| Objectives:(1)Given there is no effective clinical indicator to evaluate the prognosis of patients with ccRCC.This research was designed to predict risk stratification of patients with ccRCC based on imaging information and prognosis-related genes.(2)In SRMs,radiomics as a pivot,constructing a nomogram for preoperative differentiation of ccRCC and lp-AML.Material and methods:(1)Prognosis-related research:Gene-related information and enhanced CT image of 96 patients with clear cell renal cell carcinoma(ccRCC)were obtained from TCGA and TCIA respectively.The 96 patients were screened for differential genes between normal individuals and patients with tumors at transcriptional level.And then,the 96 patients were randomly allocated into training and testing groups with a 1:1 ratio.Prognosis-related genes were selected by COX proportional hazards regression to construct a genetic model.Radiomics features were extracted from the cortical phase and the most valuable radiomics features were identified to construct the radiomics model.Finally,34 patients with ccRCC in our institution were selected as external validation to reveal the robustness of the models and the better performance of the combination of genetic model and radiomics model.(2)Differential diagnosis related research:Additional information was collected about patients with SRMs.87 patients with small renal masses(SRMs)with tumor size≤4 cm by pathologically identified were involved in this retrospective study,including 54 patients with ccRCC and 33 patients with lp-AML.All patients received unenhanced phase(UP),corticomedullary phase(CMP)and nephrographic phase(NP)CT examinations and radiomics features were extracted from the CT images.Due to the slight variation in renal function and clinical requirements of patients,the duration of excretory phase scans varies from patient to patient and excretory phase images were not used in this study.Radiomics features were selected and three radiomics models were constructed,including UP model,CMP model and NP model.The corresponding radiomics scores were calculated.And then screened out the optimal model.Statistically significant clinical variables were used to construct clinical model,the selected clinical variables and radiomics signature were used to construct combined model and nomogram.ROC curves were used to evaluate the diagnostic efficacy of clinical model,radiomics model and combined model.Calibration curves and clinical impact curves were used to evaluate the reliability and clinical value of nomogram respectively.Results:(1)Prognosis-related research:In terms of three-year and five-year OS for the general data of TCGA,the AUCs of the radiomics model were 0.70(95%CI:0.66-0.94),0.75(95%CI:0.69-0.89),the AUCs of the genetic model were 0.86(95%CI:0.75-0.97),0.78(95%CI:0.65-0.92)and the AUCs of the combined model were 0.89(95%CI:0.80-0.93),0.88(95%CI:0.82-0.97).In terms of three-year and five-year OS of external validation group,the AUCs of the radiomics model were 0.81(95%CI:0.73-0.94),0.76(95%CI:0.67-0.88),the AUCs of the genetic model were 0.83(95%CI:0.74-0.90),0.76(95%CI:0.68-0.89)and the AUCs of the combined model were 0.95(95%CI:0.70-0.98),0.80(95%CI:0.77-0.94).The information between macroscopic image and microscopic genes complemented each other which allows for synergistic effects.The K-M curves in the external validation group showed that the genetic model was still able to stratify patients at risk more accurately and the radiomics model also showed a general trend.It also suggested that the genetic model performed more robustly than the radiomics model.(2)Differential diagnosis related research:The performance of the UP-model was better than the other two models and there was strong collinearity among the UP,CMP and NP models.By Logistic regression analysis,three clinical variables with statistical significance were selected:whether the tumor enhancement degree in corticomedullary phase was higher than renal cortex,morphology of the interface and quantification of perivisceral fat at the umbilical level,which were combined with UP-score to construct a combined model.AUCs of clinical model,radiomics model and combined model were 0.70(95%Cl:0.51～0.90),0.85(95%Cl:0.70～0.99)and 0.94(95%Cl:0.88-1.00)respectively.The calibration curve showed that the predicted values of nomogram corresponded well the actual pathological results.The clinical impact curve shows that nomogram had good clinical application value.Conclusion:(1)Prognosis-related research:Imaging may provide information beyond the genetic dimension,macro and micro information are complementary.The combination of the two makes the data more comprehensive and stereoscopic.Genetic model was tended to be more robust than radiomics model.(2)Differential diagnosis related research:The nomogram based on preoperative CT imaging characteristics and clinical variables has good efficacy in the differential diagnosis of ccRCC and lp-AML in SRMs. |
PDF Full Text Request