对乳腺肿块构建多模态影像BI-RADS预测模型的研究

The study of building a multi-modal imaging BI-RADS predictive model for breast lesions

目的:构建乳腺肿块的综合乳腺影像报告和数据系统(Breast Imaging Reporting and Data System,BI-RADS)预测模型,得到不同于单一影像的BI-RADS分类。方法:回顾并分析2019年8月—2020年9月术前行超声、乳腺X线摄影及磁共振成像(magnetic resonance imaging,MRI)检查的患者图像,根据BI-RADS词典对肿块特征进行评估,取3种影像的最高BI-RADS类别为因变量,影像学特征及临床指征为自变量,根据多元logistic回归构建综合BI-RADS预测模型。结果:综合BI-RADS预测模型分类的阳性预测值(positive predictive value,PPV)在指南的参考范围内[3类(0.00%),4A类(9.61%),4B类(42.41%),4C类(88.18%),5类(97.19%)],模型的受试者工作特征(receiver operating characteristic)曲线的曲线下面积(area under curve,AUC)为0.955。结论:将3种影像的BI-RADS词典联合肿块临床特征得到综合的BI-RADS模型是可行的,极大限度避免漏诊的发生。

Objective:To build a comprehensive Breast Imaging Reporting and Data System(BI-RADS)predictive model of breast lesions for getting BI-RADS classification that different from single imaging.Methods:The images of patients undergoing preoperative ultrasound,mammography and magnetic resonance imaging(MRI)examination from August 2019 to September 2020 were retrospectively analyzed.The tumor characteristics were evaluated according to the BI-RADS lexicon.The highest BI-RADS category of the three images was taken as the dependent variable,and the imaging characteristics and clinical indications were taken as independent variables.A comprehensive BI-RADS predictive model was constructed based on multivariate logistic regression.Results:The positive predictive value(PPV)of the comprehensive BI-RADS predictive model category was within the reference range of the guide[category 3(0.00%),category 4A(9.61%),category 4B(42.41%),category 4C(88.18%),category 5(97.19%)],the area under curve(AUC)of the receiver operating characteristic(ROC)curve was 0.955.Conclusion:It is feasible to unify the BI-RADS lexicon of the three imaging and the clinical features of the patients to obtain the comprehensive BI-RADS model,which greatly avoids the incidence of missdiagnosis.