MRI-based Radiomics Signature:A Potential Biomarker For Identifying Glypican 3-positive Hepatocellular Carcinoma

Objective: To investigate the performance of MRI-based radiomics signature in identifying glypican 3(GPC3)-positive hepatocellular carcinoma(HCC).Materials and Methods: An initial cohort of 293 patients after liver resection with pathologically confirmed HCC was involved in this retrospectively study,and patients were randomly divided into training(195)and validation(98)cohorts.The data of 293 patients was collected from April 2014 to December 2017 in Tianjin Medical University Cancer Institute and Hospital.All of the patients underwent preoperative dynamic magnetic resonance imaging(MRI)with completed pathological report.(1)we analyze the MRI with HCC by using a quantitative radiomic method to predict the glypican3(GPC3)-positive hepatocellular carcinoma,and proposed a MRI-based radiomics signature.A set of 853 radiomic features(shapes features,intensity features,texture features,wavelet features)were extracted from delay-phase of dynamic MR images for each patient.Univariable analysis and fisher information were utilized for feature reduction.Subsequently,further forward stepwise feature selection and radiomics signature building were performed based on support vector machine(SVM).(2)A set of clinical risk factors(age,gender,Diameter,α-fetoprotein,ALB,ALT,TBIL,Hepatitis,Cirrhosis,Ascites)were analyzed.Univariate analysis and multivariate logistic regression analysis were used to select independent clinical risk factors related to GPC3 expression,and then we proposed a clinical model.(3)Based on the previously constructed radiomics signature and clinical model,a combined model incorporating with radiomics and clinical risk factors was constructed.(4)Based on the previously combined model,a nomogram was developed by multivariable logistic regression modelling for the individualized prediction of GPC3-positive.receive operating characteristic(ROC)curve was used to evaluate the performance of these 3 models.And these 3 models were validated by the validation cohorts.The predictive performance of the nomogram was calculated using area under the receive operating characteristic curve(AUC)and the calibration curve.Decision curve analysis(DCA)was applied to estimate the clinical usefulness.Results: 293 patients with HCC in this study were randomly divided into training cohorts 195(GPC3(+)135,GPC3(-)60)and validation cohorts 98(GPC3(+)68,GPC3(-)30)according to the results of their pathological reports.A set of 853 radiomic features(shapes features,intensity features,texture features,wavelet features)were extracted from delay-phase of dynamic MR images for each patient.The radiomics signature consisting of 10 selected features achieved satisfying prediction efficacy.The area under ROC curve(AUC)of radiomics signature reached 0.879 with a accuracy of 83.1%,a sensitivity of 85.2% and a specificity of 78.3% in training cohorts,and the AUC in validation cohorts reached 0.871 with a accuracy of 75.5%,a sensitivity of 72.1% and a specificity of 83.3%.Two independent clinical risk factors(tumor diameter,AFP)selected from 10 clinical risk factors were correlated with GPC3 expression,and a clinical prediction model was constructed.The AUC of clinical model reached 0.815 with a accuracy of 72.3%,a sensitivity of 68.2% and a specificity of 81.7% in training cohorts,and the AUC in validation cohorts reached 0.758 with a accuracy of 71.4%,a sensitivity of 67.7% and a specificity of 80.0%.Furthermore,The AUC of combined model reached 0.926 with a accuracy of 86.7%,a sensitivity of 86.7% and a specificity of 86.7% in training cohorts,and the AUC in validation cohorts reached 0.914 with a accuracy of 79.6%,a sensitivity of 73.5% and a specificity of 93.3%.The calibration curve and DCA confirmed the clinical usefulness of our nomogram.Conclusion: MRI radiomics can effectively distinguish between GPC3(+)and GPC3(-)HCC before the surgery.The proposed nomogram may provide an effective tool for noninvasive and individualized prediction of GPC3-positive,and it could help clinicians to choose an optimal treatment strategy for patients with HCC.