A web-based prognostic nomogram incorporating VETC-associated imaging markers for predicting postoperative recurrence in hepatocellular carcinoma.

backgroundVessels that encapsulate tumor clusters (VETC) is a powerful predictor of aggressive hepatocellular carcinoma (HCC) and associated with poor outcomes of HCC. Imaging surrogates of VETC potentially help predict postsurgical recurrence. PURPOSE: To explore the noninvasive predictive potential of contrast-enhanced computed tomography (CE-CT) for VETC of HCC (VETC-HCC). A web-based prognostic nomogram model including VETC-associated imaging markers was subsequently created to predict postoperative recurrence-free survival (RFS) in HCC patients.

methodsA retrospective evaluation was performed on 393 patients with HCC who underwent CE-CT and immunohistochemical staining for CD34 at three different institutions. Patients from institution 1 (n = 241) were split into training (n = 169) and internal test (n = 72) sets. The remaining 152 patients from institutions 2 and 3 were used as the external test set. Univariate logistic regression analyses and six machine learning algorithms were performed on the training set, and the performance of 6 ML models (AUC, DeLong test, etc.) was compared to identify VETC-associated imaging markers (which were calculated as the VETC score), which were validated in the internal and external test sets. An interactive prognostic nomogram model including VETC-associated imaging markers and clinical data was used to predict RFS in the training and internal test sets. The association between the model’s stratification and postoperative recurrence after radical resection or liver transplantation was also assessed.

resultsNonsmooth margins (P = 0.011), tumor size > 5 cm (P = 0.030), and intratumoral necrosis (P = 0.038) were identified as independent predictors of VETC-HCC, and were combined into 6 machine learning models. Logistic regression (LR) was the final selected model and the VETC score was calculated. In the training, internal test, and external test sets, the VETC score demonstrated effective predictive performance for VETC (AUC: 0.768, 0.742, and 0.724, respectively). The interactive prognostic nomogram model( https://radiology.shinyapps.io/DynNomapp/ ) including the neutrophil-to-lymphocyte ratio (NLR), serum alpha-foetoprotein (AFP), and VETC score yielded C-index values ranging from 0.805 to 0.783 in the training and internal test sets and produced three prognostically distinct groups. Among patients classified as those associated with medium risk by the model, those who underwent liver transplantation had significantly improved RFS (P < 0.05). In contrast, radical resection had no significant effect on RFS in patients classified as having either low or high risk (both P > 0.05).

conclusionPreoperative CE-CT features can be used to characterize VETC-HCCs. The prognostic nomogram model has prognostic value for the preoperative prediction of RFS can aid in the selection of the appropriate surgical approach.