Prediction of mRECIST tumor response at firstfollow-up after DEB-TACE using a combined radiomics-clinical model and explainability methods

To investigate the value of preoperative enhanced computed tomography (CT) radiomics combined with clinical features in predicting the tumor response according to mRECIST of initial drug-eluting bead transarterial chemoembolization (DEB-TACE) in hepatocellular carcinoma (HCC) patients by utilizing the Shapley additive explanations (SHAP) algorithm for model interpretation. A retrospective analysis was conducted on 110 patients. Treatment response was evaluated using the modified Response Evaluation Criteria in Solid Tumors. Radiomic features were extracted from arterial phase computed tomography images and reduced using the least absolute shrinkage and selection operator. These features were combined with clinical indicators to construct predictive models, including logistic regression, naive Bayes, support vector machine, random forest, and XGBoost. The model performance was evaluated using the area under the curve, calibration curve, and decision curve. The combined model showed optimal predictive performance, with area under the curve values of 0.838 and 0.802 for the training and test sets, respectively, statistically outperforming the single models (DeLong test, P < 0.05).Furthermore, SHAP analysis revealed key predictors and their directional effects. Preoperative enhanced computed tomography radiomics combined with clinical indicators can effectively predict the initial tumor response to DEB-TACE. SHAP visualization analysis enhances interpretability and identifies key predictors, offering support for precise treatment and individualized decision-making.