Application Of MRI Radiomics Model In Differentiating Benign And Malignant Breast Tumors And Predicting The Histological Grade Of Invasive Breast Cancer

Objective To study the clinical application value of imaging Radiomics model based on T2WI and DCE images in differentiating benign and malignant breast tumors and predicting the histological grade of invasive breast cancer.Methods A retrospective analysis of general hospital of Ningxia medical university between January 2017 and December 2019 postoperative pathology confirmed for 185 cases of breast benign and malignant tumor patients,including 131 cases of patients with breast cancer and 54 patients with benign breast tumor,invasive breast cancer were divided into low grade(Ⅰ+Ⅱgrade)78 cases,high grade(Ⅲgrade)38 cases.GE Signa 1.5T superconducting magnetic resonance scanner was used to obtain T2WI,DWI and DCE images.The conventional quantitative parameters(ADC value and IER value)were measured by radiologists with experience,and the ADC value and IER of breast were analyzed by Logistic regression to establish a model for predicting the tissue grading of invasive breast cancer and differentiating benign and malignant breast tumors.Use ITK-SNAP in T2WI and DCE,on the drawing sketch areas of interest in the AK software(Artificial Intelligence Kit,version 3.6.3,GE Healthcare)on Radiomics analysis on the image,based on T2WI,DCE,T2WI and DCE key features of sequence was extracted respectively using Logistic regression,support vector machine(SVM),a decision tree(Tree)and random forest(RF)four different algorithm model of machine,and as the gold standard by pathological examination results,draw the receiver-operating characteristic(ROC)curve of each model,calculate the area under the ROC curve(AUC),degree of sensitivity,specificity,accuracy,positive predictive value and negative predictive value and evaluate model of prediction efficiency.De Long test was used to compare the AUC values of different machine algorithm models,and decision curve analysis(DCA)was used to evaluate the clinical application value of different models.Results(1)The AUC values of ADC value and IER value were 0.980(95%CI0.962-0.995)and 0.797(95%CI0.723-0.865)for differentiating benign and malignant breast tumors,respectively.The AUC values of ADC value and IER value for predicting the histological grade of invasive breast cancer alone were 0.828(95%CI0.765-0.886)and 0.546(95%CI0.449-0.641),respectively.(2)Random forest(RF)algorithm was the best for the identification of benign and malignant breast tumors with T2WI and DCE sequences.The AUC value was 0.982(95%CI0.966-0.994),and the sensitivity,specificity,accuracy,positive predictive value and negative predictive value were 0.759,1.000,0.930,1.000 and 0.910,respectively.There was statistical difference between the AUC values of T2WI and DCE combined sequence using RF algorithm model and Logistic regression,SVM and Tree(P<0.001).In decision curve analysis,combination sequence of T2WI and DCE diagnostic efficiency of RF algorithm model is higher than that of Logistic regression,SVM and Tree algorithm model.(3)The Tree algorithm for the combination sequence of T2WI and DCE was the best in predicting the histological grading of invasive breast cancer.The AUC value was0.961(95%CI0.931-0.985),and the sensitivity,specificity,accuracy,positive predictive value and negative predictive value were 0.711,1.000,0.905,1.000 and 0.876,respectively.There was statistical difference between the AUC values of T2WI and DCE combined sequence using Tree algorithm model and Logistic regression,SVM and RF(P<0.001).In the analysis of decision curve,combination sequence of T2WI and DCE the grading efficiency of Tree algorithm model is higher than that of Logistic regression,SVM and RF algorithm model.Conclusion(1)In the routine quantitative parameters of breast magnetic resonance,ADC value has a higher clinical value in differentiating benign and malignant breast tumors,while IER value has a relatively lower clinical value.Random forest(RF)algorithm model based on DCE and T2WI joint sequence has high value in the identification of benign and malignant breast tumors.It can be used as an effective supplement to conventional quantitative parameters and provide reference for accurate diagnosis of benign and malignant breast tumors.(2)In the application of breast cancer grading,ADC value has certain clinical application value in differentiating invasive breast cancer grade;IER is of low clinical application value.The imaging cluster model based on breast MRI images can noninvasively predict the histological grade of invasive breast cancer in vivo,and the decision Tree machine algorithm model for combined T2WI and DCE sequences has the best prediction performance,which can provide valuable information for clinical decision.