Research And Implementation Of A Deep Learning Based Grading Model For Clear Cell Renal Cell Carcinoma

Clear cell renal cell carcinoma is the most common subtype of renal cell carcinoma,which accounts for 10% to 15% of renal malignancies and 2% to 3% of malignant tumors in adults.Clear cell renal cell carcinoma pathology grade identification is essential to both monitoring patients’ conditions and constructing individualized subsequent treatment strategies.However,biopsies are typically used to obtain the pathological grade,entailing tremendous physical and mental suffering as well as heavy economic burden,not to mention the increased risk of complications.In order to provide a safer,reliable and non-invasive detection method for patients with clear cell renal cell carcinoma,thesis constructs a pathological grade determination method for renal clear cell carcinoma based on patient CT images using deep learning as a tool.The application of traditional pre-training methods in the medical image field will result in little correlation between the pre-training task and the actual training task,thus failing to effectively improve the feature extraction capability of the network model.Therefore,the thesis innovatively proposes a self-supervised pre-training method,which achieves the same CT images used in the pre-training task and the actual training task without leaking the original semantics of the CT images.In addition,for the noise problem in the dataset,the thesis proposes a noise correction strategy for the loss function to mitigate the impact of noisy data on the network model? For the class imbalance problem in the dataset,the thesis proposes a simple and effective sample assignment method based on the number of images? finally,according to the model ensemble method based on model performance proposed in the thesis,combining different network models makes the pathological grade prediction method constructed in the thesis more reliableThe thesis demonstrates through extensive experiments that the prediction method proposed in the thesis can accurately identify CT images of patients with different pathological grades of clear cell renal cell carcinoma.With the method proposed in the thesis,the network model can finally achieve 82.0% accuracy,85% sensitivity,75% specificity,and 88.2% area under the curve.The promising diagnostic performance indicates that our deep learning framework is an effective,non-invasive and labor-saving method for decoding CT images,offering a valuable means for clear cell renal cell carcinoma grade stratification and individualized patient treatment.