温馨,左立平,王勇,田子玉,卢飞,石硕,常玲玉,纪宇,张冉,于德新.影像组学鉴别脊柱骨岛与成骨型转移癌[J].中国医学影像技术,2024,40(5):758~763
影像组学鉴别脊柱骨岛与成骨型转移癌
CT radiomics for differentiating spinal bone island and osteoblastic bone metastases
投稿时间:2023-12-11  修订日期:2024-01-15
DOI:10.13929/j.issn.1003-3289.2024.05.026
中文关键词:  脊柱  骨硬化  肿瘤转移  影像组学  体层摄影术,X线计算机
英文关键词:spine  osteosclerosis  neoplasm metastasis  radiomics  tomography, X-ray computed
基金项目:
作者单位E-mail
温馨 山东大学齐鲁医院放射科, 山东 济南 250012
滨州市第二人民医院放射科, 山东 滨州 256800 
 
左立平 山东大学齐鲁医院放射科, 山东 济南 250012  
王勇 滨州市第二人民医院放射科, 山东 滨州 256800  
田子玉 山东大学齐鲁医院放射科, 山东 济南 250012  
卢飞 潍坊医学院医学影像学院, 山东 潍坊 261000  
石硕 山东大学齐鲁医院放射科, 山东 济南 250012  
常玲玉 山东大学齐鲁医院放射科, 山东 济南 250012  
纪宇 山东大学第二医院放射科, 山东 济南 250033  
张冉 慧影医疗科技(北京)股份有限公司, 北京 100089  
于德新 山东大学齐鲁医院放射科, 山东 济南 250012 yudexin0330@sina.com 
摘要点击次数: 153
全文下载次数: 121
中文摘要:
      目的 观察CT影像组学鉴别脊柱骨岛(BI)与成骨型转移癌(OBM)的价值。方法 回顾性分析来自3个医疗机构的98例BI患者109个病灶及158例OBM患者282个病灶(包括48例肺癌103个转移灶、52例乳腺癌86个转移灶及58例前列腺癌93个转移灶);以机构1数据为内部数据集并按7 ∶ 3比例分为内部训练集与内部验证集,以机构2、3数据作为外部数据集;并以性别分为女性数据子集(包括肺癌与乳腺癌OBM)及男性数据子集(包括肺癌与前列腺癌OBM)。基于CT图像提取、筛选影像组学特征并构建支持向量机(SVM)模型,包括模型1(鉴别BI与OBM)、模型2(鉴别女性肺癌与乳腺癌OBM)及模型3(鉴别男性肺癌与前列腺癌OBM)。绘制受试者工作特征曲线,计算曲线下面积(AUC),评估并比较模型1、单一CT值及3名医师(甲、乙、丙)鉴别BI与OBM的效能,以及模型2、3鉴别不同OBM的效能。结果 模型1鉴别内部训练集、内部验证集及外部数据集脊柱OBM与BI的AUC分别为0.99、0.98及0.86。针对内部训练集,模型1鉴别BI与OBM的AUC高于医师甲(AUC=0.78)、乙(AUC=0.87)、丙(AUC=0.93)及单一平均CT值(AUC=0.78,P均<0.05)。模型2鉴别内部训练集、内部验证集及外部数据集女性肺癌与乳腺癌OBM的AUC分别为0.79、0.75及0.73;模型3鉴别各集男性肺癌与前列腺癌OBM的AUC分别为0.77、0.74和0.77。结论 根据 CT影像组学SVM模型能可靠鉴别BI与OBM。
英文摘要:
      Objective To observe the value of CT radiomics for differentiating spinal bone islands (BI) and osteoblastic metastases (OBM). Methods Data of 109 BI lesions in 98 patients and 282 OBM lesions in 158 patients (including 103 OBM in 48 lung cancer cases, 86 OBM in 52 breast cancer cases and 93 OBM in 58 prostate cancer cases) from 3 medical institutions were retrospectively analyzed. Data obtained from institution 1 were used as the internal dataset and divided into internal training set and internal validation set at a ratio of 7 : 3, from institution 2 and 3 were used as external dataset. All datasets were divided into female data subset (including OBM of female lung cancer and breast cancer) and male data subset (including OBM of male lung cancer and prostate cancer). Radiomics features were extracted and screened to construct 3 different support vector machine (SVM) models, including model1 for distinguishing BI and OBM, model2 for differentiating OBM of female lung cancer and breast cancer, and model3 for differentiating OBM of male lung cancer and prostate cancer. Diagnostic efficacy of model1, CT value alone and 3 physicians (A, B, C) for distinguishing BI and OBM were assessed, as well as differentiating efficacy for different OBM of model2 and model3. Receiver operating characteristic (ROC) curves were drawn, and area under the curves (AUC) were calculated and compared. The differential diagnostic efficacy of model2 and model3 were also assessed with ROC analysis and AUC. Results AUC of model1 for distinguishing spinal OBM from BI in internal training set, internal validation set and external dataset was 0.99, 0.98 and 0.86, respectively. In internal training set, model1 had higher AUC for distinguishing BI and OBM than that of physician A (AUC=0.78), B (AUC=0.87) and C (AUC=0.93) as well as that of mean CT value (AUC=0.78, all P<0.05). AUC in internal training set, internal validation set and external dataset of model2 for identifying female lung cancer and breast cancer OBM was 0.79, 0.75 and 0.73, respectively, of model3 for discriminating male lung cancer from prostate cancer OBM was 0.77, 0.74 and 0.77, respectively. Conclusion CT radiomics SVM model might reliablely distinguish OBM and BI.
查看全文  查看/发表评论  下载PDF阅读器