve analysis (DCA) demonstrated that the nomogram had the highest net benefit for predicting GPC3-positive. The dynamic nomogram is freely available as a mobile application at https://zheyuzhou.shinyapps.io/GPC3nomogram/.

Since the intra-tumor heterogeneity of HCC and potential complications brought by liver biopsy, our clinical prediction tool identified GPC3 status satisfactorily and might be helpful in clinical decision-making.