The Value Of DCE-MRI Radiological Features Combined With Radiomics Signature In Predicting Molecular Subtypes Of Invasive Breast Cancer

Objective:To explore the value of dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI)radiological features combined with intratumoral and peritumoral radiomics signature in predicting molecular subtypes of invasive breast cancer.Methods:The clinical data and preoperative MRI of 103 patients with pathologically confirmed invasive breast cancer were retrospectively analyzed.According to the immunohistochemical results,breast cancer was divided into four molecular subtypes:Luminal A,Luminal B,HER2 overexpression and triple negative,and the patients were randomly divided into training set(n=71)and validation set(n=31)according to the ratio of 7:3.Firstly,according to the Breast Imaging Reporting and Data System(BI-RADS),ten DCE-MRI-related radiological features were included and analyzed,and the time-signal intensity curve(TIC)was drawn and it’s semi-quantitative parameters were measured.In the training set,univariate and multivariate Logistic regression were used to select statistically significant clinical and radiological features for predicting molecular subtypes,and a clinical-radiological features model(CRM)was established.Then 3D-Slicer and PyRadiomics software were used to extract three-dimensional radiomic features in the second enhancement phase of DCE-MRI in the tumor and the 5 mm area around the tumor.ANOVA and least absolute shrinkage and selection operator(LASSO)regression were used for feature selection,and radiomic signatures were selected and the radiomics score(radscore)were calculated.Finally,the Logistic regression classifier was used to establish the intratumoral radiomics model(IRM),peritumoral radiomics model(PRM)and clinical-radiological-radiomics features combined model(CRRM)for predicting molecular subtypes of breast cancer.Visualizing the combined models by nomogram.Receiver operating characteristic curve(ROC)was drawn for each model and the area under the curve(AUC)was calculated to evaluate the predictive performance of the model,and the AUC of each model were compared by Delong test.The calibration curve was used to test the goodness of fit of the model.Results:(1)Univariate and multivariate Logistic regression analysis showed that in radiological features,the mass margin was independently correlated with Luminal A(OR=0.361,P=0.027)and Luminal B(OR=2.801,P=0.006)breast cancer,time to peak(TTP)(OR=0.989,P=0.037)was independently correlated with triple-negative breast cancer.HER2-overexpressing breast cancer had no statistically significant radiological features.(2)Intratumoral radiomic signatures of Luminal A,Luminal B,HER2 overexpression,and triple negative breast cancer included 4,2,5,2 radiomic features,and peritumoral radiomic signatures contained 6,1,7,2 radiomic features,respectively.(3)For predicting Luminal A vs.non-Luminal A breast cancer,Luminal B vs.non-Luminal B breast cancer,triple-negative vs.non-triple-negative breast cancer,the AUCs of CRM in the training set and validation set were 0.738,0.711,0.727 and 0.643,0.526,0.654,respectively.For predicting Luminal A vs.nonLuminal A breast cancer,Luminal B vs.non-Luminal B breast cancer,HER2 overexpression vs.non-HER2 overexpression breast cancer,triple-negative vs.nontriple-negative breast cancer,the AUCs of IRM in the training set and validation set were 0.813,0.683,0.805,0.870 and 0.821,0.620,0.485,0.646,respectively;the AUCs of PRM in the training set and validation set were 0.915,0.678,0.916,0.739 and 0.821,0.654,0.654,0.569,respectively;the AUCs of CRRM in the training set and validation set were 0.919,0.756,0.916,0.906 and 0.833,0.577,0.654,0.700,respectively.Conclusion:There is a certain correlation between DCE-MRI radiological features and molecular subtypes of breast cancer.Models based on DCE-MRI radiological features,intratumoral and peritumoral radiomics signature have certain potential value in predicting molecular subtypes of breast cancer,especially the combined model has the highest performance in predicting Luminal A and triplenegative breast cancer.