Objective To explore the value of radiomics features based on digital breast tomosynthesis(DBT)images for preoperative prediction of triple-negative breast cancer(TNBC).Methods DBT images and clinical data of breast cancer patients were retrospectively analyzed,and a total of 198 patients were included according to the inclusion and exclu-sion criteria,which were classified into 146 cases of non-triple-negative breast cancer(NTNBC)and 52 cases of TNBC ac-cording to the pathological findings.The regions of interest(ROI)were delineated using 3D Slicer software,and radiomics features were extracted using Python,t-test and LASSO algorithm were used to select the features,and univariate analysis was used to select the clinical and imaging features.The models were constructed by support vector machine(SVM),and the ar-ea under the curve(AUC)and other indicators were used to evaluate the prediction performance of the models.Delong test was used to compare the AUC of each model.Results The age of menarche,mass with calcification,and BI-RADS grade(P＜0.05)were selected to establish the clinical and imaging model,four important radiomics features were selected to es-tablish the radiomics model,and features of the two groups were combined to construct a combined model.The AUCs of the above models in training set were:0.818,0.886 and 0.896,and in validation set were:0.785,0.866 and 0.870.Conclu-sion The radiomics features based on DBT images can effectively predict TNBC preoperatively,providing a new method for the clinical diagnosis of TNBC.
Digital breast tomosynthesisTriple-negative breast cancerRadiomicsSupport vector machine
NETL
NSTL
万方数据
