【摘要】 目的 探讨动态对比增强磁共振成像(DCE-MRI)影像组学特征对乳腺良恶性病变的诊断价值。方法 回顾性分析经病理证实的244例乳腺病变患者,按照7:3的比例将患者随机分为训练组(n=170)和验证组(n=74)。从DCE-MRI图像的第2期中提取影像组学特征。采用最小绝对收缩和选择算子(LASSO)回归方法筛选出训练组的最佳特征并构建影像组学标签,采用受试者工作特征(ROC)曲线评价影像组学标签的诊断效能。结果 共筛选出8个特征用于建立影像组学标签,训练组和验证组影像组学标签的曲线下面积(AUC)分别为0.891和0.878,分别以0.632和0.699为最佳诊断阈值,其敏感性和特异性分别为89.00%和74.29%;86.05%和83.87%。其中浸润性导管癌、黏液癌、髓样癌、纤维腺瘤和小管腺瘤的分类准确度较高,准确度分别为91.82%、100%、100%、83.72%和100%。结论 利用DCE-MRI提取的影像组学特征有助于乳腺良恶性病变的鉴别诊断,联合LASSO回归方法筛选出的影像组学特征建立的影像组学标签具有较高的诊断预测性。更多还原

【Abstract】 Objective To explore the value of dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI) radiomics in differentiating benign and malignant breast lesions.Methods 244 patients with breast lesions confirmed by pathology were analyzed retrospectively.They were randomly divided into the training cohort(n=170) and validation cohort(n=74) at a ratio of 7:3.The radiomics features were extracted from the second phase of DCE-MRI images.The least absolute shrinkage and selection operator(LASSO) regression method was used to screen optimal features and construct a radiomics signature in the training cohort.The receiver operating characteristic(ROC) curve was used to evaluate the diagnostic capability of the radiomics signatures.Results 8 features were selected to construct the radiomics signature.In the training and validation cohorts,the area under the curve(AUC) of radiomics signature differentiating benign and malignant lesions were 0.891 and 0.878,respectively,with the optimal thresholds of 0.632 and 0.699,sensitivity and specificity of 89.00% and 74.29%,86.05 % and 83.87% respectively.The classification accuracies of invasive ductal carcinoma,mucinous carcinoma,medullary carcinoma,fibroadenoma and tubular adenoma were high,with the accuracy of 91.82%,100%,100%,83.72%,and 100%.Conclusion The radiomic features extracted from DCE-MRI are helpful in differential diagnosis of benign and malignant breast lesions.The radiomics signature established by combining the radiomics features screened by the LASSO algorithm has high diagnostic predictability.更多还原