The Application Of Radiomics Analysis On Magnetic Resonance Diffusion Weighted Image In Breast Cancer

Chapter one Texture analysis for differentiating breast invasive ductal carcinoma and breast fibroadenoma on magnetic resonance diffusion weighted image:a preliminary studyObjective To determine the feasibility of using texture analysis to differentiate breast invasive ductal carcinoma and fibroadenoma on diffusion weighted imaging(DWI).Methods This study included 38 patients who underwent breast MRI prior to operation and had pathological findings,with a total of 40 breast tumors including 20 cases of breast invasive ductal carcinoma and 20 cases of breast fibroadenoma.The textural features of DWI images were analyzed.The receiver operating characteristic(ROC)curve was calculated to evaluate the diagnostic efficiency of texture parameters.Fisher discriminant model was used to determine the efficiency of texture parameters for distinguishing the two types of breast tumors.Results A total of 19 texture feature parameters,such as variance,standard deviation,intensity,and entropy,out of 76 texture parameters were statistically significant in the two sets of data(P<0.05).By comparing ROC curves,we found that the mean and relative deviations exhibited high diagnostic values in differentiating between invasive ductal carcinoma and fibroadenoma.The accuracy of Fisher discriminant analysis for the two types of breast tumors was 92.5%,and the accuracy rate of cross validation was 85.0%.Conclusion The texture characteristics on DWI images of invasive ductal carcinoma and fibroadenoma are significantly different.Texture analysis based on DWI images is feasible in identifying them,and the diagnostic efficiency is higher than that in conventional DWI.It can be used as an auxiliary method to identify them.Chapter two Radiomics analysis on magnetic resonance diffusion weighted image for prediction of triple-negative breast cancer:a feasibility studyPurpose This work aimed to explore whether radiomics features on magnetic resonance diffusion weighted image can be used to identify triple-negative breast cancer(TNBC)and other subtypes(non-TNBC),Methods This study were collected 80 patients in the training dataset from March 2017 to October 2017,and 67 patients in the independent training dataset from January 2018 to April 2018.In the training group.In training dataset,there are 20 cases of Luminal A,Luminal B,HER2+and triple-negative respectively.Radiomics features of DWI images were compared between TNBC and non-TNBC patients.Logistic regression and Fisher discriminant model were constructed to distinguish TNBC from non-TNBC patients.67 patients were used as the independent validation group to test the diagnostic efficiency of the model.Results A total of 76 imaging features were extracted from DWI images,and 12 radiomics feature parameters were statistically significant between TNBC and non-TNBC patients(P<0.05).In the Logistic regression model,the area under the ROC curve of the training group was 0.817,and AUC of 0.734 in validation dataset.The accuracy of Fisher discriminant model was 95%for distinguishing TNBC from non-TNBC in training group,83.8%for cross-validation and 83.3%for validation group.Conclusion The radiomics features of DWI images are significantly different between TNBC and non-TNBC,and it is feasible to distinguish TNBC from non-TNBC.Chapter three Radiomics analysis on magnetic resonance diffusion weighted image for prediction of triple-negative breast cancer:a feasibility studyPurpose To explore the feasibility of preoperative prediction of Ki-67 expression in breast cancer using diffusion-weighted magnetic resonance imaging.Methods The DWI images of breast cancer patients who underwent preoperative MRI examination were selected and the expression of Ki-67 in pathological tissues was measured after examination.A total of 114 lesions were obtained,of which 38 were negative for Ki-67 and 76 were positive for Ki-67.They were randomly divided into training set and verification set.The image histology characteristics of DWI images were analyzed,and the image histology characteristics were screened by LASSO regression model.The image histology label was constructed.The prediction model for evaluating the expression level of Ki-67 in breast cancer was constructed by using image histology label combined with pathological detection.The area under the ROC curve was used to represent the performance of the training set and training set.Results Statistical results showed that 25 of the 76 image histology parameters,such as entropy,energy and contrast,had statistical significance in the two groups of data(P<0.05).LASSO method was used to select 4 label features from 67 image histology features,and combined with pathological results for two-class modeling.Among them,AUC in training concentration reached 0.83,and AUC in verification concentration reached 0.76.Conclusion Ki-67-negative and Ki-67-positive breast cancer have different imaging histological characteristics in DWI images.The imaging histology of DWI is feasible in identifying the two,which is helpful to predict the expression level of Ki-67 in breast cancer before operation.