The Clinical Application Of Quantitative Analysis Of DCE-MRI And Radiomics In Evaluating The Characteristics Of Invasive Breast Carcinoma

The first part The correlation study of quantitative DCE-MRI parameters from linear reference region model with prognostic factors of invasive breast cancerObjective:The dynamic contrast enhanced MRI?DCE-MRI? is suggested to be a conventional MRI examination of breast cancer in clinical practice.The quantitative analysis of DCE-MRI could accurately evaluate the blood supply of breast cancer by using the pharmachokinetic model.However,the most common pharmachokinetic model in the previous quantitative analysis is the two-compartment model,which calls for a high temporal resolution but out of spatial resolution and image signal-to-noise ratio.The linear reference region model is suitable for analysis of MR images with low temporal resolution,and without sacrifice of spatial resolution.The study used the standard clinical data that explored the feasibility of the linear reference region model in the quantitative analysis of breast cancer DCE-MRI,and analyzed the relationship between DCE-MRI quantitative parameters and the prognostic factors of invasive breast cancer.Methods:1.Clinical data We retrospectively collected the invasive breast cancer patients who confirmed by surgery and pathology,and with breast MRI examination 10 days before surgery.84patients with 87 lesions were included.2.MR imaging Siemens 3.0 T superconducting magnetic resonance equipment was used.Imaging sequences included transverse axis T1WI,transverse axis T2WI,sagittal bilateral breast T2WI,transverse axis diffusion weighted images?DWI?and dynamic enhancement?DCE? T1WI.The DCE-MRI included 8 sequential phases.3.Image analysis and post-processing DCE-MRI data were analyzed using the Omni Kinetics Manual software?GE Pharmaceuticals?.The amplitude of the arterial input function?AIF? was acquired after images registration of different phases.The region of interest?ROI? was drawn manually layer-by-layer around the tumor on the maximum enhanced phase,avoiding necrosis,cystic region and bleeding area.Then,the ROIs of all layers were combined three dimensionally.Blood perfusion parameters were computed using the linear reference region model.The quantitative parameters were measured twice with an interval of at least one month.The two measurements were recorded as K₁^trans and k_ep1,K₂^trans and k_ep2,respectively.The average of the two measurements were taken as the final results which were recorded as K^trans and k_ep.4.Evaluation of clinical prognostic factors of invasive breast cancer The prognostic factors included tumor size,lymph node metastasis,distant metastasis,histological grade and molecular type.5.Statistical analyses The classified variables were compared using the chi-square test or the R�C list test.The parameters of two measurements were compared using the Bland-Altman analysis and Spearman correlation analysis.The continuous variables of two groups were compared using Student t-test or Mann-Whitney U-test.The continuous variables of multiple groups were compared using one-way ANOVA with Bonferroni test or Kruskal-Walis H-test.The diagnostic efficacies of K^trans and k_ep on prognostic factors were analyzed by the receiver operating characteristic?ROC?curve.All statistical analyses were 2-sided and were performed using SPSS 22.0 software.Results:1.There was a good consistency of K^trans and k_ep of the two measurements.The correlation coefficient of K^trans was 0.795?P<0.001?.The correlation coefficient of k_epp was 0.904?P<0.001?.2.Correlation analysis between DCE-MRI quantitative parameters and prognostic factors of invasive breast cancer K^trans had a moderate positive correlation with histological grade,maximal diameter and triple negative type,weak positive correlation with lymph node metastasis,and no correlation with distant metastasis.K_ep had a low positive correlation with histological grade,tumor maximal diameter and triple negative type,weak correlation with distant metastasis,and no correlation with lymph node metastasis.3.Diagnostic efficacies of DCE-MRI quantitative parameters on the prognostic factors The diagnostic efficiencies of K^trans and k_ep were low in lymph node metastasis.The diagnostic efficiencies of K^trans and k_ep were 0.704 and 0.756?P=0.037,P=0.009?in distant metastasis,respectively.K^trans and k_ep had high diagnostic efficiencies in distinguishing between high and medium-low grade tumors,and the area under curves of ROC were 0.959 and 0.942?P<0.001,P<0.001?,respectively.The diagnostic efficiencies of K^trans and k_ep for distinguishing triple negative and non-triple negative breast cancer were 0.857 and 0.756?P<0.001,P<0.001?,respectively.Conclusion:1.The quantitative analysis of DCE-MRI in breast cancer provided a good reproducibility using the linear reference region model.2.The routine clinical MRI data of breast cancer was maximally used by quantitative analysis based on linear reference region model.This method could be used to estimate the risk of