In order to accurately grade the nuclei of renal clear cell carcinoma in whole slide images and improve the treatment and prognosis of renal cancer, an International Society of Urological Pathology nuclear grading method based on CSFNet for ccRCC pathological images was proposed. In CSFNet, the semantic information of different stages was fused through multi-scale channel information splicing, thereby extracting more shallow features without losing depth information and subsequently achieving better classification performance. The renal histological sections of 90 patients were collected in the experiment. Afterwards, the WSI images were divided into training set and test set in a ratio of 4∶1 after being cut and enhanced. The CSFNet convolutional neural network model was then optimized iteratively on the training set and verified on the test set. The experimental results showed that the proposed CSFNet model achieved a macro-AUC of 0.975 8, a micro-AUC of 0.979 4, an accuracy of 88%, a precision of 88.36%, a recall of 86.67% and a F1-score of 87.32% for classifying ISUP Ⅰ, ISUP Ⅱ, ISUP Ⅲ and normal. Furthermore, our model was superior to other traditional classification network models, which proved that the proposed ISUP nuclear grading model for ccRCC had satisfied diagnostic effectiveness and potential clinical application value.