Analysis Of Dynamic Enhancement Magnetic Resonance Imaging In The Stroma Surrounding The Tumor For The Prediction Of Molecular Subtypes In Breast Cancer

Breast cancer is the most common disease harmful to women health with high morbidity,which has been widespread concerned.Early diagnosing and treatment are the key to reduce breast cancer mortality of patients.However,most patients diagnosed with breast cancer have developed into the middle and late,leading to delayed treatment in the best period.With dynamic Contrast-enhanced Magnetic Resonance Imaging(DCE-MRI)technique,which has higher soft-tissue resolution in breast cancer early diagnosis than other imaging diagnostic techniques,we can obtain the dynamic information distribution of contrast agent in the capillary networks and tissue interspace.A classification system categorizes breast cancer into four distinct molecular subtypes at the genetic level.Distinct differences exist between patients with various molecular subtypes in prognosis and chemotherapeutic effect.The index of proliferation marker Ki-67,which reflects the proliferation activity of tumor cells,is one of the most important factors in breast cancer diagnosis and treatment.In previous studies,most researchers only focused on lesion area,while the study about tumor stroma region,providing the nutrition and metabolic demands for cancer cells,is relatively less in breast cancer.MRI and clinical findings reveal the stromal region is rich of more diagnostic information.In this study we investigated the role of 3D volumetric features derived from lesion areas and tumor stroma regions with dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI)in breast cancer early diagnosis.The article is mainly focuses on the enhancement patterns of dynamic information in tumor stroma regions and features extracted from tumor stroma regions in MRI to detect the expression of Ki-67 gene for prediction of breast cancer molecular subtypes.Detailed of this study as follows:(1)Acquisition of the region of interest and feature extraction.First,the semi-automatic computerized method was used to segment the lesion area in DCE-MRI images based on the richexperienced doctor’s annotation.Second,the lesion area was divided into different sub-regions according to dynamic enhancement rate and the Tpeak(time-to-peak)value of the pixels.Finally,imaging features including dynamic enhancement features,texture features and statistical features were extracted from the sub-regions.A total of eight stroma sub-regions were obtained with threedimensional stroma region segmentation method from the breast gland region.And the dynamic enhancement features were extracted from these eight stroma sub-regions respectively.(2)Analysis of enhancement patterns of dynamic information in tumor stroma region.In this paper,PE averages were used to characterize the dynamic enhancement signals for the stroma subregions.Enhancement patterns were explored with the different increasing radial distances and different directions in the same stroma sub-region.The results showed that the dynamic enhancement signals gradually decreases with the increment of radial distance.At the same time,there is no difference in directivity of the dynamic enhancement signal on the same stroma sub-region.(3)Prediction of Ki-67 gene expression status based on imaging features extracted from tumor stroma region in DCE-MRI.In order to predict the expression of Ki-67 gene with imaging features extracted from tumor stroma regions in DCE-MRI we designed logistic regression classification models in three aspects:(1)Logistic regression classifiers were built and compared in eight stroma sub-regions respectively.And then the stroma sub-region which has the best classification effect was selected;(2)A logical regression classifier was constructed for all stroma region based on multiple features fusion method and compared with the classifier built with the imaging features extracted from the lesion area;(3)A logical regression classifier was built with features extracted both from the lesion areas and the stroma regions.Finally,we used the area under the receiver operating characteristic(ROC)curve(AUC)and other evaluation index to assess the performance of models.The results showed that the best features for classification performance among them was extracted from the SB region in all stroma sub-regions.At the same time,the classifier contracted with features extracted from all stroma regions has preferable classification effect as the classifier built with features computed from lesion areas.After fused the features extracted both from lesion areas and stroma regions,the AUC value is significantly improved to 0.920.In this paper,we investigated the application of imaging features extracted from breast tumor stroma region with DCE-MRI in breast cancer early diagnosis.Results showed that the imaging features computed from the stroma region have rather high classification performance in prediction of molecular subtypes in breast cancer.This fully confirmed the potential clinical value of the stroma region.By fusing features extracted both from the lesion area and the stroma region in DCE-MRI may provide a new method for research early diagnosis in breast cancer.