Radiomic biomarkers for the recurrence prediction of hepatocellular carcinoma treated with postoperative TACE: A multicenter retrospective study

Purpose

The study aims to evaluate the value of radiomics signature in predicting the recurrence risk in hepatocellular carcinoma (HCC) patients treated with postoperative adjuvant transarterial chemoembolization (PA-TACE).

Patients and Methods

In this retrospective study, 204 HCC patients treated with PA-TACE between November 2014 and May 2023 from three centers were included and stratified into the training (n = 91), the internal(n = 21) and external validation cohorts (n = 92). Based on multi-parametric magnetic resonance imaging (mpMRI), radiomics features were extracted and radiomics models were established by using 101 combinations of 10 machine learning algorithms. The most valuable radiomics model with the highest average C-index was identified and subsequently used to calculate the Rad-score. All patients were then stratified into low- and high-risk radiomics signature (LRS and HRS) groups based on the median value of the Rad-score and subgroup analyses were performed to explore the potential association between recurrence risk and benefit from PA-TACE. Furthermore, a combined nomogram was subsequently developed by integrating the Rad-score with relevant clinicopathological variables.

Results

The radiomics model developed by CoxBoost + survival-SVM method was regarded as the optimal model with C-index (95 % CI) of 0.828 (0.777–0.879), 0.796 (0.622–0.933), and 0.718 (0.647–0.781) in three cohorts. The RFS of the HRS group was superior to that of the LRS group in the training(52.4 months v.s. 27.5 months; P < 0.0001) and external validation cohorts(45.8 months v.s. 43.0 months; P < 0.0001). The combined nomogram presented better predictive performance in the training set (0.848, 0.793–0.903), internal validation (0.825, 0.689–0.962) and external validation (0.722, 0.657–0.787). The decision curve analysis indicated that the combined nomogram had relatively higher clinical net benefits.

Conclusion

mpMRI-based radiomics features are associated with recurrence risk in HCC patients receiving PA-TACE and may serve as the potential imaging biomarkers for stratifying candidates who are more likely benefit from PA-TACE.