The Value Of Based On Dual-model Ultrasound Radiomics In The Differential Diagnosis Of Benign And Malignant Breast Masses

Purpose:Early diagnosis of benign and malignant breast masses is vital for subsequent therapy,and radiomics can be utilized to characterize tumors by quantitatively analyzing ultrasound images.The objective of this study is to construct a Nomogram model for preoperative visualization to predict the benignity and malignancy of breast masses by analyzing the semantic features and radiomics features of two-dimensional ultrasound images and strain elastography images.Materials and Methods:A total of 219 breast masses were included in this study in 219 patients who presented to our hospital for ultrasound examination with breast masses from March2021 to March 2023,and all breast masses were pathologically confirmed by puncture biopsy or surgical excision.Two ultrasonographers retrospectively analyzed two-dimensional ultrasound images and strain elastography images of each breast mass.Using a stratified sampling method,all breast masses were randomly divided into a training set(n = 153)and a testing set(n = 66)in a ratio of 7:3.The two groups were compared for statistical significance in terms of patients’ age,mass maximum diameter,mass location,shape,margin,orientation,posterior features,calcification,fat strain ratio(FLR),elastography score,and blood flow classification.Independent predictors of breast malignancy were screened by univariate and multivariate logistic regression analysis of semantic features,and a semantic feature model was developed.A two-dimensional radiomics model,a elastography radiomics model and a bimodal radiomics model were developed using analysis of variance(ANOVA),the maximum correlation minimum redundancy algorithm(m RMR),the least absolute shrinkage and selection operator(LASSO)feature selection method,and a logistic regression(LR)classifier were selected,respectively.The semantic features and the Radscore of bimodal radiomics model were combined by univariate and multivariate logistic regression analysis to construct Nomogram.the area under the curve(AUC),accuracy(ACC),sensitivity(SEN),specificity(SPE)and F1 scores of the subject characteristic curve(ROC)were calculated for each of the five prediction models to evaluate the diagnostic accuracy of the models.The differences between the ROC curves of the different models were compared using the De Long test.The agreement between the predicted and actual results of the five models was assessed using calibration curves and the Hosmer-Lemeshow goodness-of-fit test was applied to assess the degree of fit of the five prediction models.To further assessed the value of the five prediction models for clinical application,decision curve analysis(DCA)was used to quantify the standardized net benefit at different threshold probabilities.Results:There were no statistically significant differences between groups in the training and testing sets in terms of patients’ age,mass maximum diameter,mass location,shape,margin,orientation,posterior features,calcification,FLR,elastography score and blood flow classification(p > 0.05).Age,maximum diameter,margins,FLR and blood flow classification were independent predictors of breast malignancy(p < 0.05).The AUCs of the training set were 0.83,0.91,0.92,0.96,and 0.99 for the semantic feature model,elastography radiomics model,two-dimensional radiomics model,bimodal radiomics model,and Nomogram,respectively,and the AUCs of the testing set were 0.86,0.87,0.91,0.93,and 0.95,respectively.Incorporating Radscore and semantic features Nomogram had the best diagnostic efficacy,and the bimodal radiomics model outperformed the semantic feature model,the two-dimensional radiomics model,and the elastography radiomics model.In the training set,the De Long test showed statistically significant differences(p < 0.05)between the ROC curves of the semantic features model and the elastography radiomics model,the bimodal radiomics model,and Nomogram,respectively.The calibration curves for the five models were tested using the Hosmer-Lemeshow test,with the testing set for the elastography radiomics model yielding significant(p < 0.05)results and the remaining models yielding non-significant(p > 0.05)results,providing evidence of good calibration.DCA showed that Nomogram had higher net clinical benefit.Conclusion:The two-dimensional ultrasound radiomics model and the elastography radiomics model are of diagnostic value in identifying benign and malignant breast masses.The bimodal radiomics model based on two-dimensional images and strain elastography images is superior to two unimodal radiomics models.Nomogram based on the bimodal radiomics model and semantic features can further improve its diagnostic efficacy in benign and malignant breast masses and provide valuable information for clinical decision making.