| Objective:A radiomics model based on MR was constructed and combined with clinical features to achieve preoperative noninvasive diagnosis of prostate cancer with a PSA of4~10 ng/ml and a PI-RADS score of ≥3,so as to further guide clinical decision-making.Methods:Review the analysis from June 2018 to January 2022 at Shanxi Medical University First Hospital(first institution)and from January 2021 to April 2022 for a prostate MRI before Shanxi Provincial Hospital of Traditional Chinese Medicine(the second institution).Patients and clinical data of patients with pathological data and PSA levels between 4~10ng/ml and PI-RADS score ≥3 points.Patients from the first organ are randomly divided into a training set and a test set in a 7:3 ratio to build and test the model.The patients of the second institution are used as external verification sets.The T2 WI,ADC and DCE of each case manually sketched the region of interest in ITK-SNAP,and the generated VOI and original images were saved respectively.Use the Python platform to extract the characteristics of the group.FAE software completes data pre-processing,feature screening,and group learning model construction.Individual factors and multifactor logistic regression analysis for clinical characteristics and prostate cancer correlation,independent predictive factors are determined for the construction of clinical models.Through the AUC value evaluation model performance,the best group learning model is selected and the clinical model is jointly constructed to build a comprehensive model.And to verify the generalization of each model through external verification sets.Finally,the comprehensive model line diagram is drawn through R software,and the degree of fitting and clinical application value of the calibration curveand decision-making curve are used.Results:A total of 232 patients were included in this study,among them,a total of 59 patients with prostate cancer and 123 patients with benign prostatic hyperplasia were collected in the first institution.A total of 23 patients with prostate cancer and 27 patients with benign prostatic hyperplasia were collected in the second institution.Among the 182 patients of the first institution,127 training sets and 55 test sets;the second institution was 50 cases as external verification sets.After feature screening and verification,the T2WI+ADC+DCE multi-parameter arrays based on the six largest characteristics of the six characteristics of the six most characteristics is the best performance.After statistical analysis,age and Pi-Rads V2.1 score are related to prostate cancer,which is a clinical predictive factors for prostate cancer.The comprehensive model is based on T2WI+ADC+DCE screening.The combined age and PI-RADS V2.1 score are jointly constructed.In the final comprehensive model,group learning model,and clinical models,the AUC is 0.933,0.915,0.802;0.887,0.872,0.677;0.866,0.812,0.709,of which the diagnosis effectiveness of the comprehensive model is the best.The calibration curve results of the nomogram showed that the prediction results were in good agreement with the pathological results.The decision curve shows that the comprehensive model has good clinical application value.Conclusion:The comprehensive model based on multi-parameter MR-omics model combined with clinical model can be used to non-invasively diagnose prostate cancer with PSA "gray zone" and PI-RADS score of 3 or above before surgery,so as to further guide clinical decision-making. |
PDF Full Text Request