| 杨鑫,黄京城,胡景卉,侯承师,王文剑,罗先富,徐军.三维重建CT影像组学诊断椎体急性轻度压缩性骨折[J].中国医学影像技术,2023,39(11):1710~1715 |
| 三维重建CT影像组学诊断椎体急性轻度压缩性骨折 |
| 3D reconstruction CT radiomics for diagnosis of acute mild vertebral compression fractures |
| 投稿时间：2023-07-01 修订日期：2023-09-10 |
| DOI：10.13929/j.issn.1003-3289.2023.11.026 |
| 中文关键词: 骨折,压缩性 脊柱 体层摄影术,X线计算机 成像,三维 影像组学 |
| 英文关键词:fractures, compression spine tomography, X-ray computed imaging, three-dimensional radiomics |
| 基金项目: |
| 作者 | 单位 | E-mail | | 杨鑫 | 苏北人民医院医学影像科, 江苏 扬州 225001
大连医科大学研究生院, 辽宁 大连 116044 | | | 黄京城 | 苏北人民医院医学影像科, 江苏 扬州 225001
大连医科大学研究生院, 辽宁 大连 116044 | | | 胡景卉 | 苏北人民医院医学影像科, 江苏 扬州 225001 | | | 侯承师 | 苏北人民医院医学影像科, 江苏 扬州 225001
大连医科大学研究生院, 辽宁 大连 116044 | | | 王文剑 | 苏北人民医院医学影像科, 江苏 扬州 225001 | | | 罗先富 | 苏北人民医院医学影像科, 江苏 扬州 225001 | | | 徐军 | 苏北人民医院医学影像科, 江苏 扬州 225001 | 58500372@qq.com |
|
| 摘要点击次数: 1663 |
| 全文下载次数: 560 |
| 中文摘要: |
| 目的 观察三维重建CT影像组学模型诊断椎体急性轻度压缩性骨折的价值。方法 回顾性分析362例胸/腰椎(T1~L5)轻度压缩性骨折患者CT及MRI资料,包括274例单发及88例多发椎体骨折。以MRI为参考,于轴位、矢状位及冠状位三维重建CT图中标记骨折椎体为"1",将相邻正常椎体标记为"0"。以60例(含120个椎体)为验证集,将其余302例604个椎体按7 ∶ 3比例分为训练集(211例,422个椎体)及测试集(91例,182个椎体);于三维重建CT图中勾画椎体ROI,获得感兴趣容积(VOI),提取并筛选其最优影像组学特征,以logistic回归、随机森林及支持向量机分类器进行分类(1或0)并构建影像组学模型。绘制受试者工作特征曲线,评估模型诊断椎体急性轻度压缩性骨折的效能。结果 基于轴位、矢状位、冠状位ROI及VOI构建模型在测试集中的曲线下面积(AUC)均＞0.80,以logistic回归分类器所构模型的校准度较优。以logistic回归基于矢状位ROI及VOI所构建模型的AUC均高于其他模型(P均＜0.05),其在验证集的AUC分别为0.83及0.88。结论 三维重建CT影像组学能有效诊断椎体急性轻度压缩性骨折,以logistic回归基于椎体矢状位ROI及VOI所构建模型的诊断效能较佳。 |
| 英文摘要: |
| Objective To explore values of 3D reconstructive CT radiomics models for diagnosing acute mild vertebral compression fractures. Methods CT and MRI data of 362 patients with acute mild compression fractures of thoracic/lumbar vertebral bodies (T1-L5), including 274 cases of single and 88 cases of multiple vertebral body fractures were retrospectively analyzed. Taken MRI as the reference standard, the fractured vertebra was marked as "1", and the adjacent normal vertebra was marked as "0" on the axial, sagittal and coronal 3D reconstructed CT images. Then 120 vertebral bodies in 60 cases were taken as verification set, the other 302 cases with 604 vertebral bodies were divided into training set (211 cases, 422 vertebral bodies) or test set (91 cases, 182 vertebral bodies) at the ratio of 7 ∶ 3. ROI of vertebral body was drawn on 3D reconstructive CT images, and volume of interest (VOI) was obtained. The optimal radiomics features were extracted, screened and classified (1 or 0) using logistic regression, and random forest and support vector machine classifiers and radiomics models were constructed. Receiver operating characteristic curves were drawn to evaluate the efficacy of each model for diagnosing acute mild vertebral compression fracture. Results The area under the curve (AUC) of models constructed based on axial, sagittal, coronal ROI and VOI were all greater than 0.80 in testing set. The model constructed with logistic regression classifier had better calibration than the others. AUC of the model constructed with logistic regression based on sagittal ROI and VOI were all higher than other models (all P＜0.05), with AUC in validation set of 0.83 and 0.88, respectively. Conclusion 3D reconstructed CT radiomics could be used to effectively diagnose acute mild vertebral body compression fractures. Models constructed using logistic regression based on vertebral sagittal ROI and VOI had better diagnostic efficacies. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
