林奕军,杨光耀,蒋耀先,常晓丹.基于深度学习测量X线片髋关节外侧中心边缘角和Sharp角评估髋关节发育不良[J].中国医学影像技术,2022,38(11):1710~1714
基于深度学习测量X线片髋关节外侧中心边缘角和Sharp角评估髋关节发育不良
Measuring lateral center edge angle and Sharp angle of hip joint on X-ray films based on deep learning for diagnosing hip dysplasia
投稿时间:2022-06-01  修订日期:2022-08-24
DOI:10.13929/j.issn.1003-3289.2022.11.026
中文关键词:  髋关节  放射摄影术  深度学习
英文关键词:hip joint  radiography  deep learning
基金项目:
作者单位E-mail
林奕军 大连大学附属中山医院放射科, 辽宁 大连 116001
齐齐哈尔市第一医院CT诊断室, 黑龙江 齐齐哈尔 161000 
 
杨光耀 东北林业大学信息与计算机工程学院, 黑龙江 哈尔滨 150040  
蒋耀先 重庆医科大学附属儿童医院放射科, 重庆 400014  
常晓丹 大连大学附属中山医院放射科, 辽宁 大连 116001 302647771@qq.com 
摘要点击次数: 907
全文下载次数: 357
中文摘要:
      目的 探讨基于深度学习测量X线片髋关节外侧中心边缘(LCE)角和Sharp角评估髋关节发育不良的价值。方法 回顾性收集384例成人双髋关节正位数字化X线片。由2名影像科医师标注髋臼关键点和股骨轮廓用于模型训练,经十折交叉验证得到预测结果;以设定程序自动测量双侧髋关节LCE角和Sharp角,由上述医师于X线片中手动测量LCE角和Sharp角并诊断髋关节发育不良及交界性发育不良。对比手动测量与自动测量LCE角、Sharp角结果的差异;观察手动测量与自动测量的相关性;以组内相关系数(ICC)评估测量结果的一致性;采用受试者工作特征(ROC)曲线评估自动测量LCE角、Sharp角诊断髋关节发育不良的效能。结果 共测量768侧髋。手动测量与自动测量左侧LCE角、右侧LCE角、右侧Sharp角结果差异无统计学意义(P均>0.05),手动测量左侧Sharp角小于自动测量结果(P<0.05)。对左、右侧LCE角及左、右侧Sharp角,手动测量与自动测量结果的ICC分别为0.944、0.904、0.890及0.887,r值分别为0.948、0.924、0.910及0.887(P均<0.05)。共诊断73侧髋关节发育不良、93侧交界性髋关节发育不良。基于自动测量诊断髋关节发育不良的敏感度为87.67%(64/73),准确率为95.57%(734/768);诊断交界性发育不良的敏感度为66.67%(62/93),准确率为91.54%(703/768)。自动测量LCE角(阈值为25.24°)和Sharp角(阈值为43.47°)诊断髋关节发育不良的曲线下面积(AUC)分别为0.928和0.906(P均<0.05)。结论 基于深度学习自动测量X线片中双侧髋关节LCE角和Sharp角可诊断髋关节发育不良。
英文摘要:
      Objective To investigate the value of measuring the lateral center edge (LCE) angle and Sharp angle of hip joint on X-ray films based on deep learning for diagnosing hip dysplasia. Methods Bilateral-hip anteroposterior digital radiographs of 384 adults were retrospectively analyzed. The key points of the acetabulum and the outline of the femur were labeled by 2 radiologists for model training. Ten-fold cross-validation was used to obtain the prediction results, while LCE angle and Sharp angle of bilateral hip joints were automatically measured with the program. Then LCE angle and Sharp angle on X-ray films were manually measured by the above radiologists for diagnosing hip dysplasia and borderline dysplasia. The manually measured and automatically measured LCE angle and Sharp angle were compared, and the correlations were observed. Intra-class correlation coefficient (ICC) was used to evaluate the consistency of the measurement results. Receiver operating characteristic (ROC) curve was used to evaluate the efficacy of automatic measurement of LCE angle and Sharp angle for diagnosing hip dysplasia. Results Totally 768 hip joints were measured. No significant difference of the left and right LCE angle nor right Sharp angle was found between manual measured and automatic measured results (all P>0.05), while the manually measured left Sharp angle was smaller than automatically measured ones (P<0.05). For left and right LCE angle, left and right Sharp angle, the ICC of manual measurement and automatic measurement was 0.944, 0.904, 0.890 and 0.887, and the r value was 0.948, 0.924, 0.910 and 0.887, respectively (all P<0.05). Hip dysplasia was diagnosed in 73 hips, while borderline hip dysplasia was diagnosed in 93 hips. The sensitivity and accuracy of diagnosing hip dysplasia based on automatic measurement was 87.67% (64/73) and 95.57% (734/768), of diagnosing borderline dysplasia was 66.67% (62/93) and 91.54% (703/768), respectively. The area under the curve (AUC) of automatic measurement of LCE angle (threshold value of 25.24°) for diagnosing hip dysplasia was 0.928 (P<0.05), and of Sharp angle (threshold value of 43.47°) was 0.906 (P<0.05). Conclusion Automatic measurement of bilateral hip LCE angle and Sharp angle on radiographs based on deep learning could be used to diagnose hip dysplasia.
查看全文  查看/发表评论  下载PDF阅读器