Gadoxetic acide-enhanced magnetic resonance imaging radiomicsfeatures for predicting histopathological grade of hepatocellular carcinoma

Purpose: To create preoperative models to predict histological grade of hepatocellular carcinoma (HCC) based on gadoxetic acid (Gd-EOB-DTPA)-enhanced magnetic resonance imaging (MRI) radiomics. Materials and Methods: 68 patients who were diagnosed with hepatocellular carcinoma histopathologically by trucut needle biopsy in Çukurova University Hospital between September 2015 and November 2021 were retrospectively included in our study. Histopathological grading was performed according to the Edmondson-Steiner system, and the patient group was examined in two groups as well and moderate+poor differentiated. Radiomics features were extracted from hepatobiliary phase images. In addition to the radiomics features, some clinical and laboratory features of the patients were examined in order to create a combined model. The LASSO method was combined with the logistic regression model to select the most important radiomics features. Clinically significant features were included in the logistic regression analysis using the backward LR method and the clinical model was constructed. In addition, a combined model was obtained by adding the rad-score to the clinical model. The performance of the Rad-score and models were evaluated by receiver operating characteristic (ROC) curve analysis, and performance measures (area under the curve-AUC, sensitivity, selectivity) were obtained. Results: Of the 68 patients included in the study, 10 (%14.7) were female, 58 (%85.3) were male, and the mean age was 64.1±10.2. It was determined that 48.5% (n=33) of the patients had well, and 51.5% (n=35) had non-well (moderate+poor) differentiation. The distribution of laboratory results (AFP, total bilirubin, Neutrophil/Lymphocyte ratio (NLR), ALT, INR) was similar between the differentiation groups (p>0.05). Patients with well differentiation have significantly lower rad-scores than patients with moderate+poor differentiation. The AUC value of the Rad-score in predicting moderate+poor differentiation was obtained as 0.803 (95% CI: 0.697-0.909), with a sensitivity of 0.77 and a specificity of 0.776. The AUC value of the Rad-score in predicting moderate+poor differentiation was obtained as 0.803 (95% CI: 0.697-0.909), with a sensitivity of 0.77 and a specificity of 0.776. The AUC of the clinical model was 0.749 (95% CI: 0.615-0.883, p=0.002), which increased to 0.827 (95% CI: 0.716-0.937, p<0.001) with the addition of rad-score information. Conclusion: Combined model consisting of clinical parameters and Gd-EOB-DTPA contrast (hepatobiliary phase only) enhanced MR radiomics features can distinguish well and moderately+poorly differentiated HCCs with high success.