| Objective: To investigate the ability of MRI-DCE-based intratumoral and peritumoral radiomics features in identifying benign and malignant in breast BI-RADS category 4tumors,and further analyze the feasibility of DCE-MRI-based intratumoral and peritumoral radiomics models in preoperatively predicting molecular subtypes of breast cancer.Methods: Retrospective analysis of 404 patients（327 malignant and 77 benign）with breast tumors confirmed by surgical pathology from January 2016 to March 2021 at the First Affiliated Hospital of Bengbu Medical College（Center 1）and 95 patients with breast cancer from December 2018 to August 2019 at Sir Run Run Shaw Hospital,Zhejiang University（Center 2）.The 191 patients with breast single mass diagnosed as BI-RADS 4 by breast MRI in Center 1 were screened for inclusion in Group A,including 77 benign and 114 malignant cases.The included cases were randomly divided into training and test groups in the ratio of 8:2.The one-layer image with the largest area of the lesion of the patients DCE-MRI images were selected to outline the ROI,and automatically conformal extrapolated by 5 mm;the 422 invasive ductal breast carcinoma（IDBC）patients with pathological immunohistochemistry were screened for inclusion in Group B,including 327 cases in center 1 and 95 cases in center 2.Patients from center 1 were randomly divided into training and test groups in the ratio of 7:3,and patients from center 2 were used as the external test group.3D intratumoral ROIs were manually outlined layer by layer on DCE-MRI images,and automatically conformal extrapolated by 2 mm,4mm,6mm and 8mm.The intratumoral and peritumoral radiomics features of groups A and B were automatically extracted,and the statistical and machine learning methods were used to select the optimal radiomics signatures of groups A and B.Using logistic regression and SVM as classifier,radiomics models were developed to identify benign and malignant of BI-RADS 4breast tumors（Group A-based）and to predict molecular subtypes of IDBC（Group B-based）.The clinical features were screened out as independent clinical independent risk factors by univariate and multivariate logistic regression to develop clinical model.The combined identification/prediction model was established by combining the intratumoral and peritumoral radiomics signatures and clinical independent risk factors and visualized by nomogram.The performance of the model was evaluated by receiver operating characteristic curve（ROC）of each model in group A and group B.The stability of the model was assessed by 10-fold cross validation.And the value of the model was evaluated by calibration curves and decision curves.Results:1.Preoperatively identifying the benign and malignant of breast tumor: The combinedidentification model based on intratumoral and peritumoral radiomics signatures andclinical characteristics showed better diagnostic performance compared with othermodels.The AUC,sensitivity,specificity in the training set were 0.932,90.1%,86.9%,respectively;and the test set were 0.875,95.7%,62.5%,respectively.2.Preoperatively predicting the molecular subtypes of IDBC: In the tasks of predictingHR-positive vs non-HR-positive,HER2-enriched vs non-HER2-enriched,andTNBC vs non-TNBC,the combined prediction model obtained the optimalperformance compared to the individual model with AUCs of 0.838,0.848 and 0.930in the training set,respectively;0.827,0.813 and 0.879 in the internal test set,respectively;0.791,0.707 and 0.852 in the external test set,respectively.Conclusion:1.The combined identification model based on the intratumoral and peritumoralradiomics signatures and clinical characteristics of DCE-MRI has good performanceand stability in identifying the benign and malignant in BI-RADS 4 breast tumors,which can provide guidance for clinical decision in a non-invasive manner.2.The DCE-MRI-based intratumoral and peritumoral radiomics model can be used asa non-invasive tool for preoperatively predicting of molecular subtypes in breastinvasive ductal carcinoma. |
PDF Full Text Request