Predicting Response To Transarterial Chemoembolization In Unresectable Hepatocellular Carcinoma Based On PyRadiomics

BackgroundTransarterial chemoembolization(TACE)is the standard treatment for patients with unresectable hepatocellular carcinoma(uHCC),which is recommended in the guidelines for the management of hepatocellular carcinoma(HCC).However,in clinical practice,not all patients respond well to TACE therapy,and some patients are still in progressive disease after several TACE procedures.Radiomics is an emerging field which could be applied to extract quantitative features from medical images.In previous studies,radiomics-based signature or model has been proven to be associated with gene mutation,treatment response and prognosis in a variety of tumors.This suggests a new question:whether radiomics methods could be applied to predict TACE response and help with clinical decision-making in patients with uHCC.ObjectiveTo establish a prediction signature to precisely predict if a uHCC patient would respond well to TACE treatment based on three-dimensional reconstructed medical images,and then help physicians to identify patients who could benefit from TACE procedure using pretreatment medical images.Thus,the clinicians would make precise and individual clinical decisions based on the prediction results to prolong overall survival in uHCC patients.MethodsThere were 313 patients enrolled into this multi-center retrospective study who met the inclusion criteria.We collected the contrast-enhanced computed tomography images(arterial and portal venous phase images).Then the tumor was segmented to regions of interest(ROI)and three-dimensionally reconstructed by 3D-Slicer.Radiomics data were extracted in a comprehensive open source platform,PyRadiomics,which is implemented in Python,and is capable of extracting features from images and analyzing them.After the radiomics features were extracted,logistical model was developed using the least absolute shrinkage and selection operator(LASSO)algorithm.A formula for the PyRadiomics score was generated using a linear combination of the selected features,weighted using the LASSO method.Based on these results and coefficients,a PyRadiomics signature was developed.Receiver operating characteristic curve(ROC)analysis was used to explore the effectiveness of the prediction signature.In the present study,confusion matrix and some evaluation measures of classifier(Accuracy,Precision,Recall and F-1 Score)were also generated to explore the results in more detail.Furthermore,all the evaluations were validated in the independent validation cohort.We also compared the area under the curve(AUC)of traditional predicting parameters with PyRadiomics signature,and ran decision curve analysis(DCA)to compare the net benefit of PyRadiomics prediction signature and traditional clinical predicting parameters.ResultsA total of 1167 PyRadiomics features were extracted by the open-source platform.The final PyRadiomics signature,which consisted of 24 features selected using LASSO regression,showed significant capability to predict TACE response.The AUC of ROC analysis obtained by employing the signature were 0·896(95%confidence interval:0·848-0·944)and 0·887(95%confidence interval:0·833-0·941)in the training and validation cohort,respectively.As for the evaluation index of classifier,the accuracy and precision rate of PyRadiomics prediction signature is also high.The accuracy was 81·4%and 80·1%in the training and validation cohorts,respectively.The F-1 score was 0.778 and 0.721 respectively.The DCA showed that the PyRadiomics signature had the best ability to provide the overall net benefit among all tested strategies when the threshold probability was>6%,which indicated that its promising value in clinical application.ConclusionThe 3D-PyRadiomics signature possesses excellent identification and discrimination ability,and could predict the reponse to transcatheter arterial chemoembolization in patients with unresectable hepatocellular carcinoma well.It was able to gain higher net benefit in clinical scenarios and assist physicians to select patients before operation,so as to develop more personalized and accurate treatment plans for uHCC patients,and thus could improve the prognosis of patients.The present study of radiomics provides a new idea for the accurate treatment of unresectable hepatocellular carcinoma in the future.