基于动态增强MRI的反向传播神经网络模型预测BI-RADS 4类乳腺良恶性病变

Predicting the Benignity and Malignancy of BI-RADS 4 Breast Lesions by Dynamic Contrast-Enhanced MRI Based Back propagation Neural Network Model

目的初步探索基于动态增强磁共振成像(DCE-MRI)的反向传播神经网络(BPNN)模型预测乳腺影像报告和数据系统(BI-RADS)4类乳腺良恶性病变的价值。方法纳入本院2018年2月至2022年12月经乳腺MRI诊断为BI-RADS 4类乳腺病变的患者260例,并随机分配至训练集(n=182)和验证集(n=78)。选取对比剂注射后的第二期DCE-MRI图像勾画病变感兴趣区体积,提取病变MRI影像组学特征。采用Logistic回归分析、最小绝对收缩与选择算子(LASSO)等方法,筛选出对预测乳腺BI-RADS 4类良恶性病变有意义的临床因素和影像组学特征,并分别建立临床模型、影像组学模型以及临床联合影像组学的结合模型、BPNN模型。通过受试者工作特征曲线、曲线下面积(AUC)、校准曲线和德隆(Delong)检验评比预测模型的性能。结果病灶长径是BI-RADS 4类乳腺良恶性病变的独立临床相关因素[优势比(OR):1.906;95%可信区间(CI):1.359~2.731];12个影像组学特征与BI-RADS 4类乳腺良恶性病变相关。对比临床模型、影像组学模型、结合模型,BPNN模型具有更高的预测性能(训练集AUC:0.784 vs.0.801 vs.0.855 vs.0.976,验证集AUC:0.725 vs.0.776 vs.0.813 vs.0.971),与其他三个模型之间的AUC比较差异均有显著统计学意义(P<0.05)。校准曲线表明,模型预测BI-RADS 4类乳腺病变的稳定性好。结论基于DCE-MRI的BPNN模型在预测BI-RADS 4类乳腺良恶性病变方面有较高的性能,可以为临床治疗决策提供参考依据。

Objective This study aimed to explore the value of Dynamic Contrast-Enhanced Magnetic Resonance Imaging(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 injection were selected to delineate the volume of the lesions'regions of interest,and radiomics features of the lesions were extracted.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).Additionally,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 indicated 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-making.