Hepatocellular Carcinoma (HCC) is most common malignant neoplasm of the liver and the third leading cause of cancerous death worldwide$$$^1$$$. Lesion characterization based on imaging features is essential to the successful treatment of HCC. Various methods have been proposed based on different image features, such as texture, derived from the first- and second- order intensity statistics, to demonstrate the intrinsic characterization of HCC $$$^2$$$. To our knowledge, the correlation between the image feature and malignant of HCC has not been fully investigated in a quantitative manner before $$$^3$$$. In this work, we investigate the malignant of HCC from Contrast-enhanced MR images based on the analysis of texture features, in order to contribute the procedure of HCC diagnosis. This retrospective study was approved by the local Institute of Review Board. Twenty-four MR images were acquired with a 3.0T MR scanner (Signa Excite HD 3.0T, GE Healthcare, Milwaukee, WI, USA) using eight-channel phase-array coil with a BH Ax LAVA+C(1iver acquisition with volume acceleration, LAVA) sequence. The malignant differentiation of HCCs for all the anticipated patients was pathologically verified as the ground-truth and categorized as well-differentiated and moderately HCCs. Regions of interest (ROIs) of HCCs were manually drawn on the contrast-enhanced MR image (arterial phase) for all the anticipated patients (Fig.1). The texture parameters of the average intensity value and grey level nonuniformity (GLN) were chosen as the discriminate feature to characterize the malignant of HCCs. Note that the texture parameter GLN was measured in four different directions($$$0^\circ$$$, $$$45^\circ$$$, $$$90^\circ$$$, $$$135^\circ$$$), denoted as GLN_0, GLN_45, GLN_90, and GLN_135, respectively. The average intensity value and GLN values of HCCs were expressed as the mean standard deviation. A Student $$$t$$$ test was used to differentiate the average intensity value and GLN values of HCCs. Give the quantitative measurements, the optimal threshold value of differentiate HCC malignant was determined by a receiver operating characteristic (ROC) analysis. Computer software packages (SPSS software, version 21; SPSS, Chicago, IL, USA) were used for the statistical analyses. P values less than 0.05 were considered statistically significant. The mean intensity value of the well differentiated HCCs (1324.95$$$\pm$$$397.47) was significantly larger than that of the moderately differentiated HCCs (701.75$$$\pm$$$185.85) (p=0.004, unpaired $$$t$$$ test) (Table I). With regard to the texture parameter GLN, all values of GLN towards four directions of well differentiated HCCs were relatively smaller than those values of the moderately differentiated HCCs (Table I). By ROC analysis, the optimal threshold value of the average intensity was 739 for optimal sensitivity (100$$$\%$$$, 13/13) and specificity (81.8$$$\%$$$, 9/11) of HCC differentiation. The optimal threshold values of the GLN in four different directions were 38.72, 66.59, 31.51 and 77.06, for optimal sensitivity (90.9$$$\%$$$, 90.9$$$\%$$$, 81.8$$$\%$$$, 81.8$$$\%$$$) and optimal specificity (92.3$$$\%$$$, 92.3$$$\%$$$, 100$$$\%$$$, 100$$$\%$$$), respectively (Table 2). The area under the ROC curve (AUC) for the average intensity was 0.951 (Fig.2a), followed by 0.823, 0.916, 0.888 and 0.930 for the GLN in four different directions (Fig.2b).Our study demonstrated that the texture feature (average intensity value and GLN) of HCC in contrast-enhanced MR images was a good predictor to characterize the malignant of HCC. The reason that we chose contrast-enhanced MR images for texture extraction was that 3.0T MR with Gd-EOB-DTPA provided higher-resolution images of HCCs, especially for their vascular. By quantitatively comparing the texture parameters in well differentiated and moderately differentiated HCCs, our quantitative results imply the values of average intensity remarkably decreased and GLN significantly increased according to the increasing degree of malignant for HCCs. These findings of the texture analysis in contrast enhanced MR images for HCCs may reflect changes of malignant as the HCC develops.