Objective This study aimed to explore the value of Dynamic Contrast-Enhanced Magnetic Resonance Ima-ging(DCE-MRI)based back propagation Neural Network(BPNN)model in predicting Breast Imaging Reporting and Data System(BI-RADS)category 4 breast lesions.Methods A total of 260 patients diagnosed with BI-RADS category 4 breast lesions through breast MRI at our hospital from February 2018 to December 2022 were included and randomly divided into a training set(n=182)and a validation set(n=78).The second phase of DCE-MRI images after contrast agent in-jection were selected to delineate the volume of the lesions'regions of interest,and radiomics features of the lesions were ex-tracted.Logistic regression analysis,Least Absolute Shrinkage and Selection Operator(LASSO),and other methods were used to select clinically meaningful factors and radiomics features for predicting BI-RADS 4 breast lesions.Clinical model,radiomics model,combined model and BPNN model of clinical and radiomics features were established.The performance of the prediction models was evaluated using Receiver Operating Characteristic(ROC)curves,Area Under the Curve(AUC),calibration curves,and DeLong's test.Results The long diameter was identified as an independent clinical factor associated with BI-RADS 4 breast lesions(odds ratio(OR):1.906;95％confidence interval:1.359-2.731).Additional-ly,12 predictive radiomics features were found to be associated with BI-RADS 4 breast lesions.When compared with clinical model,radiomics model,and combined model,the BPNN model exhibited the highest predictive performance(training set AUC:0.784 vs.0.801 vs.0.855 vs.0.976,validation set AUC:0.725 vs.0.776 vs.0.813 vs.0.971).The differences in AUC between the BPNN model and the other three models were statistically significant(P＜0.05).Calibration curves indi-cated that the model had good stability in predicting BI-RADS 4 breast lesions.Conclusion The BPNN model exhibits higher performance in predicting BI-RADS 4 breast lesions,providing a reference basis for clinical treatment decision-mak-ing.
Breast lesionsBack propagation neural networkMagnetic resonance imagingPrediction model
维普
万方数据
