| objective:To investigate the feasibility of MRI radiomics for the prediction of Luminal and non-Luminal breast invasive ductal carcinomas.The aim is to provide objective and accurate imaging reference for preoperative prediction of Luminal classification of breast invasive ductal carcinoma.Methods:A retrospective review of 241 patients admitted to the Third Affiliated Hospital of Kunming Medical University between January 2014 and December 2016 who were surgically or biopsy-proven for breast invasive ductal carcinoma was included in this study.The patient's clinical and pathological data were collected,including the age of onset,menarche age,menopausal status,number of pregnancies,number of births,lymph node metastasis,maximum tumor size,immunohistochemistry,and FISH results.According to the results of IHC and FISH,they were divided into four molecular subtypes:LuminalA,LuminalB,HER2 overexpression,and basal-like.LuminalA and LuminalB are collectively referred to as Luminal type,and HER2 overexpression and basal-like are collectively referred to as non-Luminal types.All patients underwent routine MRI,DWI-MRI,and DCE-MRI scans before surgery or biopsy,and the routine MRI,ADC-MRI and DCE-MRI images of all patients were collected.In the ADC group,all patients were randomly divided into 138 training sets and 103 validation sets.In the DCE-MRI subtraction group,all patients were randomly divided into 162 training sets and 79 validation sets.Univariate analysis was used to compare the clinical pathological features between Luminal and non Luminal breast invasive ductal carcinoma patients of training set and validation set in the same group.The 3D-ROIs of the lesions on the ADC image and the most obvious stage image of the tumor enhancement on DCE-MRI were manually delineated.The first-order statistical features,shape and size features,texture features,and wavelet features of the region of interest were extracted by computer automatically.Using Lasso's regression model to reduce dimensions,select,and build prediction models.Finally,the validation set is used to validate the prediction models,and the receiver working curve(ROC)is used to evaluate the diagnostic efficiency of the prediction model in training set and validation set.Results:1.There was no significant difference in the age of onset,menarche age,menopausal status,frequency of pregnancy,frequency of production,lymph node metastasis and maximum diameter of tumor between Luminal and non Luminal breast invasive ductal carcinoma in in the ADC group training set(P>0.05).There was a significant difference in the age of onset and lymph node metastasis between the two types of breast invasive ductal carcinoma in the ADC group validation set(P<0.05),and there was no statistically significant difference in menarche age,menopausal status,frequency of pregnancy,number of times of production,and maximum diameter of tumor between the two types of breast invasive ductal carcinoma in the ADC group validation set(P>0.05).2.There was no significant difference in the age of onset,menarche age,menopausal status,frequency of pregnancy,frequency of production,lymph node metastasis and maximum diameter of tumor between Luminal and non Luminal breast invasive ductal carcinoma in in the DCE-MRI subtraction training set(P>0.05).There was a significant difference in the maximum diameter of tumor between the two types of breast invasive ductal carcinoma in the DCE-MRI subtraction group validation set(P<0.05),and there was no statistically significant difference in the age of onset,menarche age,menopausal status,frequency of pregnancy,requency of production and lymph node metastasis between the two types of breast invasive ductal carcinoma in the DCE-MRI subtraction group validation set(P>0.05).3.The 9 radiomics features selected from the ADC chart group were significantly associated with the Luminal classification of breast invasive ductal carcinoma in the training set and validation set of both groups(P<0.01).The AUC value of the predictive model in the ADC group was 0.853(95%Cl:0.790-0.917)on the training set,the diagnostic sensitivity was 0.800,the specificity was 0.753,the accuracy was 0.775,and the positive predictive value was 0.641,the negative predictive value 0.892,and the AUC value on the validation set was 0.764(95%CI:0.664-0.864),The diagnostic sensitivity was 0.800,the specificity was 0.712,the accuracy was 0.738,the positive predictive value was 0.533,and the negative predictive value was 0.897.4.The 15 radiomics features selected from the DCE-MRI subtraction group were significantly associated with the Luminal classification of breast invasive ductal carcinoma in the training set and validation set of both groups(P<0.01).The AUC value of the DCE-MRI substraction group prediction model on the training set was 0.867(95%:0.806?0.928),the diagnostic sensitivity was 0.714,the specificity was 0.863,the accuracy was 0.821,the positive predictive value was 0.690,the negative predictive value was 0.894,and the AUC value on the validation set was 0.815(95%:0.705-0.925),the diagnostic sensitivity was 0.714,the specificity was 0.863,the accuracy was 0.810,the positive predictive value was 0.741,and the negative predictive value was 0.846.Conclusions:1.Imaging features based on MRI dynamic enhancement and ADC images are useful for reflecting the biological features of breast cancer immunohistochemistry.2.MRI radiomics has a high classification and prediction ability for luminal classification of breast invasive ductal carcinoma. |
PDF Full Text Request