| 王光松,石大发,郭秋,张浩然,王思远,任振东,任克.基于乳腺X线影像组学辅助诊断BI-RADS 4和5类良性病变[J].中国医学影像技术,2022,38(4):540~544 |
| 基于乳腺X线影像组学辅助诊断BI-RADS 4和5类良性病变 |
| Radiomics based on mammography for assisting diagnosis of BI-RADS 4 and 5 benign lesions |
| 投稿时间：2021-08-10 修订日期：2021-11-17 |
| DOI：10.13929/j.issn.1003-3289.2022.04.015 |
| 中文关键词: 乳腺肿瘤 乳腺X线摄影 影像组学 乳腺影像报告和数据系统 |
| 英文关键词:breast neoplasms mammography radiomics breast imaging reporting and data system |
| 基金项目:厦门大学附属翔安医院引进高层次人才科研启动基金（PM201809170011）。 |
|
| 摘要点击次数: 1338 |
| 全文下载次数: 505 |
| 中文摘要: |
| 目的 评价乳腺X线影像组学辅助诊断乳腺影像报告和数据系统（BI-RADS）4和5类良性病变的价值。方法 回顾性分析经病理证实的344例乳腺病变患者，包括194例良性病变、150例恶性病变；根据接受乳腺X线检查时间将前70%归为训练集（n=240）、后30%归为验证集（n=104）。基于头足（CC）位和内外斜（MLO）位图像提取影像组学特征，以组内相关系数（ICC）、Spearman相关性及最小绝对收缩和选择算子（LASSO）筛选最佳影像组学特征，应用支持向量机建立组学模型，预测BI-RADS 4和5类中的良性病变。由2名放射科医师对训练集及验证集数据进行判断，采用受试者工作特征（ROC）曲线评估组学模型与放射科医师的诊断效能，并比较其差异。结果 分别基于CC位和MLO位图像提取92个影像组学特征，并最终选出7个及2个最佳影像组学特征。ROC曲线显示，组学模型诊断训练集及验证集良性乳腺病变的曲线下面积（0.92、0.87）均大于放射科医师（0.76、0.75，Z=-4.20、-2.40，P均<0.05），诊断特异度（0.93、0.85）均高于放射科医师（0.71、0.70，χ2=9.26、5.11，P均<0.05），而诊断敏感度（0.81、0.81）与放射科医师差异均无统计学意义（0.81、0.82，χ2=0、0.10，P均>0.05）。结论 乳腺X线影像组学有助于诊断BI-RADS 4和5类良性病变。 |
| 英文摘要: |
| Objective To explore the value of radiomics based on mammography for assisting diagnosis of breast imaging reporting and data system (BI-RADS) 4 and 5 benign lesions. Methods Preoperative mammography data of 344 patients with pathologically confirmed breast lesions, including 194 benign cases and 150 malignant ones, were retrospectively analyzed. The first 70% patients who underwent mammography were divided into training set (n=240), while the last 30% were divided into the validation set (n=104). Radiomics features were extracted based on craniocaudal (CC) position and mediolateral oblique (MLO) position images. Then intra-class correlation coefficient (ICC), Spearman correlation and least absolute shrinkage and selection operator (LASSO) were used to select the optimal radiomics features, support vector machines was used to build a radiomics model for predicting benign lesions in BI-RADS 4 and 5. Data of training set and validation set were judged by 2 radiologists. Receiver operating characteristic (ROC) curves were drawn to assess the diagnostic efficacy of radiomics model and radiologists, respectively, and the effectiveness were compared. Results Ninety-two radiomics features were separately extracted, and finally 7 and 2 best radiomics features were screened out from CC position and MLO position images. ROC curves show that, in training and validation set, the area under the curve (AUC) of radiomics model for diagnosing benign breast lesions (0.92, 0.87) were larger than those of radiologists (0.76, 0.75; Z=-4.20, -2.40, both P<0.05), the diagnostic specificity)of radiomics model (0.93, 0.85) were higher than those of radiologists (0.71, 0.70, χ2=9.26, 5.11, both P<0.05), while the sensitivity of radiomics model (0.81, 0.81) were not significant different with those of radiologists (0.81, 0.82; χ2=0, 0.10, both P>0.05). Conclusion Mammography-based radiomics could help to diagnose benign lesions in BI-RADS 4 and 5. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
