Can quantitative CT texture analysis be used to differentiate subtypes of renal cell carcinoma?

To investigate whether computed tomography (CT) texture analysis (TA) can be used to differentiate non-clear-cell renal cell carcinoma (non-ccRCC) from clear-cell RCC (ccRCC) and classify non-ccRCC subtypes.

One hundred ccRCC and 27 non-ccRCC (12 papillary and 15 chromophobe) were analysed. Texture parameters quantified from multiphasic CT images were compared for the objectives. Receiver operating characteristic (ROC) analysis was performed and the area under the ROC curve (AUC) was calculated. The optimal discriminative texture parameters were used to produce support vector machine (SVM) classifiers. Diagnostic accuracy and 10-fold cross-validation was performed.

Compared to ccRCC, non-ccRCC had significantly lower mean grey-level intensity (mean), standard deviation (SD), entropy, mean of positive pixels (MPP), and higher kurtosis (p<0.001). A model incorporating SD, entropy, MPP, and kurtosis produced an AUC of 0.94±0.03 with an accuracy of 87% (sensitivity=89%, specificity=92%) to identify non-ccRCC from ccRCC. Compared to chromophobe RCC, papillary RCC had significantly lower mean and MPP (p=0.002). A model incorporating SD, MPP, and skewness resulted in an AUC of 0.96±0.04 with an accuracy of 78% (sensitivity=87%, specificity=92%) to differentiate between papillary and chromophobe RCC.

CT TA could potentially be used as a less invasive tool to classify histological subtypes of RCC.