Different Neural Networks Based On CT-enhanced Images In Predicting The Pathological Grade Of Renal Clear Cell Carcinoma

[Objective]To investigate the feasibility and diagnostic performance of different neural networks based on CT-enhanced images in predicting the pathological grade of renal clear cell carcinoma.[Materials and methods]134 patients with clear cell renal cell carcinoma admitted to our hospital from January 2015 to December 2020 and 77 patients with clear cell renal cell carcinoma in the TCIA database were retrospectively collected.All patients underwent standard abdominal enhancement scanning procedures before surgery,and all patients had WHO/ISUP pathological grading data after surgery.179 Low-grade clear cell renal cell carcinoma and 32 high-grade renal clear cell carcinoma were included.Preoperative thin-slice enhanced CT images of kidneys were downloaded from PACS,and the region of interest was delineated using 3D Slicer software.The delineated images are preprocessed,and 9 different CNN networks are applied to build deep learning models.The ROC curve was drawn to evaluate the model,and its accuracy,precision,sensitivity,specificity,F11 score and area under the curve(AUC)were calculated.[Results]The test accuracies of the 9 CNN networks,Googlenet,mobilenetv2 and Xception networks tested in this experiment are:67.71±3.168,79.73±2.813 and 81.73±2.244.Its AUC values were 0.7059±0.0202,0.8889±0.03479 and 0.9129±0.01395.The test set test accuracy of resnetl8 and resnet50 networks are:79.01±0.7128,73.25±6.775;AUC values are 0.9085±0.0009493,0.846±0.05762.The test set test accuracies of darknet19,vgg16 and vgg19 networks are:74.96±2.744,75.64±1.48 and 72.9±0.6587;AUC values are 0.8741±0.01933,0.8817±0.01918 and 0.8295±0.04467.The test set test accuracy of densenet201 network is:85.14±1.56,and the AUC value is 0.9512±0.008386.[Conclusion]This study shows that the deep learning convolutional neural network densenet201 based on CT-enhanced images is feasible to predict the pathological grade of renal clear cell carcinoma,and has high diagnostic performance.