磁共振纹理动态特征鉴别乳腺良恶性肿块  

Differentiated diagnosis between malignant and benign breast mass using textural kinetic MRI features

作　　者：林帆[1] 胡若凡[1] 梁超[1] 余娟[1] 刘侠静 雷益[1] 

机构地区：[1]深圳市第二人民医院放射科,广东518035

出　　处：《放射学实践》2017年第10期1037-1040,共4页Radiologic Practice

基　　金：深圳市科技研究资金(JCYJ20150330102720117)

摘　　要：目的:提取乳腺病灶的时空变化特征作为新的DCE-MRI标记(称为纹理动态特征)并证明其鉴别良恶性肿块的能力。方法:回顾性分析52个乳腺肿块,其中恶性肿瘤30个,良性肿块22个,提取并对动态特征信号强度特征、纹理特征、形态特征、边缘特征进行分组。为了更好评估这些特征,采用不同的特征类建立分组模型,计算正确率,敏感度,特异性及曲线下面积(AUC)。结果:结合纹理动态特征所建立的良恶性肿瘤分类器具有最大的AUC=0.94,准确率90%,敏感度92%,特异性85%,优于其他各组分类器,与信号强度特征所建立的模型差异有统计学意义(AUC=0.80,P<0.05)。结论:磁共振纹理动态特征有助于鉴别良恶性肿块,甚至优于临床上最流行的DCE-MRI标记信号强度动态特征。Objective:In this study,we proposed a new DCE-MRI descriptor(called textural kinetics)to capture the spatiotemporal changes of lesion texture for the better discrimination of benign and malignant lesions.Methods:We retrospectively collected 52 breast mass lesions(30 malignant and 22 benign).Kinetic features were extracted from breast DCEMRI and divided into four categories:signal intensity,texture,morphology,margin feature.In order to evaluate these features,we created classification model using different feature classes.Accuracy,sensitivity and specifity and area under curve(AUC)were calculated.Results:When textural kinetic features were combined with other features,the classifier yielded90% accuracy,92% sensitivity and 85% specificity,and 0.94 AUC,showing the best differentiation between benign and malignant lesions.The difference for AUC between combined classifier and signal intensity classifier is significant(AUC=0.80,P<0.05).Conclusions:Textural kinetic features in DCE-MRI is helpful in distinguishing malignant from benign lesions,which is even better than the most popular DCE-MRI descriptor of signal intensity kinetics feature in clinic.

关 键 词：磁共振成像 纹理 乳腺肿瘤 人工智能 

分 类 号：R445.2[医药卫生—影像医学与核医学] R737.9[医药卫生—诊断学]