CT-Based Radiomics Nomogram Improves Risk Stratification and Prediction of Early Recurrence in Hepatocellular Carcinoma After Partial Hepatectomy

Cuiyun Wu, Shufeng Yu, Yang Zhang, Zhu Li, Shuangxi Chen, Yang LiuFrontiers in Oncology 2022 · DOI

TL;DR

This study introduces a novel CT-based radiomics nomogram that significantly improves the prediction and risk stratification of early recurrence in hepatocellular carcinoma (HCC) patients after partial hepatectomy. By integrating radiomics features with clinical and radiological data, this nomogram offers a powerful, non-invasive tool for personalized risk assessment, potentially leading to better patient management and outcomes.

What Problem Does This Paper Address?

Hepatocellular carcinoma (HCC) is a prevalent cancer with a high recurrence rate after surgical resection, particularly within the first year post-surgery (early recurrence or ER). Early recurrence is associated with a worse prognosis, making its accurate prediction crucial for guiding treatment decisions and improving patient survival. While existing staging systems are valuable, they often fall short in precisely predicting ER and quantifying risk. Previous attempts to use radiomics for HCC recurrence prediction have shown promise, but few have specifically focused on ER and even fewer have developed an intuitive, clinically applicable nomogram for risk stratification. This paper aims to bridge this gap by developing and validating a CT-based radiomics nomogram for predicting and stratifying the risk of ER in HCC patients.

Methods

The researchers employed a retrospective approach, enrolling 132 HCC patients who underwent partial hepatectomy. These patients were divided into training (n=91) and test (n=41) sets.

Key methodological steps included:

  • Radiomics Feature Extraction: Contrast-enhanced CT images were preprocessed (resampling, denoising, grey-normalization) and segmented to define the region of interest. A total of 396 radiomics features were extracted from both arterial (AP) and portal venous (PP) phases, encompassing various feature types like histogram, GLCM, RLM, and GLSZM.
  • Feature Selection: To identify the most predictive features, a combination of ANOVA, correlation analysis, and Least Absolute Shrinkage and Selection Operator (LASSO) was used for dimension reduction. This was further refined using a gradient boosting decision tree algorithm.
  • Model Development: Six predictive models were constructed:
  • Clinical Model: Based on clinical factors (e.g., Edmondson grade).
  • Radiological Model: Based on radiological features (e.g., tumor size).
  • Radiomics Model: Based on a derived radiomics score (Rad-score).
  • Radiomics-Clinical Model: Combining Rad-score and clinical factors.
  • Radiomics-Radiological Model: Combining Rad-score and radiological factors.
  • Combined Model: Integrating Rad-score, Edmondson grade, and tumor size. Machine learning algorithms (logistic regression, Bayes, SVM, KNN) were used to evaluate model performance.

  • Nomogram Construction: The best-performing model (the combined model) was used to develop an intuitive nomogram for predicting individual ER probability.

  • Performance Evaluation: Model and nomogram performance were assessed using Receiver Operating Characteristic (ROC) curves (Area Under the Curve - AUC), calibration curves, and Decision Curve Analysis (DCA).

Datasets

The study utilized a dataset of 132 HCC patients who underwent preoperative contrast-enhanced CT at a single institution. This dataset was split into a training set (n=91) and a test set (n=41). No new benchmark datasets were introduced.

Key Insights and Ideas

  • Radiomics Captures Crucial Information: The study strongly suggests that radiomics features, particularly those extracted from both arterial and portal venous phases, capture significant information about tumor heterogeneity and biological behavior that is predictive of early recurrence.
  • Combined Model Superiority: Integrating radiomics features with established clinical (Edmondson grade) and radiological (tumor size) factors yielded the most powerful predictive model, significantly outperforming models based on individual feature types. This highlights the synergistic value of multi-modal data.
  • Intuitive Nomogram for Clinical Use: The development of a user-friendly nomogram transforms complex radiomics data into a practical tool for clinicians. This visual aid allows for straightforward risk stratification, making the findings more accessible and actionable in clinical practice.
  • High-Risk Group Identification: The nomogram effectively stratified patients into high- and low-risk groups for ER, with a statistically significant difference in recurrence rates between these groups. This is a critical step towards personalized treatment strategies.

Key Results

  • Radiomics Score (Rad-score): The Rad-score alone achieved AUCs of 0.870 (training) and 0.890 (test) for predicting ER.
  • Combined Model Performance: The combined model (Rad-score, Edmondson grade, tumor size) demonstrated superior performance, achieving AUCs of 0.907 in the training set and 0.948 in the test set.
  • Nomogram Validation: The nomogram, based on the combined model, showed excellent calibration and good net benefits in clinical decision-making, as indicated by DCA.
  • Risk Stratification: The nomogram successfully stratified patients into high- and low-risk groups, with a highly significant difference in ER risk (p <0.001).

Limitations and Future Work

The authors acknowledge several limitations:

  • Retrospective Design: The study's retrospective nature may introduce selection bias.
  • Single-Center Study: The findings need validation in larger, multi-center prospective studies to confirm generalizability.
  • Specific Patient Population: The study focused on patients with hepatitis B virus-associated HCC, and findings may not directly apply to patients with other etiologies or different clinicopathological features.
  • Lack of Survival Analysis: While risk stratification was performed, actual survival analyses were not conducted.

Future work should focus on prospective validation, expanding the study to diverse patient populations, and incorporating survival data to further refine the nomogram's clinical utility.

Summary updated

Abstract

The CT-based radiomics nomogram developed in this study exhibits outstanding performance for ER prediction and risk stratification. As such, this intuitive nomogram holds promise as a more effective and user-friendly tool in predicting ER for HCC patients after partial hepatectomy.
BibTeX
@inproceedings{wu2022ct,
  author = {Cuiyun Wu and Shufeng Yu and Yang Zhang and Zhu Li and Shuangxi Chen and Yang Liu},
  title = {{CT-Based Radiomics Nomogram Improves Risk Stratification and Prediction of Early Recurrence in Hepatocellular Carcinoma After Partial Hepatectomy}},
  booktitle = {Frontiers in Oncology},
  year = {2022},
  doi = {10.3389/fonc.2022.896002},
}
Figure