Application Of Based On Contrast-enhanced CT Radiomics In Differentiating Pathological Grade Of Hepatocellular Carcinoma

Objective: To explore the feasibility of based on preoperative contrast-enhanced CT radiomics for predicting pathological grade of hepatocellular carcinoma(HCC).Methods: A total of 242 HCC patients who underwent partial hepatectomy confirmed by postoperative pathology were analyzed retrospectively in our hospital from 1 January2016 to 31 December 2020.Patients were randomized in 7:3 ratio into training(n=170)and test(n=72)groups,and who’s clinical data and radiological data were collected.The3 D slicer software was applied to the enhanced CT images to delineate the region of interest(ROI)along the largest slice of the tumor contour and at the slice level up and down 5mm respectively.The radiomics features of Arterial Phase(AP)and portal Venous Phase(VP)were extracted after preprocessing the ROI images.First,univariate analysis was performed on the extracted radiomics features.Radiomics features with P<0.05 after univariate analysis were subjected to dimensionality reduction using least absolute shrinkage and selection operator(LASSO).After filtering out the radiomics features that were most valuable for predicting the pathological grade of hepatocellular carcinoma,the dataset was divided into well differentiated and poorly differentiated groups based on surgical pathology results,and radiomics models based on AP,VP,and AP+VP were constructed,from which the best predictive efficacy radiomics model was selected.At the same time,the clinical model was constructed after the most meaningful clinical features of the predicted the pathological grade of HCC.Then,the combined models AP+Clinical,VP+Clinical,and AP+VP+Clinical were constructed by combining clinical and radiomics features,and the best predictive efficacy combined models were selected for comparison among the three combined models.Finally,the ROC curves of clinical model,radiomics models,combined models these three models for predicting pathological grade in the training and validation sets were plotted,and the predictive efficacy of each model was evaluated.Results: Univariate and multivariate analysis of the clinical data show that: AFP and neutrophil count were independent risk factors for poorly differentiated hepatocellular carcinoma.A clinical model was built based on the above independent risk factors.In the training group,it’s AUC is 0.673(95%CI: 0.597～0.743),sensitivity: 88.57%,specificity:40.00%,Youden index: 0.286,and the AUC in the test group is 0.567(95%CI:0.445～0.683),sensitivity:68.57%,specificity: 51.35%,Youden index: 0.199.Among the AP,VP,and AP+VP radiomics models,the AP+VP model had the best efficacy in predicting differentiation grade.In the training group,it’s AUC is 0.825(95%CI:0.760～0.879),sensitivity:80.00%,specificity:74.00%,Youden index: 0.540,and the AUC in the test group is 0.748(95%CI: 0.632～0.843),sensitivity:74.29%,specificity: 75.68%,Youden index: 0.500.Among the AP+Clinical,VP+Clinical,and AP+VP+Clinical radiomics models,the AP+VP+Clinical model had the best efficacy in predicting differentiation grade.In the training group,it’s AUC is 0.837(95%CI: 0.773～0.889),sensitivity:85.71%,specificity:69.00%,Youden index:0.547,and the AUC in the test group is 0.745(95%CI: 0.629～0.841),sensitivity:57.14%,specificity:86.49%,Youden index: 0.436.Finally,we compared three models: Clinical,AP+VP,and AP+VP+Clinical,in which the AP+VP model have the best performance and the model was composed of14 radiomics features that were significantly associated with pathologic differentiation grade.In the training group,the AUC of AP+VP model is 0.825(95%CI: 0.760～0.879),sensitivity:80.00%,specificity:74.00%,Youden index: 0.540,and the AUC in the test group is 0.748(95%CI: 0.632～0.843),sensitivity:74.29%,specificity:75.68%,Youden index: 0.500.Conclusions: Contrast-enhanced CT radiomics is feasible in the preoperative prediction of the pathological grade of hepatocellular carcinoma,in which the AP+VP model have the best efficacy in predicting differentiation grade.The model based on contrastenhanced CT radiomics can predict the pathological grade of HCC noninvasively,simply and accurately,which is of great significance to guide personalized medicine.