The Application Of Radiomic Features Based On Magnetic Resonance Imaging In Identifying Benign And Malignant Breast Tumors And Molecular Subtypes Of Breast Cancer

Objective:1.To assess the diagnostic performance of radiomic features based on diffusion-weighted imaging(DWI)and dynamic contrast agent-enhanced(DCE)MRI in breast tumors classification.2.To investigate the correlation between radiomic features based on DWI and DCE MRI and molecular subtypes of breast cancer.Materials and Methods:From June 2015 to June 2016,a total of 153 female patients were enrolled,including 71 cases of breast fibroadenoma(benign)and 82 cases of breast cancer(malignant).All patients underwent diffusion-weighted imaging(DWI)and DCE MR at a 1.5-T MR imager,and ADC images were automatically generated in the workstation.The ADC and IER values of the breast tumors were measured by two experienced radiologists.Manual segmentation was performed for the tumors on ADC and DCE images,then texture features were extracted from each region of interest(ROI).After reduction,the number of radiomic signatures selected from the ADC images and DCE images were both five.Patients were randomly assigned to training set and validating set with a ratio of 60%vs.40%.Statistical analysis was performed using ADC values,IER values and radiomic signatures.To identify the conventional parameters and radiomic features that optimally discriminated malignant from benign breast tumors,univariate analysis and logistic regression analysis were performed.Four molecular subtypes of breast cancer were classified by immunohistochemical detection of pathological specimens,including Luminal A,Luminal B,human epidermal growth factor receptor 2(HER2)overexpression and triple negative(TN).Univariate analysis and logistic regression analysis were performed,predicting models were established to classify molecular subtypes groups.The receiver operating characteristic(ROC)curves were plotted,and the area under the ROC(AUC)were calculated to compared the diagnostic performance of each model.The Hosmer-Lemeshow test were perfrmed to test the goodness of model fitness.Results:1.Univariate analysis showed that ADC values,IER values and all radiomic signatures except R_variance in benign and malignant breast tumors were significantly different.The ADC,D_entropy,L_min,L_mean and R_sum variance values were significantly higher in benign breast tumors compared with malignant ones,other parameters were the opposite.The results of logistic regression revealed that the AUCs of ADC and IER values alone for classifying benign and malignant breast tumors in the training group and the validating group were 0.96 vs.0.94 and 0.79 vs.0.71,respectively.And the AUC of the models for classifying benign and malignant breast tumors in the training group and the validating group based on ADC and DCE images were 0.98 vs.0.97 and 0.95 vs.0.91,respectively.Delong analysis showed that the differences of AUC of IER value were significant lower than other models(P<0.05),while there were no significant differences between ADC value and the models based on ADC or DCE images(P>0.05),whether in the training group or the validation group.2.Univariate analysis showed that the overall distribution of ADC value,IER value,L_min,R_IMC1,L_mean and L_homogeneityl among four molecular subtype groups were not significantly different(P>0.05),while the other six radiomic signatures were significantly different(P<0.05).The results of univariate logistic regression analysis showed that the AUC values of ADC and IER values were both less than 0.70 for classifying molecular subtypes,while at least one radiomic signature had AUC greater than 0.70 for identifying each molecular subtypes,and AUC of L_autocorrelation achieved the highest value of 0.941 in identifying TN and non-TN subtypes.Models were established by multivariate logistic regression analysis,results showed that the AUCs for classifying Luminal A and non-Luminal A,Luminal B and non-Luminal B,TN and non-TN subtypes were 0.786 and 0.733 And 0.941,respectively.The Hosmer-Lemeshow test showed that the P values of all models were large than 0.10,indicating that these models were all fitted good.Conclusion:1.ADC value,as one of the traditional parameters of breast MRI,was valuable in classifying benign and malignant breast tumors,while IER value was valueless.Radiomic features based on DWI and DCE MRI can be effective supplements to the traditional quantitative parameters of MRI(especially for IER value)in classifying benign and malignant breast tumors.2.The traditional parameters of breast MRI(including ADC and IER value)was valueless in classifying molecular subtypes of breast cancer,while radiomic features based on DWI and DCE MRI were valuable in classifying molecular subtypes of breast cancer,and may be potential biomarkers.