Research On Nuclear Grading And Staging Diagnosis Model For Clear Cell Renal Carcinoma Using Enhanced CT

With the aggravation of population aging,the incidence of cancer in China has been increasing year by year,with 77,410 new cases of kidney cancer in 2022.Clear cell Renal Cell Carcinoma(ccRCC)is the most common subtype of kidney cancer,accounting for more than 75%of cases,and its main prognostic evaluation criterion is pathological nuclear grading.Preoperative pathological nuclear grading evaluation of ccRCC tumors is a key factor in improving the early screening rate of kidney cancer and improving patient prognosis.Medical image analysis based on artificial intelligence technology can quantitatively describe the intratumoral heterogeneity,providing a solution for early non-invasive auxiliary diagnosis of ccRCC.Currently,there are still difficulties in the clinical application of preoperative non-invasive nuclear grading-assisted diagnosis of ccRCC based on enhanced CT images:the abundance of enhanced CT images obtained from multiple scans may result in redundant information for diagnosing nuclear grading,leading to low efficiency of clinical assisted diagnosis.Furthermore,ccRCC tumors and lesions of specific nuclear grades may have specific enhancements in different phases of the image.Therefore,this paper constructs a ccRCC nuclear grading classification model based on radiomics and deep learning,and compares the performance of the model in diagnosing nuclear grading in the corticomedullary phase(CMP),nephrographic phase(NP),and excretory phase(EP)of multiphase enhanced CT,analyzing the differences in imaging features of different nuclear grades in the three enhancement phases,and obtaining the optimal enhancement phase for diagnosing nuclear grading as NP.The main work completed in this paper is as follows:(1)A radiomics diagnostic model of ccRCC pathological nuclear grading was constructed based on the three phases of enhanced CT,and the model achieved better preoperative diagnosis of nuclear grading in NP.By extracting radiomics features and training the model on CMP,NP,and EP,the optimal feature subset specific to the three enhancement phases for diagnosing nuclear grading was screened out,which could effectively guide clinical selection of multiphase scan images.Meanwhile,the model based on NP had good robustness and generalization ability in differentiating ccRCC pathological nuclear grading(SVM:AUC=0.82±0.05,RF:AUC=0.82±0.05,XGBoost:AUC=0.81±0.04).(2)Deep learning diagnostic models of ccRCC pathological nuclear grading were constructed for the three phases,and transfer learning optimization training was carried out.Finally,the transfer learning model based on NP obtained the best classification performance and outperformed the radiomics model.By randomly initializing training and transfer learning optimization training of deep learning nuclear grading classification models based on CMP,NP,and EP,the best-performing model was a transfer learning model based on NP using ResNet18(AUC=0.87),which achieved better performance than the radiomics model in diagnosing nuclear grading and has clinical value as an auxiliary diagnostic tool.In summary,this paper proposes that the optimal enhancement phase for preoperative noninvasive evaluation of ccRCC nuclear grading is the nephrographic phase(NP),and the transfer learning model based on NP can more effectively evaluate pathological nuclear grading,providing a reliable preoperative nuclear grading-assisted diagnostic tool for clinical practice.