Abstract
OBJECTIVES: Noninvasive evaluation and treatment of vessels encapsulating tumor cluster (VETC) HCCs remain challenging. Herein, a new Clinic-Radiologic-Intratumor Radiomics (CRIR) model was investigated for the preoperative prediction of VETC-HCCs and prognosis based on CE-CT, then compared therapeutic outcomes between predicted VETC and nonVETC-HCCs after different treatment methods. METHODS: Total 456 HCC patients who underwent radical resection (RR), liver transplantation (LT) or TACE were retrospectively included in this multicenter (Center 1–4) study between January 2014 and November 2022. The intratumor and 1 cm peritumor VOI were segmented in the three phases of CE-CT imaging. Radiomics features were selected using LASSO and multivariable logistic regression (LR) to filtered the useful features. Clinical, radiological qualitative and quantitative features, intratumor, peritumor and combined radiomics, were established using LR into Clinic-radiological (CR), intratumor radiomics (IR), peritumor radiomics (PR), CRIR, Clinic-radiological- peritumor radiomics (CRPR)and CR-intra and peritumor radiomics (CRIPR) models. Diagnostic performance was calculated and compared for the models. Kaplan–Meier survival analysis was used to assess progression state in model-predicted VETC-HCCs and non-VETC-HCCs in TACE, or early recurrence in both pathologic and model-predicted in RR or LT groups. Additionally, outcomes between the RR and LT groups were compared to determine the optimal treatment approach. RESULTS: Neutrophil-to-lymphocyte ratio (NLR) (P = 0.031), gamma-glutamyl transferase (GGT) (P = 0.043), intratumor necrosis (P = 0.026), Arterial enhancement fraction (AEF) (P = 0.038), and intra-tumoral artery (P = 0.035) were independent predictors of VETC-HCC. CRIR model showed best area under the ROC curve value (0.85-080 across training, internal test, and external test), statistically significant improvement over the clinico-radiologic model, but not the CRIPR model. In patients with pathologic VETC-HCC, those treated with RR exhibited higher early recurrence rate compared to those treated with LT (P = 0.029). On the contrast, the early recurrence rates in patients with pathologic non-VETC-HCC, were similar between the RR and LT groups (P > 0.05). Similarly, the early recurrence rates in patients with CRIR model predicted VETC-HCC or non-VETC-HCC presented same trend as pathologic group. In application (TACE) group, CRIR model predicted VETC-HCC had lower tumor response rate (50.00% vs. 75.56%, P < 0.001) and worse PFS (17 months vs. 30 months; P = 0 0.039) than those with non-VETC HCCs. CONCLUSION: The CRIR model provides accurate preoperative identification of VETC-HCC and offers prognostic value for early recurrence following RR or LT, tumor response after TACE and surgical approach selection between RR and LT in solitary HCC. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12885-025-14408-1.