| Objective: To study the risk factors of early postoperative recurrence of hepatocellular carcinoma(HCC)and to explore the role of MR radiomics in predicting the early postoperative recurrence of patients with HCC after radical resection.Finally,a nomogram prediction model for postoperative early recurrence risk was eatablished based on the independent risk factors affecting postoperative early recurrence of HCC.That is,to explore whether non-invasive imaging methods can be used to predict the risk of early recurrence of HCC after radical resection.Methods: The clinicopathological and MR imaging data of 129 HCC patients who received radical hepatectomy from Department of hepatobiliary surgery,the First Affiliated Hospital of Guangxi Medical University from January 2015 to December 2018 were analyzed retrospectively.The MR images of the study cases were imported into huiyihuiying radiomics cloud platform,and the ROI of the MR enhanced images(arterial phase,portal phase,delayed phase,hepatobiliary phase)of the lesions was manually sketched.The image features based and filter based features of all ROI were extracted.Then we use the Lasso feature selection algorithm to select the most significant features.The data of clinical pathology and MR imaging characteristics,such as sex,BMI,child Pugh score,AFP level,PIVKA-II level,pathological tissue classification,MVI,Ki-67,CK19,VEGF,were analyzed by Logistic regression univariate analysis.And indicators with statistical significance(P < 0.05)were further included into the Logistic regression multivariate analysis.Finally,the nomogram early recurrence prediction model was established for the clinical pathology,MR imaging features or the selected MR radiomics features with excellent diagnostic efficacy.Results: Logistic regression analysis showed that AFP(P = 0.040),MVI(P = 0.020),tumor number(P = 0.035),tumor maximum diameter(P = 0.004),intratumoral necrosis(P = 0.011),satellite nodule(P =0.028),and whether the tumor margin was smooth(P = 0.019)were significantly correlated with the early recurrence of HCC after radical operation.The risk factors with significant correlation in the above Logistic regression univariate analysis were further included in the Logistic regression multivariate analysis.The results showed that only two factors had significant correlation with the early recurrence of HCC after operation,namely AFP level(P = 0.033;or =1.653,95% CI: 1.042-2.624)and the tumor maximum diameter(P = 0.018;or =3.224,95% CI: 1.371-7.584).Six of the most significant radiomics features were screened out by MR imaging.However,diagnostic performances of MR radiomics feature models for early recurrence of HCC after operation were not ideal,and the AUC of the optimal model was only 0.656.The prediction model of early recurrence of nomogram HCC was established by AFP level and tumor maximum diameter.Conclusion: The combination of MR enhanced quantitative features and machine learning can not accurately predict the early recurrence of HCC after radical operation,but the preoperative imaging and AFP examination may be helpful to assist the clinical decision making or the follow-up planning. |