invasiveness and metastasis of invasive breast cancer.K^trans and k_ep may be warranted in evaluating the invasive breast cancer.The second part Application of radiomics in the evaluation of biological characteristics of invasive breast cancerObjective: Breast cancer is a highly heterogeneous tumor.The heterogeneity of different tumor biological characteristics may vary significantly,and influence treatment efficacy and prognosis.The breast DCE-MRI image can show the heterogeneity of the whole tumor,however,it is hard to quantification through observation of naked eye.Radiomics can transform the image into data information,and maximally use the spatial information of image by quantifying the spatial distribution of pixels.So far,however,there has not reported on the application of radiomics to evaluate the biological characteristics of breast cancer which have a important impact on comprehensive treatment.The study aimed to investigate whether the radiomics of breast MR images can differentiate the biological characteristics of the tumor,and to compare the discriminating abilities of different sequences.Methods: 1.Clinical data The DCE-MRI data of 112 breast cancers were collected retrospectively from April 2013 to September 2016.The inclusion and exclusion criteria were the same as in the first part.2.MR imaging The same with the first part.3.Segmentation and feature extraction The segmentation was carried out on T1 WI,the peak enhanced phase of T1WI?about 80 s ? 120s?,T2 WI and DWI.Ma Zda software was used for analysis.The ROI was delineated manually at the largest slice of tumor at the peak enhanced phase of T1 WI,including necrosis.Then the ROI was copied to other sequences,with properly edge adjusting.The texture features were calculated from six algorithms include histogram,gray level co-occurrence matrix,run-length matrix,absolute gradient,autoregressive model and wavelet transform.4.Dimensionality reduction The extracted texture features were standardized first.Then the dimensional reduction was obtained using principal component analysis?PCA?on MATLAB.The cumulative contribution rate from the first principal component was calculated.The set of n principal components was considered to be sufficient to reflect the information of the original data when the contribution rate of the current n principal components reaches 85% or more.5.Support Vector Machine?SVM? classifier to identify categories The principle components of 112 lesions were used as input of SVM.74 lesions were used as training group and 38 lesions as independent testing group.The "k-fold cross-validation" method was used to select different variables?including axillary lymph node metastases,distant metastases,histologic grade,and triple-negative subtyping?.Linear kernel function was used to construct SVM classifier model.Then verified the model in the testing group.6.ROC analysis The diagnostic accuracies of the SVM classifier model were evaluated with ROC method using the distances of every lesion to the decision hyperplane of SVM.The overall diagnostic efficiency of the different classifier models were compared by the AUC of each curve.Results: 1.Dimensionality reduction The principal component analysis reduced the original data of each sequence into 11-13 principal components,which can represent more than 85% of the original data.2.Discrimitive performance of SVM classifier For axillary lymph node metastasis,the classification accuracy of SVM classifiers constructed from the four sequences were all higher than 81%,and higher than the SVM classifier constructed from all the features of the four sequences.For distant metastasis,the classification accuracy of SVM classifier from DWI was the highest and the SVM classifier from all features of the four sequences was the lowest.For histological grade,the classification accuracy of SVM classifiers from the four sequeces were all higher than 84%,and higher than the SVM classifier constructed from all the features of the four sequences.For triple-negative subtyping,the classification accuracy of SVM classifiers from the four sequeces separately and combinedly were all higher than 94%.3.Comparison of diagnostic effects of SVM classifiers from different sequences The AUC of T1 WI,the peak enhanced T1 WI and T2 WI were all higher than 0.9 for axillary lymph node metastasis.The sensitivity of T1 WI was the highest.The diagnostic efficacies of the four sequences for distant metastasis were all low with a AUC of 0.5-0.7.The AUC of the combination was 0.81,but the sensitivity was low.The diagnostic efficacies of the T1 WI,the peak enhanced T1 WI,DWI and the combination for histologica grade were all low with a AUC of 0.6-0.7.The AUC of T2 WI was 0.84,but the sensitivity was not available.The AUC of T1 WI and DWI were low while the sensitivity of the peak enhanced T1 WI,T2WI and the combination were low.Conclusion: The texture analysis of clinical common breast MRI images can discriminate the axillary lymph node metastasis of invasive breast cancer,and can provide additional information for clinical treatment.