基于MR早期动态增强的影像组学标签鉴别乳腺良恶性病变的价值  

DCE-MRI-based radiomics signature in distinguishing benign and malignant breast diseases

作　　者：徐凡 彭丽君 梁莹莹 程梓轩 梁治平 曾旭文 XU Fan;PENG Li-jun;LIANG Ying-ying(Department of Radiology,Guangzhou Red Cross Hospital,Medical College,Jinan University,Guangzhou 510220,China)

机构地区：[1]广州市红十字会医院,暨南大学医学院附属广州红十字会医院放射科,广州510220 [2]广州市第一人民医院,华南理工大学附属第二医院放射科,广州510180

出　　处：《放射学实践》2021年第1期66-70,共5页Radiologic Practice

基　　金：广州市卫生健康科技项目(20191A011006)。

摘　　要：目的:探讨基于MR早期动态增强的影像组学标签鉴别乳腺良恶性病变的价值。方法:回顾性搜集通过乳腺动态对比增强MRI(DCE-MRI)检查,发现乳腺结节或肿块的144例患者(146个病变),146个病变按照样本量7:3随机抽样选取良性病变与恶性病变(102个作为训练组,44个作为验证组)。所有病例基于病变的三维图像对影像组学特征进行提取,然后采用Lasso logistic回归模型进行特征降维及筛选,以建立影像组学标签。采用ROC曲线对乳腺良恶性病变的诊断效能进行评价。结果:3个主要的影像组学特征与乳腺良恶性病变的鉴别诊断相关(P=0.0005)。建立的影像组学标签对鉴别乳腺良恶性病变具有较高的诊断效能,在训练组中的ROC曲线下面积(AUC)为0.909(95%CI:0.843~0.975),在验证组中的AUC为0.877(95%CI:0.743~1.000)。结论:基于MR早期动态增强构建的影像组学标签对乳腺良恶性病变具有较高的鉴别诊断效能,可辅助临床进行更精准的良恶性分层,为临床治疗方案的制订提供参考。Objective:To explore the value of early dynamic contrast-enhanced imaging radiomics signature in the diagnosis of benign and malignant breast lesions.Methods:The data of 144 patients with breast nodules or masses(n=146)detected through breast DCE-MRI were retrospectively collected.The benign and malignant breast lesions were randomly assigned into training group(n=102)and verification group(n=44)according to a sample size of 7:3.In all cases,the radiomics signature were extracted based on the three-dimensional images of the lesions,and then Lasso logistic regression model was used to reduce the dimensions and filter the features to establish the radiomics signature.The diagnostic efficacy of benign and malignant breast lesions was evaluated by ROC curve.Results:The three main imaging radiomics features were related to the differential diagnosis of benign and malignant breast lesions(P=0.0005).The established imaging radiomics signature has a good diagnostic ability for identifying benign and malignant breast lesions.The AUC in the training group is 0.909(95%CI:0.843 to 0.975)while the AUC in the verification group is 0.877(95%CI:0.743~1.000).Conclusion:Radiomics signature based on early dynamic magnetic resonance enhanced imaging can be used to identify benign and malignant breast lesions and enhance precise benign and malignant stratification in clinical practice,which can provide a reference for the formulation of clinical treatment plans.

关 键 词：乳腺病变 乳腺肿瘤 影像组学 磁共振成像 诊断 鉴别 

分 类 号：R655.8[医药卫生—外科学] R445.2[医药卫生—影像医学与核医学]