Correlation of radiomic features based on diffusion weighted imaging and dynamic contrast-enhancement MRI with molecular subtypes of breast cancer

Objective To explore the relationship between radiomics signatures based on DWI and dynamic contrast-enhanced MRI (DCE-MRI) and molecular subtypes of breast cancer. Methods A retrospective analysis of 79 female breast cancer patients, with single mass, clear molecular subtypes and preoperative breast MRI scanning (obtaining DCE-MRI and ADC images), of Guangdong General Hospital from June 2015 to June 2016, were performed. Traditional quantitative parameters, including ADC value and initial enhancement rate (IER), were recorded. Texture analysis were performed on ADC map and DCE map, with manual segmentation and extraction of radiomic features, and Manual segmentation was performed on ADC map and DCE map, radiomics features were extracted and 10 radiomics signatures were finally selected after dimension reduction. 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 logistic regression analysis was used for assessing the performance of ADC values, IER values and radiomics signatures to independently predict molecular subtypes groups. Multivariate logistic regression analysis was performed to establish predicting models, then receiver operating characteristic curves (ROC) were drawn and areas under ROC curve were calculated to compare the diagnostic performance of each model. The Hosmer-Lemeshow test was performed to test the goodness of model fitness. Results There were 29 cases of Luminal A, 39 cases of Luminal B, 5 cases of HER2 overexpression and 6 cases of TN breast cancer patients. Univariate logistic regression analysis was used to assess the ability of traditonal MRI parameters of ADC and IER values and ten of the radiomics siganitures in classifying molecular subtypes, results showed that the AUC values of ADC and IER values, were both less than 0.70 (range 0.516 to 0.605), which indicated valueless;at least one radiomic signature had AUC greater than 0.70 when identifying each molecular subtype, and AUC of DCE_L_G_2.5_autocorrelation achieved the highest value of 0.941 in identifying TN and non-TN subtypes. Multivariate logistic regression analysis were performed to obtain the best model, 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 larger than 0.10 (0.156, 0.204 and 0.820, respectively), indicating that there was no significant difference between the predicted and observed values of each model established, these models were all fitted good. Conclusion The radiomics features based on ADC map and DCE map can help to identify the molecular subtypes of breast cancer, especially for the identification of TN type breast cancer. Key words: Breast neoplasms; Radiomics; Texture analysis; Molecular subtypes; Prediction