| BackgroundAt present,the incidence of breast cancer is the highest in female malignant tumors.In the clinical diagnosis and treatment of breast cancer,the accurate diagnosis of benign and malignant lesions is the most important step in the treatment of breast diseases,which is crucial to patient prognosis and reducing mortality.According to the BI-RADS classification description of the MRI of breast lesions reported by the American Society of Radiology(ACR),the lesions with atypical malignant signs but have sufficient reason to be classified as category BI-RADS 4(> 2% but <95% probability of malignancy).Due to the large range of likelihood of malignant lesions,all BI-RADS 4 breast diseases recommend biopsy of suspicious areas to clarify their pathological characteristics.ObjectiveDue to the possibility of malignant lesions in BI-RADS 4 breast diseases,we can see that many cases of BI-RADS 4 patients are overdiagnosed and puncture,which causes a degree of traumatic [4] to the body.And the clinical diagnosis inevitably has a certain false-positive rate.Therefore,we proposed the hypothesis of establishing a predictive imaging omics model based on preoperative imaging information to predict the malignant risk probability of BI-RADS 4 breast masses.Therefore,a radiomics(Radiomics)technology based on establishing the correlation between image features and clinical information based on extracting rich and deep image features from medical images has emerged,which can improve the accuracy of tumor diagnosis,prognosis and prediction.In addition,radiomics is also an important part of precision medicine and individualized therapy,especially in oncology.Currently,dynamic contrast-enhanced MRI(DCE-MRI),DWI,and ADC sequences have been applied to distinguish between benign and malignant breast tumors,determine tumor size,and detect occult lesion.However,the DWI and ADC are the images after removing the direction weight,which represent the average value of the different directions,especially the three-dimensional orthogonal directions.Due to the human breast acini,cavity,ducts and fibers are radial to the nipple,mammary gland lesions are also easily along the physiological channels expansion,spread,so the change of diffusion breast lesions in three-dimensional direction,but presents the characteristics of anisotropy,but DWI can not reflect this situation.To distinguish between the diffusion differences in the 3 D direction,DTI was introduced and introduced into the study of breast lesions,and showed a very high stability and reproducibility.DTI can not only provide the mean ADC value(also known as mean dispersion rate,meandiffusion,MD),but also provide the anisotropy fraction of diffusion difference(FA),maximum intrinsic(λ1),intermediate intrinsic(λ2),minimum eigenquantity(λ3)and maximum anisotropy index(λ 1-λ 3)Therefore,this study aims to establish a nomogram based on DCE-MRI sequence combined with DTI parameters to predict the malignant probability of BI-RADS 4 breast masses before surgery.Materials and MethodsWith the approval of the hospital ethics committee,the imaging and clinical data of 216 female patients presenting with breast disease and diagnosed with a BI-RADS 4 mass from December 2017 to December 2021.All patients received a conventional(DCE-MRI)sequence at the 1.5TMRI and added a DTI sequence.All MRI images were reviewed separately by two highly seniority radiologists.Clinical data included age,menopausal status,breast background parenchymal enhancement(BPE)category,tumor size,and fertility.DTI parameters include ADC value(also known as average dispersion rate,meandiffusion,MD),anisotropy fraction(FA),maximum eigenvalue(λ 1),intermediate eigenvalue(λ 2),minimum eigenvalue(λ 3)and maximum anisotropy index(λ 1-λ 3).A hundred and seventeen radiomics features were extracted from the dynamic contrast enhancement(DCE-MRI)images,using the minimum absolute shrinkage and selection operator(LASSO)and logistic regression.Using postoperative pathology as the gold standard,all lesions were randomly divided into training and validation groups according to the ratio of(7:3),By Logistic regression analysis,using the benign and malignant pathological outcome as the dependent variable,Univariate analysis was performed first in the training group with all recorded risk factors as independent variables,After a multivariate analysis was performed,Logistic regression prediction models were then built with statistically significant risk factors,Finally,the clinical model,DTI parameter model,and Radscore model were established respectively,The Radscore model + DTI parametric model,Radscore model + clinical model,DTI parametric model+ clinical model,Radscore model + DTI parametric model + clinical model,Construction of 7predictive models,Finally,a nomogram to predict the benign and malignant BI-RADS 4 breast masses was established.Draw the subject working curve(receiver operating characteristic curve,ROC)and find the area AUC under the curve;draw the calibration curve(Calibration curve)to test the model prediction accuracy;and draw the decision curve(Decision Curve Analysis,DCA)to evaluate the clinical fitness of the model.ReslutsFifty-two unrelated features were removed using the ICC consistency analysis and Pearson correlation,and then the remaining feature coefficients were screened for each model we constructed using LASSO and logistic regression.Finally,the best performing nine related features were obtained in the model.The radiomics model was then constructed based on these nine features.Patient clinical data and univariate analysis results of DTI parameters showed that patient age,menopause,breast background parenchymal enhancement(BPE)category,tumor size,ADC,FA,λ 1,λ 2,and λ3 in the DTI parameters of breast cancer tissue were the risk factors for the diagnosis of benign and malignant breast mass(P <0.05).The combined model based on Radscore model + DTI parameter model + clinical model achieved the highest AUC scores compared with Radscore model only,DTI parameter model,clinical model and pairwise combined model,with an AUC of 0.877 [95% CI,0.834-0.892] and 0.859 [95% CI,0.767-0.932] in the training and validation cohorts,respectively.Model calibration diagram results obtained by Hosmer-Lemeshow test showed that the training group and validation group were well consistent.Training group Slope 1,validation group Slope=0.917,and the area under the(DCA)curve was 0.81,indicating that the model had good clinical fitness.ConclusionsRadiomic features extracted from the primary tumor region of DCE-MRI images can be used to distinguish between benign and malignant breast lesions,while combining DTI parameters and clinical models can significantly improve the diagnostic power of benign and malignant lesions of breast MRI.This can provide very valuable evidence to support clinical manipulations and treatment decisions.The diagnosis of good and malignant breast tumors has good sensitivity and specificity and has good clinical fitness,and is expected to provide a brand new and objective method for the clinical diagnosis of benign and malignant breast tumors. |
PDF Full Text Request