DCE-MRI-based Breast Tumor Heterogeneity Region Segmentation Method And Its Application In Ki-67 Expression Prediction

Breast cancer is a highly malignant tumor,according to the American Cancer Society(ACS),the number of new patients with breast cancer is on the rise and has been the highest incidence of malignant tumors in women.The study of gene expression profiles showed that breast cancer had different molecular subtypes,and there were significant differences in prognostic response and survival outcomes with different molecular subtypes.In addition to the differences between individuals,breast cancer is also highly heterogeneous within the tumor.At present,the biopsy sample is taken by the hospital to obtain the pathological information of the patient.The results of different areas of the tumor sampling may be different,so the diagnosis and treatment of the disease may be misjudged.Biopsy puncture on the human body has invasive damage,and can not describe the heterogeneity of the tumor through multiple biopsies.Dynamic enhanced magnetic resonance imaging(DCE-MRI)technology is widely used because of its soft tissue high-resolution and kinetic information.In this paper,the heterogeneity of internal tumor was studied by using the dynamic information of DCE-MRI,and the heterogeneity of the tumor was described.Ki-67 is one of the important indicators in the diagnosis,treatment and prognosis evaluation of breast cancer,therefore,Ki-67 expression prediction is studied in this paper through the results of breast tumor heterogeneity.At present,the relationship between the imaging features of DCE-MRI and the expression of Ki-67 is mainly based on the analysis of the entire tumor,and the relationship between heterogeneous regional characteristics and gene expression is poorly studied.In this dissertation,heterogeneous region segmentation and heterogeneity of tumor were described by the difference of dynamic parameters of breast DCE-MRI.Meanwhile,the heterogeneity region was compared with the overall tumor region to investigate whether the quantitative heterogeneity region study can achieve a more accurate prediction of Ki-67 expression in ER-positive patients.The specific research contents were as follows:(1)Data screening: According to the purpose of this study,we screened the data and statistically analyzed the basic information of the experimental samples,including age,menopause,size of the lesion and so on.Chi-square test and analysis of variance(ANOVA)were used to study the effect of basic information on the expression of Ki-67.(2)Image processing and feature extraction: The lesion area was extracted from the DCE-MRI by computer semi-automatic method,and the heterogeneous region was segmented by the dynamic information of DCE-MRI.Using computer semi-automatic method to extract the lesion region,through the kinetic information of the DCE-MRI to the tumor heterogeneity region segmentation,the segmentation was based on 1)Time to Peak(TTP);2)Peak Enhancement Rate(PER);3)Kinetic Pattern Clustering(KPC).The texture features,morphological features,statistical features and enhancement features of the segmented heterogeneous region and the whole tumor were extracted.In the extraction of texture features,two-dimensional texture feature and three-dimensional texture feature were extracted for comparative analysis.(3)Ki-67 predictive expression based on heterogeneous regional features: Logistic regression classification models were designed from univariate and multivariate perspectives,and the Receiver Operating Characteristic Curves(ROC curves)were plotted and calculated the area under ROC curve(AUC).Considering the interrelation between heterogeneous regions,the models of the classifier for heterogeneous regions were fused.Finally,Bootstrap was used to calculate the significant P-value to analyze whether the fusion model was significantly superior to the model constructed by the whole lesion.The precision,recall,F1-Measure and accuracy were calculated to evaluate the model comprehensively.Based on DCE-MRI,this article was studied on breast tumor heterogeneity segmentation method,heterogeneous regional characteristics and Ki-67 expression prediction model was constructed.Experiments showed that the texture features of heterogeneity regions significantly better than the entire tumor.After classifiers fusion,the classification models of heterogeneous regions were superior to the model of entire tumor in all indicators.Among them,the model of heterogeneity region by TTP-segmented showed that the AUC,after fusion was 0.922 ± 0.030,which was much larger than the model of entire tumor(AUC = 0.802 ± 0.053).The significance was tested by Bootstrap and the result was 0.04.Therefore,quantitative analysis of tumor heterogeneity can improve the accuracy of Ki-67 expression prediction,which may be a valuable clinical marker.