Radiomics Analysis On Hepatobiliary Phase MRI Of Liver Parenchyma:to Predict The Presence Of Small Hepatocellular Carcinoma In Patients With Liver Cirrhosis

ObjectiveTo determine whether radiomics analysis of hepatobiliary phase(HBP)images of liver parenchyma can predict the presence of small hepatocellular carcinoma(SHCC)in patients with viral hepatitis–related cirrhosis.MethodsThis retrospective study was approved by the ethics committee,and was granted a waiver of informed consent.There were 92 patients with viral hepatitis–related cirrhosis included in this study.According to whether they had SHCC(Diameter?3cm,diagnostic criteria refer to Guidelines for Diagnosis and Treatment of Primary Liver Cancer in China-2017 Edition)when receiving Gd-EOB-DTPA MRI,they were divided into two groups: non-HCC and SHCC group.Clinical information of each patient was collected,including age,sex,the type of viral hepatitis and the Child-pugh grade.Their HBP images of Gd-EOB-DTPA MRI were derived,and then bias correction was done in A.K.software(Artificial Intelligence Kit,A.K.,GE Healthcare,China).A circular region on interest(ROI,diameter =30 pixels)was placed in the predetermined locations of the right hepatic lobe to extract the radiomics features.Randomly,70% samples were selected as the training set for the extraction of radiomics features and the establishment of machine learning model,and 30% samples were selected as the testing set to verify the accuracy of the model.During the extraction of radiomics features,intergroup statistical difference test,univariate logistic regression and spearman test(threshold=0.8)were used in sequence.The logistic regression model was selected as the classifier for machine learning.The accuracy of the model is illustrated by ROC curve,and the area under the curve(AUC),specificity and sensitivity were calculated in the train set and test set as evaluation indexes of the classifier.ResultsThe age,sex and the type of hepatitis virus in the clinical information of the patients in the two groups were not correlated with the occurrence of SHCC in patients who had viral hepatitis–related cirrhosis.There were significant differences in liver function between the two groups.In the SHCC group,there were more cirrhosis patients whose Child-pugh grade was B.Totally,385 radiomics features were extracted in A.K.software.After selection,7 of them were associated with the occurrence of HCC in patients with viral hepatitis–related cirrhosis.Logistic regression model showed that the AUC,specificity and sensitivity in the train set and test set were 0.809(95%CI: 0.737-0.944),0.800,0.727 and 0.805(95%CI: 0.574-0.940),0.643 and 0.800,respectively.ConclusionThe radiomics analysis based on HBP images of liver parenchyma can predict SHCC in patients with viral hepatitis–related cirrhosis,and has a helpful predictable performance.