| Objective: By analyzing the MRI radiomics features and clinical related factors of patients with advanced prostate cancer,to identify a clinical-radiomics model for early prediction of rapid progression to CRPC after endocrine therapy,to provide more individualized treatment options for prostate cancer patients.Methods: A retrospective collection of patients with advanced prostate cancer who underwent MRI examination with endocrine therapy as initial treatment from June 2013 to June 2020 was collected.According to whether castration resistance occurred within 2 years,it was divided into rapid progression group and slow progression group.Using 3D Slicer software to manually outline ROIs layer by layer on axis T2 WI,DWI,and ADC plots.The radiomics package(Py Radiomics 3.0)of python(version 3.8.8)was used to extract radiomics features in the ROI of each lesion.Randomly dividing patients into training and testing groups in a 6:4 ratio.In the training group,using the selection of variance method,Spearman correlation analysis,Mann-Whitney rank sum test,and LASSO to sifting the radiomics features.According to the LASSO regression model,calculating the rad-score,and then establishing the radiomics model.Using univariate and multivariate logistic regression analysis to evaluate clinically relevant independent media actors,and then establishing clinical models.Basing on rad-scores and clinically relevant independent media actors,create multiple logistic regression partner models and a nomogram is drawn,and then the model is evaluated in the training group and verified in the test group.Using the ROC curve to assess the diagnostic performance.Appling the Hosmer-Lemeshow goodness-of-fit test and calibration curve to assess the calibrationability.Using decision curve analysis(DCA)was to assess the clinical net benefit.Results: In this study,163 cases were collected,83 in the rapid progression group and 80 in the slow progression group.The results of univariate and multivariate logistic regression analysis show that lymph node metastasis,anterior prostate fat thickness and Gleason score are independent clinically relevant risk factors(P<0.05)for rapid progression to CRPC,and bone metastasis is constructed with a clinical model of0.800 in the training group and 0.806 in the test group.Based on the radiomics integration of T2 WI,DWI,ADC and T2WI+DWI+ADC,the AUCs of each model were 0.864,0.823,0.794 and 0.902 in the training group and 0.818,0.795,0.791 and 0.852 in the test group,respectively.The AUC value of the prediction model based on T2WI+DWI+ADC in the training group and the test group was higher than that of the prediction model based on T2 WI,DWI and ADC sequences.Combining the radiomics characteristics and clinically relevant independent risk factors,a clinical-radiomics multivariate logistic regression joint model is constructed,and a nomogram is drawn.In the training group and the test group,the AUC of the joint model are respectively 0.931 and 0.904,which are good than those of the clinical model and the radiomics model.The Hosmer-Lemeshow goodness test show that the joint model have nice goodness of fit(χ2=11.057,P=0.399).The calibration curve indicates that the joint model has good predictive ability.The DCA curve suggests that the combined model has a higher net clinical benefit in both the training and testing groups.Conclusion: The combined model combining clinical independent risk factors and MRI radiomics features of primary prostate cancer has a good diagnostic efficacy in early predicting the rapid progression of CRPC after endocrine therapy in patients with prostate cancer,which is helpful to make more valuable clinical decisions for patients. |
PDF Full Text Request