Clinical Application Study Of Multimodal MR Imaging And Radiomics Prediction Model In Evaluation Of Microvascular Invasion And Prognosis Of Hepatocellular Carcinoma

AimMicrovascular invasion（MVI）is an important predictor of recurrence and prognosis in hepatocellular carcinoma（HCC）,and is the basis for determining the scope of surgery and other interventions for HCC patients.However,the diagnosis of MVI depends on the invasive techniques such as biopsy and postoperative pathology.Therefore,finding the effective and new noninvasive diagnostic methods for MVI is an urgent problem to be solved in clinical practice.Using Radiomics features to diagnose MVI has great potential,but its methods and value need to be further explored and clarified.The aim of this study is to clarify the application value of Radiomics features in preoperative evaluation of MVI in HCC patients.MethodsPatients with HCC who underwent surgical resection between September 2015 and November 2021 were screened and included to analyze the clinical and imaging characteristics of HCC.Firstly,the potential diagnostic ability of MR signal intensity for the pathological grading,MVI,and prognosis of HCC patients was evaluated.Subsequently,the eligible patients from two independent institutions who were surgically and pathologically diagnosed as HCC and underwent preoperative dynamic enhanced MRI with gadoxetic acid disodium（Gd-EOB-DTPA）were divided into the training cohort and the validation cohort.In the training cohort,multiple sequences of dynamic contrast-enhanced MRI based on Gd-EOB-DTPA[fat suppression T1weighted sequence（T1WI-FS）,fat suppression T2 weighted image（T2WI-FS）,apparent diffusion coefficient（ADC）map,arterial phase,portal vein phase,transitional phase,hepatobiliary phase]and including tumor interior Multiple regions of the five regions of interest around the tumor(VOI:VOI^tumor,VOI^50%,VOI^10mm,VOI^20mm,VOI^liver)extract Radiomics features and construct relevant Radiomics models to predict MVI.A nomograph model was constructed to accurately predict MVI before surgery based on the clinical parameters,imaging manifestations,and Radiomics characteristics of HCC patients before surgery.The predictive ability of the established Clinical-Imaging models,Radiomics models,Clinical-Radiomics models,and Clinical-Imaging-Radiomics models for MVI were compared.The Kaplan Meier method was performed to obtain survival curves,and the log rank test was used to compare relapse free survival.Results1.The SI_Ap/Al,and SI_At in HCC patients are independently related to pathological grading,and had certain discriminating abilities in preoperative diagnosis of HCC pathological grading（AUC values are 0.652,0.652,respectively,P<0.05）.Compared with patients with SI_Ap/Al<1.1,the patients with SI_Ap/Al≥1.1 had a significantly lower relapse free survival rate after radical HCC resection（P<0.05）.2.SI_Ap/Al was associated with MVI in HCC patients,with a correlation coefficient of r=0.607（P<0.05）.Multivariate analysis showed that SI_Ap/Al was independently associated with MVI positive in HCC patients（P<0.05）,and SI_Ap/Al had a better ability in distinguishing MVI in HCC patients（AUC=0.851,P<0.05）.3.For the effectiveness of extracting features from different sequences to predict MVI,the features in T1WI-FS and hepatobiliary phase sequences were superior to other sequences in the single sequence model.In the training cohort（AUC=0.842 vs.0.747）and the validation cohort（AUC=0.804 vs.722）,the Radiomics model showed potentially better ability of MVI discrimination than the Clinical-Imaging model.In combination with alpha fetoprotein,envelope integrity,periarterial enhancement,and imagegenic venous invasion,the nomograms of combined Radiomics features showed excellent predictive ability in training（AUC=0.891）and validation cohorts（AUC=0.836）,which were superior to Radiomics or Clinical-Radiomics models.The calibration curve shows that the Clinical-Imaging-Radiomics nomogram model has a good consistency between the predicted probability and the actual probability.In addition,the recurrence free survival time of the nomograph model was similar to the histopathological results.ConclusionsThe SI_Ap/Al in preoperative contrast-enhanced MR images can distinguish MVI in HCC patients,and can be used as a predictor of postoperative prognosis in HCC.A multi region,multi sequence Clinical-Imaging-Radiomics models based on serum alpha fetoprotein,MR imaging signs,and Radiomics characteristics have good abilities to predict MVI in HCC patients.This helps clinicians develop the optimal treatment strategies to improve the clinical outcome of HCC.