The predictive value of a fusion model constructed from Gd-BOPTA-enhanced MRI radiomics and clinicopathological data for early recurrence after surgery of hepatocellular carcinoma

CAI Bing; WANG Xinying; WEI Hong; ZHANG Yifan; YANG Yunfei; ZHENG Wanjing; XING Zhen; CAO Dairong; XIONG Meilian

doi:10.20258/j.cnki.1006-9011.2025.12.015

Abstract：Objective To construct an radio-clinical fusion model based on Gd-BOPTA enhanced MRI radiomics and clinico-pathological data,and to explore the predictive value of this model for early postoperative recurrence in patients with hepatocellu-lar carcinoma(HCC).Methods A total of 110 HCC patients were randomly divided into a training set(n=76)and a testing set(n=34)at a 7∶3 ratio.All patients underwent partial hepatectomy.In the training set,there were 32 cases of early postoperative recurrence and 44 cases without recurrence,while,in the testing set,there were 14 cases of early postoperative recurrence and 20 cases without recurrence.All the research subjects underwent fMRI examination before the operation.By segmenting and ex-tracting the intratumoral and peritumoral radiomics features,intratumoral,peritumoral,and intratumoral+peritumoral ra-diomics models were constructed,and the optimal radiomics model was selected by comparing the ROC curve.Multivariate Logis-tic regression was used to screen the independent risk factors for postoperative recurrence in the training set of patients,and a clinical model was constructed based on this.Subsequently,the screened clinicopathological data were combined with the opti-mal radiomics model to construct a fusion model.The ROC curve was used to evaluate the predictive efficacy of the radiomics model,clinical model,fusion model and the ALBI and PALBI scores of the classic serological models for early recurrence after partial hepatectomy.Then,the consistency between the predicted probability and the actual probability of the radiomics model,clinical model and fusion model was analyzed through the calibration curve,and the clinical benefits of the three were compared through the decision curve analysis.Results In the training set and validation set,pairwise comparisons of the AUC of intratu-moral,peritumoral,and intratumoral+peritumoral fusion radiomics models for predicting early postoperative recurrence of HCC showed no statistically significant differences(all P>0.05).To simplify the model,only the intratumoral radiomics model(here-inafter referred to as the radiomics model)was included in the subsequent process as the optimal radiomics model to participate in the construction of the fusion model.Multivariate Logistic regression model analysis showed that age,γ-glutamyl transferase,microvascular invasion,and maximum tumor diameter were independent risk factors for predicting early recurrence of liver can-cer after surgery(P<0.05).A fusion model was constructed based on clinical independent risk factors and radiomics models.The AUC of this model for predicting early recurrence after HCC surgery in the training set was 0.913,and its predictive efficacy was superior to the ALBI and PALBI scores of the classical serological models(all P<0.05).However,there were no statistically sig-nificant differences when compared with the radiomics model and the clinical model(all P>0.05).Calibration curves analysis showed that the predicted probability of the fusion model in the training set and validation set was closer to the actual probability.Meanwhile,DCA analysis showed that the fusion model provided greater net benefits within a reasonable threshold probability range.Conclusion In this study,a radio-clinical fusion model is successfully constructed based on Gd-BOPTA enhanced MRI radiomics and clinicopathological features.This model has a good predictive value for the early recurrence of HCC after surgery.