The Predictive Value Of Radiomic Based On Mammography In The Differential Diagnosis Of Triple Negative And Non-triple Negative Breast Cancer

Aim The purpose of this study is to differential diagnosis the molecular subtypes of triple negative breast cancer(TNBC)and non-triple negative breast cancer(nTNBC)by making use of radiomics features based on mammography.Material and Methods In this retrospective study,391 mammograms of invasive breast cancer patients were collected from the Picture Archiving and Communication Systems(PACS)of Zhongda Hospital Affiliated to Southeast University.According to the immunohistochemical results of all patients,breast cancer molecular subtype was carried out,including 51 cases of TNBC and 340 cases of TNBC.The region of interest(ROI)was segmented by manual segmentation,and mammographic features of each lesion were recorded at the same time.The radiomics features and mammographic features extracted from each ROI were selected and reduced by F test.5-fold cross validation and logical regression was used to construct models.The model performance was evaluated by the area under receiver operating characteristic curve.Result From the original image,18 first-order features,14 shape features and 22gray-scale co-occurrence matrices features were extracted,totally 54 radiomics features.Finally,20 radiomics features were selected to build a model,including 7first-order features,4 shape features and 9 gray level co-occurrence matrices features.(1)For the differential diagnosis of TNBC and nTNBC,the area under ROC curve of the radiomics model based on mammography,mammographic features model,and radiomics combined with mammographic features model was 0.524,0.627 and 0.601,respectively,and there was no statistical difference in AUC value of the three models(P >0.05).(2)In the differential diagnosis of mass-forming TNBC and nTNBC,the area under ROC curve of the radiomics model based on mammography,mammographic features model,and radiomics combined with mammographic features model was 0.518,0.682 and 0.661,respectively.There was a statistically significant difference between the radiomics model based on mammography and radiomics combined with mammographic features model(P <0.05).(3)In the differential diagnosis of non-mass TNBC and nTNBC,the area under ROC curve of the radiomics model based on mammography,mammographic features model,and radiomics combined with mammographic features model was 0.539,0.498 and 0.461 respectively,there was no significant difference between them(P >0.05).Conclusion The study shows that radiomics based on mammography combined with mammographic features model has a certain differential diagnostic performance for mass-forming TNBC and nTNBC.