CT-based Radiomics Study For Prediction Of Hcc Histologic Grade And The Prognosis After Hepatectomy

Objective:To assess the value of radiomics signature to predict the histologic grade of hepatocellular carcinoma(HCC)based on enhanced computed tomography(CT)images,and the prognostic value of radiomics features for single HCC after hepatectomy.Ma terials and Methods:1.A total of 385 patients were collected and included in the retrospective study of the pre-operative prediction model for Edmondson-Steiner grade of the hepatocellular carcinoma,divided into training group and test group according to the time of examination and proportion.Vary image filtrations and image differencing were used to smooth image and to enhance the local detail,respectively.17 gray-level co-occurrence matrix(GLCM)features,14 gray-level run-length matrix(GLRLM)features and 8 morphological features were extracted from the portal phase CT images.Feature selection was performed with the minimum redundancy maximum relevance(mRMR)algorithm in the training group.Akaike information criteria(AIC)was used to identify the best model.Radiomics scores were calculated for each patient in test group.The diagnostic performance of the model was evaluated with receiver operating characteristic(ROC)curve analysis.ROC curves were then generated in the training dataset and test dataset.The optimal cut-off threshold values were determined at the point on the ROC curve at which the positive likelihood ratio was maximal,followed by the derivation of the sensitivity and specificity.Area under the curve(AUC)and diagnostic accuracy to distinguish the grade was derived in both the training dataset and test dataset..2.A total of 61 single HCC patients were enrolled for studying the the prognostic value of radiomics features.Textural features of the CT images were quantified,including the features from the category of GLCM,Gabor and wavelet transform,after a Laplacian of Gaussian spatial band-pass filter was used.The differences between the hepatic arterial phase and the portal venous phase were obtained(the Dif.).ROC was used to screen the texture parameters and to obtain the optimal cut-off value.The parameters with statistical significance in ROCs,along with other significant clinical variables(screened by univariate regressions;p<0.10)were entered for multivariate regression analyses by Cox proportional hazards models on overall survival(OS)for each filter.Afterwards,AIC were used to identify the best model for OS.Patients were divided into two groups according to the significant factors identified by multivariate regression analyses.Survival curves for OS and disease-free survival(DFS)were obtained using the Kaplan-Meier method and log-rank tests were performed to compare the difference.Results:1.The radiomics signature score was calculated for each patient based on 8 selected features as follows:Rad Score = 1.009-0.5107xorig_orig_fos_median+0.3362xg.075-gab.20.1-GLCM-homogeneity2 + 0.3719x g.075_gab.15.2_GLRLM_LRLGLE-0.5798 ×Compactness2-0.2858xorig_diff.4_fos_krutosis +0.3261 xg.100_gab.10.1_GLCM_homogeneity2 +0.6769xg.075_lap.cut2_GLCM_IDN-0.3208xg.075_gab.15.2_GLCM_IMC2The radiomics signature was an independent predictor for staging of HCC,which could successfully categorize HCC into Edmondson-Steiner grade I-II and grade III-IV(p<0.0001)in training and validation dataset.As for the classification performance of the radiomics signature in HCC staging,the AUC was 0.766(95%CI:0.824-0.707)with sensitivity of 0.667 and specificity of 0.753.The signature in the test dataset obtained an AUC of 0.673(95%CI:0.775-0.571)with sensitivity of 0.674 and specificity of 0.683.2.ROC and Cox regression analyses identified five parameters.Filter 1.0 achieved the best performance,in which the Dif.Scalel.2 was a superior indicative independent marker for OS(p=0.05).Kaplan-Meier analyses further demonstrated that the Dif.Scale2.2 at filter 0(p=0.001),Dif.Scalel.2(p =0.006),Dif.Scale3.2(p =0.005)at filter 1.0,Dif.Wavelet 8 at filter 1.5(p<0.001),and corona(p =0.032)were associated with OS.Moreover,Dif.Scale2.2 at filter 0(p =0.039),Dif.Scalel.2 at filter 1.0(p=0.001),and Dif.Wavelet 8 at filter 1.5(p =0.007)were associated with DFS,while the Barcelona-Clinic Liver Cancer(BCLC)parameters showed no statistical correlation with OS(p =0.057).Conclusions:A radiomics signature was developed and validated to be a significant predictor for discrimination of Edmondson-Steiner grade I-II from grade III-IV HCC,which may serve as a complementary tool for the preoperative tumor staging in HCC.For patients with a single HCC treated by hepatectomy,the textural features for Gabor and Wavelet,especially the varying Dif.,potentially provided prognostic information beyond traditional indicators such as those of the BCLC.