MRI-based diffusion weighted imaging and diffusion kurtosis imaging grading of clear cell renal cell carcinoma using a deep learning classifier.

Clear cell renal cell carcinoma (ccRCC) is a malignant tumor, originating from the renal epithelium, and accounts for ~85% of RCC cases. The present study aimed to validate the efficacy of an MRI deep learning (DL) model to preoperatively predict the pathological grading of ccRCC. Therefore, a DL algorithm was constructed and trained using diffusion weighted imaging (DWI) and diffusion kurtosis imaging (DKI) sequence images. Subsequently, the apparent diffusion coefficient maps from DWI, as well as axial kurtosis (Ka), fractional anisotropy, radial kurtosis (Kr), mean kurtosis (MK) and mean diffusivity maps from DKI were calculated. The VGG-16 model was selected as the backbone architecture to validate the DL model. Based on the inclusion and exclusion criteria, a total of 79 patients with ccRCC, including 40 low- and 39 high-grade cases, were prospectively evaluated. Among the different image parameters, mean MK achieved the highest accuracy, with a precision of 81.48%, F1-score of 76.04%, recall of 74.08% and accuracy of 76.04%, followed by Kr, with values of 75.51, 75.36, 75.42 and 75.36%, respectively. Ka had precision, recall, F1-score and accuracy values of 81.39, 71.81, 68.31 and 71.81%, respectively. Overall, the results of the current study revealed that the established DL model, as a non-invasive algorithm based on MRI sequences, could accurately predict the pathological grading of ccRCC. Therefore, these findings highlighted the potential of this method to guide individualized treatment decisions for patients with ccRCC.