Prediction Of Microvascular Invasion In Small Hepatocellular Carcinoma (≤3cm) Based On Gd-EOB-DTPA Enhanced MRI Radiomics

Objective To investigate the effect of a comprehensive model nomogram based on gadolinium ethoxybenzyl diethylenetriamine pentaacetic acid（Gd-EOB-DTPA）enhanced MRI radiomics and clinical-radiological features in predicting microvascular invasion（MVI）in patients with small solitary hepatocellular carcinoma（SHCC）.To compare the predictive effectiveness of radiomics models constructed based on different volumes of interest（VOI）s of Gd-EOB-DTPA-enhanced MRI.Methods This study retrospectively analyzed 196 patients with SHCC（≤3 cm）from January 2017 to December 2020 in the Affiliated Hospital of Qingdao University who underwent preoperative Gd-EOB-DTPA enhanced MRI within one month before surgery and collected clinical and radiological data.According to pathological diagnosis,the patients were divided into the MVI-positive and MVI-negative groups.150 patients（48 in the MVI-positive group and 102 in the MVI-negative group）from Institution I were randomly divided into training cohort（n=105）and testing cohort（n=45）in a 7:3 ratio,and46 patients from Institution II（14 in the MVI-positive group and 32 in the MVI-negative group）as an independent external validation cohort.The VOI of each lesion was manually outlined on the sequence images of artery phase（AP）,hepatobiliary phase（HBP）and T2WI,and the peritumor range was enlarged by 5 mm.VOI_intratumor（containing only intratumoral region）and VOI_{intratumor+peritumor}（intratumoral and peritumoral 5 mm regions）were obtained.Radiomics features were extracted based on different VOIs,the least absolute shrinkage and selection operator（LASSO）regression model was used for feature selection,and single and fused sequences radiomics prediction models were established.The area under the curves（AUC）of the receiver operating characteristic curves（ROC）were used to compare the prediction efficiency of radiomics models based on different VOI from different sequences and select the best radiomics model.The independent risk factors for MVI in SHCC patients were determined by univariate and multivariate Logistic regression analysis,and the clinical-radiological model was established.The best radiomics and clinical-radiological models establish a comprehensive model and draw a nomogram.Used AUC,accuracy,sensitivity and specificity to evaluate the performance of each model.Used Decision curve analysis（DCA）and calibration curves to evaluate the clinical application value of the nomogram.Net reclassification improvement（NRI）and Integrated discrimination improvement（IDI）analyses validated the advantages of the comprehensive model over the clinical-radiological model.Results Among the eight radiomics models,the fused sequences model based on VOI_{intratumor+peritumor} performed best,with an AUC of 0.843,0.800 and 0.875 in training,testing and external validation cohorts,respectively.Analysis of clinical and radiological features showed that peritumoral hypointensity on HBP and low tumor apparent diffusion coefficient（ADC）value were independent risk factors for predicting MVI in SHCC patients.The AUC of the clinical-radiological model in the three cohorts was 0.853,0.767and 0.723,respectively.The best radiomics model and clinical-radiological model were integrated to establish the prediction model and draw the nomogram.The results showed that the comprehensive model had good prediction efficiency,and the AUC in the three cohorts were 0.934,0.889 and 0.875,respectively.NRI and IDI analyses showed that the predictive power of the comprehensive model was improved compared with the clinical-radiological model in training,testing and external validation cohorts.DCA showed that the comprehensive model nomogram obtained the most clinical benefit.Conclusions A comprehensive model nomogram based on Gd-EOB-DTPA enhanced MRI radiomics and clinical-radiological features can predict MVI in SHCC patients.The radiomics model based on VOI_{intratumor+peritumor} performed better than the VOI_intratumor.