Abstract
PURPOSE: To develop an integrated predictive model combining radiomics, clinical risk factors, and machine learning for prognostic assessment in hepatocellular carcinoma (HCC) patients receiving continued transarterial chemoembolization (TACE) after developing TACE resistance. MATERIALS AND METHODS: In this retrospective study, 108 HCC patients with TACE resistance treated between September 2015 and September 2024 were analyzed. The dataset was randomly divided into training (70%) and test (30%) cohorts. Radiomics features were extracted from both intratumoral and peritumoral regions, with margins of 3 mm, 6 mm, and 10 mm. Subsequently, radiomics scores (rad-scores) were computed. Multiple machine learning algorithms were evaluated for model performance. RESULTS: Multivariate analysis identified Barcelona Clinic Liver Cancer (BCLC) stage and lymphocyte-to-monocyte ratio (LMR) as independent prognostic factors. The support vector machine (SVM)-based combined model showed the highest predictive accuracy, with an area under the curve (AUC) of 0.956 (95% CI: [0.910-1.000]) in the training cohort and 0.897 (95% CI: [0.772-1.000]) in the test cohort, significantly surpassing models using only clinical or radiomics data. Survival analysis revealed a longer median survival in the low rad-score group (< 3.05) compared to the high rad-score group (33 vs. 22 months; p = 0.009) and in the high-LMR group versus the low-LMR group (32 vs. 15 months; p < 0.001). CONCLUSION: The machine learning-based combined model can effectively predict the prognosis of HCC patients who continue TACE after developing TACE resistance. Rad-score, BCLC stage, and LMR are potential prognostic indicators for repeated TACE following TACE refractoriness.