What type of study is this?

This is a Validation Study study.

What question did this study set out to answer?

To develop a predictive model that incorporates radiomics and clinical risk factors for TACE refractoriness in HCC.

April 29, 2026Open Access

Preoperative Prediction of TACE Refractoriness in Hepatocellular Carcinoma Using CT-Based Radiomics Model

Key Points

To develop a predictive model that incorporates radiomics and clinical risk factors for TACE refractoriness in HCC.
Analyzed data from 180 HCC patients and 42 external validation patients.
Used LASSO regression and Boruta algorithm for feature extraction.
Developed machine learning models, primarily random forest, to assess TACE refractoriness.
Used SHAP to interpret predictions and assess feature importance.
Random forest model showed AUC of 0.841 in testing cohort and 0.777 in validation cohort.
Combined model enhanced performance with AUCs of 0.842 and 0.847 in testing and validation cohorts, respectively.
SHAP analysis indicated significant contributions from radiomic features.

Abstract

Purpose: To develop an integrated predictive model that combines radiomics, and clinical risk factors to predict early refractoriness to transarterial chemoembolization (TACE) in patients with hepatocellular carcinoma (HCC). Methods: The study cohort comprised 180 HCC patients from Hospital A, while the external validation cohort included 42 patients from Hospital B. Optimal radiomic features extracted from computed tomography (CT) were selected using both LASSO regression and the Boruta algorithm. Eight machine learning models based on radiomics were developed. SHapley Additive Explanations (SHAP) were utilized to interpret the predictions and assess feature importance of the best model. Furthermore, independent clinical risk factors for TACE refractoriness were identified within the study cohort, leading to the construction of a combined model. The predictive performance of the model was evaluated using the area under the curve (AUC), calibration curve, and decision-curve analysis (DCA). Results: The random forest (RF) model exhibited the superior performance, achieving an AUC of 0.841 (95% CI: 0.731– 0.950) and 0.777 (95% CI: 0.624– 0.929) in the testing and validation cohorts, respectively. SHAP analysis indicated that radiomic features significantly contributed to the RF model. Subsequently, Radscore was integrated with the clinically independent risk factor (tumor diameter) identified through univariate and multivariate logistic regression to develop the combined model. The combined model exhibited superior AUC performance compared with the clinic and radiomics models, with AUCs of 0.842 (95% CI: 0.736– 0.948) and 0.847 (95% CI: 0.721– 0.973) in the testing and validation cohorts, respectively. Calibration curve and decision curve analyses confirmed the utility of the combined model nomogram in clinical practice. Conclusion: The combined model exhibits strong predictive performance for early TACE refractoriness, potentially offering improved guidance for decision-making regarding subsequent TACE treatments. Keywords: HCC, radiomics, TACE refractoriness, machine learning, nomogram

Read Full Paperexternally

Demander à l'IA

Bookmark

View Full Paper