Multiparameter MRI Features Of Breast Tumors And Its Application In Molecular Typing And Histological Grading Prediction

Breast cancer has become the world-wide number one threat to women’s physical and mental health.To achieve accurate diagnosis and treatment of breast cancer as soon as possible,it is urgent to study how to improve detection and treatment of breast cancer using molecular characteristics and pathological information of individuals and their tumors.In the clinical treatment of breast cancer,molecular subtypes and histological grading of breast cancer are basis of treatment selection and prognostic estimation.Dynamic Contrast-Enhanced Magnetic Resonance Imaging(DCE-MRI)and Diffusion-Weighted Imaging(DWI)are widely used in the diagnosis of breast cancer as non-invasive and informative imaging methods.At present,the correlation between biological indicators and histological grading of breast cancer and medical imaging is mainly based on the tumors on DCE-MRI.This paper explores the predictive performance of both tumors and glands on DCE-MRI and DWI for molecular typing and histological grading of breast cancer.Specific research contents are as follows:(1)Obtaining regions of interest(ROI)and feature extraction:The whole tumor region,tumor boundary,proximal gland region,middle and distal gland region on DCE-MRI are segmented using computer semi-automatic method,and the corresponding five regions of interest on DWI are extracted applying the maximum mutual information criterion between DWI and ROIs on DCE-MRI.A series of features including texture features,statistical features and dynamic enhancement features are extracted based on all regions of interest.(2)Predictive performance analysis of single and multiple features for molecular typing and histological grading:Firstly,a single variable logistic regression model is constructed based on individual image feature from different ROI under DCE-MRI and DWI,and the model is trained and tested under Leave-One-Out Cross Validation(LOOCV).Receiver Operating Characteristic Curve(ROC)and the area under ROC curve(AUC)were used to evaluate the model’s predictive performance.Then,a multivariable logistic regression model is constructed based on the feature sets of different sub-regions under the two parametric images.And the prediction performance of different sub-regions and the corresponding sub-regions in DCE-MRI and DWI before and after feature fusion are analyzed and compared.The experimental results show that tumors on DCE-MRI or DWI have a good classification effect.For Luminal A,the model produced AUCs of 0.814 and 0.801,respectively.And the AUC of the model based on tumors after fusion is increased to 0.929,which indicates that there exists information complementarity between DCE-MRI and DWI.At the same time,the fused gland sub-regions have equivalent or even better classification effect compared with the fused whole tumor,which fully affirms the potential value of gland in radiomic analysis of breast cancer.To sum up,this paper studies the peritumoral fibrous gland tissues and tumors based on DCE-MRI and DWI,which provides a new research idea for the prediction of biomarkers of breast tumors.