Value of CT texture analysis in the preoperative prediction of Fuhrman grade of clear cell renal cell carcinoma

Objective To detect the values of CT texture features in the preoperative prediction of Fuhrman grade of clear cell renal cell carcinoma (ccRCC). Methods The CT data of 206 patients with ccRCC admitted to the Third Affiliated Hospital of Soochow University from January 2011 to December 2016 were retrospectively analyzed, and the ccRCC cases were graded using Fuhrman grading system, including 38 cases of Grade Ⅰ, 107 cases of Grade Ⅱ, 50 cases of Grade Ⅲ and 11 cases of Grade Ⅳ. All subjects undergone plain and enhancement CT scans. There were two methods used for the extraction of texture features, including histogram (2 features: Kurtosis and Skewness) and gray-level co-occurrence matrix (6 features: Contrast, Correlation, Energy, Entropy, Homogeneity and Variance). Each texture feature during Grade Ⅰ to Ⅳ was compared using a one-way analysis of variance following the log-ratio transformation, and a Newman-Keuls test was performed for all pairwise comparisons. An independent sample t test was used to find the differences of each texture feature between low (Grade Ⅰ+Ⅱ) and high grade (Grade Ⅲ+Ⅳ) ccRCC. A Spearman Rank test was performed to quantify the correlation of each texture feature with Fuhrman grade in ccRCC. Receiver operating characteristic curve (ROC) was employed to compare the diagnostic performance of the texture features to differentiate the low grade from high grade ccRCC. Results Six texture features, including Contrast, Correlation, Entropy, Homogeneity, Variance and Kurtosis, were different during Grade Ⅰ to Ⅳ (all P 0.05). Furthermore, five textures, such as Correlation, Entropy, Homogeneity, Variance and Kurtosis, were not significantly different between Grade Ⅲ and Ⅳ ccRCC. There was no clinical application value for the features of Correlation, Energy, Entropy, Variance and Skewness with the absolute coefficients of 0.05). When those features were used to differentiate the high from low grade ccRCC, the Contrast exhibited the biggest area under ROC of 0.806 (P<0.05), followed by the Correlation of 0.641, Homogeneity of 0.687, Kurtosis of 0.668 and Variance of 0.659. Conclusion CT texture features can preoperatively predict the Fuhrman grade of ccRCC, and the Contrast will likely be the potential imaging biomarker for the clinical application. Key words: Renal neoplasms; Tomography,X-ray computed; Texture analysis; Grade