| 刘壮盛,李昌林,衣利磊,李智,陈业航,李荣岗,罗学毛,龙晚生,冯宝.MRI纹理特征分析预测浸润性乳腺癌脉管浸润[J].中国医学影像技术,2020,36(11):1637~1642 |
| MRI纹理特征分析预测浸润性乳腺癌脉管浸润 |
| MRI texture analysis in predicting lymphovascular invasion of invasive breast cancer |
| 投稿时间：2019-10-08 修订日期：2020-03-12 |
| DOI：10.13929/j.issn.1003-3289.2020.11.009 |
| 中文关键词: 乳腺肿瘤 肿瘤转移 磁共振成像 纹理分析 |
| 英文关键词:breast neoplasms neoplasm metastasis magnetic resonance imaging texture analysis |
| 基金项目:国家自然科学基金（81960324）、广东省医学科研基金（A2020622）、江门市中心医院科研杰青项目（J201904）。 |
| 作者 | 单位 | E-mail | | 刘壮盛 | 中山大学附属江门医院放射科, 广东 江门 529000 | | | 李昌林 | 桂林电子科技大学电子工程与自动化学院, 广西 桂林 541004
桂林航天工业学院电子信息与自动化系, 广西 桂林 541004 | | | 衣利磊 | 佛山市中医院放射科, 广东 佛山 528000 | | | 李智 | 桂林电子科技大学电子工程与自动化学院, 广西 桂林 541004
桂林航天工业学院电子信息与自动化系, 广西 桂林 541004 | | | 陈业航 | 桂林电子科技大学电子工程与自动化学院, 广西 桂林 541004
桂林航天工业学院电子信息与自动化系, 广西 桂林 541004 | | | 李荣岗 | 中山大学附属江门医院病理科, 广东 江门 529000 | | | 罗学毛 | 中山大学附属江门医院放射科, 广东 江门 529000 | | | 龙晚生 | 中山大学附属江门医院放射科, 广东 江门 529000 | | | 冯宝 | 中山大学附属江门医院放射科, 广东 江门 529000
桂林航天工业学院电子信息与自动化系, 广西 桂林 541004 | 475795274@qq.com |
|
| 摘要点击次数: 2711 |
| 全文下载次数: 671 |
| 中文摘要: |
| 目的 评估MRI纹理特征预测浸润性乳腺癌（IBC）脉管浸润（LVI）的价值。方法 将204例接受MR检查并经术后病理证实IBC患者分为训练组80例（LVI阳性亚组30例、阴性亚组50例）、内部验证组66例（LVI阳性亚组21例、阴性亚组45例）和外部验证组58例（LVI阳性亚组20例、阴性亚组38例）。自临床及常规MRI征象中筛选LVI独立危险因素，构建主观MRI征象模型。对MRI所示病灶进行分割、纹理特征提取和筛选，构建纹理特征模型。采用受试者工作特征（ROC）曲线评价2种模型的诊断效能。结果 瘤周水肿类型（OR=3.82）和MRI腋窝淋巴结状态（OR=7.63）是LVI的独立危险因素。自4 300个纹理特征中筛选出GLRLM_LRHGE_1_0.67_Equal_32、GLRLM_GLV_1_0.67_Equal_32和GLRLM_GLV_1_0.67_Equal_64共3个有效特征构建纹理模型。内、外部验证组中，纹理特征模型诊断LVI的准确率分别为86.36%和79.31%、敏感度为66.67%和60.00%、特异度为95.56%和89.47%、AUC为0.86和0.84，诊断效能均高于主观MRI征象模型（P均=0.04）。结论 根据MRI纹理特征可在术前有效预测IBC LVI状态。 |
| 英文摘要: |
| Objective To explore the value of MRI texture analysis in predicting lymphovascular invasion (LVI) of invasive breast cancer (IBC). Methods Breast MRI of 204 IBC patients confirmed by pathology were retrospectively collected and split into training group (n=80, 30 LVI-positive and 50 LVI-negative), internal validation group (n=66, 21 LVI-positive and 45 LVI-negative) and external validation group (n=58, 20 LVI-positive and 38 LVI-negative). The independent risk factors of LVI were screened from clinical characteristics and conventional MRI features, and subjective MRI signs model was constructed. The lesions on MRI were segmented, texture features were extracted and screened, and texture feature model was constructed. Receiver operating characteristic (ROC) curve was used to evaluate the diagnostic efficacy of models. Results Peritumoral edema type (OR=3.82) and MRI axillary lymph node status (OR=7.63) were independent predictors of LVI. Among 4 300 texture features, 3 effective features, i.e. GLRLM_LRHGE_1_0.67_Equal_32, GLRLM_GLV_1_0.67_Equal_32 and GLRLM_GLV_1_0.67_Equal_64,were selected for establishing texture features model. In internal and external validation groups, the accuracy of the texture features model for diagnosis of LVI was 86.36% and 79.31%, the sensitivity was 66.67% and 60.00%, the specificity was 95.56% and 89.47%, and the area under the ROC curve (AUC) was 0.86 and 0.84, respectively. The diagnostic performance of texture features model was prior to that of the subjective model both in internal and external validation groups (both P=0.04). Conclusion MRI texture analysis could be used to preoperatively predict LVI status of IBC. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
