影像组学和BI-RADS分类鉴别乳腺结节的效能比较

Comparison between Radiomics and BI-RADS in Differential Diagnosis of Benign and Malignant Breast Nodules

目的:对比影像组学和BI-RADS分类鉴别乳腺结节良恶性的效能;方法:收集2018年9月到2019年9月期间超声发现的、有明确病理结果的乳腺结节。利用影像组学技术建立影像组学标签,计算其敏感性、特异性和准确性,并与BI-RADS分类方法比较;结果:总共纳入112个结节(恶性n=37,良性n=75)。恶性结节的影像组学标签值高于良性结节,具有统计学差异(P<0.001)。影像组学标签鉴别乳腺结节良恶性的最佳截断值为-0.52,对应的敏感性、特异性、准确性和ROC曲线下面积(AUC)分别为81.08%、84.00%、83.04%和0.91;BI-RADS分类分别为72.97%、92.00%、85.71%和0.84。两者的鉴别效能无统计学差异(AUC比较,P=0.08);结论:影像组学具备良好的乳腺结节良恶性鉴别效能。

Objective:To compare the diagnostic efficacy of radiomics and BI-RADS in distinguishing malignant and benign breast nodules.Methods:From September 2018 to September 2019,breast nodules with definite pathological results were collected.Radiomics features of the nodules on the ultrasound images were extracted using radiomics technique,and the meaningful features were screened.A formula was established to calculate the radiomics signature.The sensitivity,specificity and accuracy of the radiomics signature were calculated and compared with the breast ultrasound imaging report and data system(BI-RADS).Results:A total of 112 nodules(malignant,n=37;benign,n=75)were included.The values of radiomics signature of malignant nodules were significantly higher than that of benign nodules(P<0.001).The optimal cut-off value of radiomics signature for differentiating malignant from benign breast nodules was-0.52,with sensitivity,specificity,accuracy and area under ROC curve(AUC)were 81.08%,84.00%,83.04%and 0.91,respectively,which were 72.97%,92.00%,85.71%and 0.84 for BI-RADS,respectively.There was no significant difference in discrimination between radiomics signature and BI-RADS(AUC comparison,P=0.08).Conclusion:Radiomics has a good ability to distinguish malignant from benign breast nodules.