Association Between DCE-MRI Features And Molecular Subtypes In Breast Cancer

Breast cancer is the top murder of the women's health in all of the malignancy.It can be divided into four molecular subtypes of Luminal A,Luminal B,HER-2 and Basal like at the genetic level.There are significant differences among the breast cancer patients with different molecular subtype in the expression of disease,response to treatment and survival outcomes.If the judgment of molecular type is made before pathological examination,it will be very advantageous to the formulation of post-treatment plan,moreover,it can avoid patients being exposed to intrusive damage.The solving means to this problem is proposed as the formation and development of imaging genomics.Different molecular subtypes of breast cancer always show difference in imaging characteristics,so it is acceptable that there is a correlation between molecular subtype and imaging characteristics,and that image examination can even become a way to identify molecular subtypes.At the same time,Dynamic Contrast-Enhanced Magnetic Resonance Imaging(DCE-MRI)stands out in all types of breast imaging diagnostic tools by virtue of high resolution of the organization and dynamic characteristics has been widely used in recent years.The research on associations between breast DCE-MRI imaging features and the molecular subtype is currently rare in our country,but it is a hot trend in foreign countries.It has been reported that many image features showed significant correlation with molecular subtypes.Compared to previous study,more image features were extracted to study the associations between them,and these features were used to build the multi-class classifier in order to classify and predict molecular subtypes.This research mainly focused on the two issues: finding out the imaging features that showed significant correlation with molecular subtypes in DCE-MRI and prediction ability of these features to classify molecular subtypes.This study specifically includes:(1)Molecular subtype judgment: 60 cases of malignant breast cancer were retrospectively analyzed whose immunohistochemistry pathology and DCE-MRI was checked before chemotherapy and surgery.According to immunohistochemistry ER,PR,HER2,Ki-67 expression results,patients were divided into four molecular subtypes.The distribution of four molecular subtypes was studied and statistical analysis was applied.Analysis of variance and chi-square test were carried out for age,menopausal status,breast tumor type and other information to study their impact on molecular subtype.(2)Image feature extraction: semi-automatic method using computer was performed for breast segmentation and lesion segmentation of DCE-MRI to acquire three areas for study such as the lesion region,normal breast side background region and abnormal breast side background region.And 65-dimensional features,including statistical features,morphology,texture features,dynamic contrastenhanced imaging features,were extracted in the respective regions.Especially in dynamic enhanced features,also extracted association between feature region,such as the ratio of dynamic enhancement rate between the lesion region and the abnormal side background region and the ratio of dynamic enhancement rate between the normal side and the abnormal side background region.(3)Association analysis of image feature and molecular subtype: the association between DCEMRI imaging features and the molecular subtypes of breast cancer was investigated.Significant P value calculated from univariate logistic regression and multivariate logistic regression model was used to assess the correlation of image features and each molecular subtype and imaging features,such as skewness,firmness,was significantly associated with molecular subtypes.Furthermore,for a single molecular subtype,classification prediction model of multivariate logistic regression was conducted,and then the model was evaluated using the area under the receiver operating characteristic curve(area under roc curve,AUC),cross-validation accuracy and other indicators.Finally,the distribution of the significant image features was studied,and it was explained based on the biological theories.(4)Study on molecular subtype prediction using image features: multi-class classifier such as logistic regression,support vector machines and decision tree was designed using DCE-MRI imaging features to predict breast cancer molecular subtypes.For five kinds of situations of three class classifier,assessment and contrastive analysis of classification results were showed.It was found that overall classification performance showed the result,support vector machines> decision tree> logistic regression,and one-versus-one mode is superior to one-versus-rest mode when two class classifier was used to conduct three class classifier.In this thesis,an objective quantitative research on the association between the molecular subtype and DCE-MRI imaging features and classification and prediction capability for molecular subtype between based on DCE-MRI imaging features was put forward.The results showed a certain correlation between the image features and the molecular subtype and classification and prediction capability of molecular subtype was good using DCE-MRI imaging features.This will further enhance the value of imaging diagnosis in breast cancer therapy.