Fig. 7: Development, validation, and clinical evaluation of the serum protein signature.

A Flowchart for serum protein signature development and validation. B Kaplan–Meier DFS curves for cohort 1 (left); AUC(t) analysis comparing DFS prediction by the serum protein signature versus conventional clinical staging systems (right). C Kaplan–Meier DFS analysis for cohort 2 (left); AUC(t) comparison (right). D ROC curve assessing the signature’s predictive accuracy for TACE clinical benefit. E The confusion matrix illustrates the correspondence between the signature-based predictions and the actual clinical outcomes at the optimal cutoff value. F Box plot of risk scores in TACE responders vs. non-responders. G Dot plot of aggregated SHAP values for global interpretability of the serum protein signature. Elevated protein expression with negative SHAP values (blue) associates with lower risk scores; elevated expression with positive SHAP values (yellow) associates with higher risk scores. H Individualized SHAP explanations for representative TACE-responsive HCC cases. Left: Bar plot visualizing direction and magnitude of serum protein contributions to risk predictions. Right: Pre-/post-treatment CT or MRI. ***p < 0.001. FAHNU the First Affiliated Hospital of Nanchang University, RCOX regularized Cox, FSSVM fast survival support vector machine, ROC receiver operating characteristic, AUC area under the curve, PPV positive predictive value, NPV negative predictive value, BCLC Barcelona Clinic Liver Cancer, CNLC China Liver Cancer Staging, TNM tumor node metastasis, SHAP Shapley Additive exPlanations, CT computed tomography, MRI magnetic resonance imaging, TACE transarterial chemoembolization.