| 张子健,马剑雄,柏豪豪,王颖,孙磊,卢斌,高之浩,于成双,孙汉宸,马信龙.基于深度学习构建模型识别下肢全长正位X线片中的下肢力线关键点及自动测量关键角度[J].中国医学影像技术,2022,38(6):901~906 |
| 基于深度学习构建模型识别下肢全长正位X线片中的下肢力线关键点及自动测量关键角度 |
| Automatic measurement model for recognizing key points and automatic measurement of key angles of lower limb alignment on anteroposterior full-length lower limb radiograms based on deep learning |
| 投稿时间：2021-09-11 修订日期：2022-03-21 |
| DOI：10.13929/j.issn.1003-3289.2022.06.025 |
| 中文关键词: 膝内翻 膝外翻 X线 下肢力线 关键角度 深度学习 自动测量 |
| 英文关键词:genu varum genu valgum X-rays lower limb alignment key angles deep learning automatic measurement |
| 基金项目:国家重点研发计划（2018YFB1307802）。 |
| 作者 | 单位 | E-mail | | 张子健 | 天津大学天津医院骨科研究所, 天津 300050
天津市骨科生物力学与医学工程重点实验室, 天津 300050 | | | 马剑雄 | 天津大学天津医院骨科研究所, 天津 300050
天津市骨科生物力学与医学工程重点实验室, 天津 300050 | mjx969@163.com | | 柏豪豪 | 天津大学天津医院骨科研究所, 天津 300050
天津市骨科生物力学与医学工程重点实验室, 天津 300050 | | | 王颖 | 天津大学天津医院骨科研究所, 天津 300050
天津市骨科生物力学与医学工程重点实验室, 天津 300050 | | | 孙磊 | 天津大学天津医院骨科研究所, 天津 300050
天津市骨科生物力学与医学工程重点实验室, 天津 300050 | | | 卢斌 | 天津大学天津医院骨科研究所, 天津 300050
天津市骨科生物力学与医学工程重点实验室, 天津 300050 | | | 高之浩 | 天津大学天津医院骨科研究所, 天津 300050
天津市骨科生物力学与医学工程重点实验室, 天津 300050 | | | 于成双 | 天津大学天津医院骨科研究所, 天津 300050
天津市骨科生物力学与医学工程重点实验室, 天津 300050 | | | 孙汉宸 | 天津大学天津医院骨科研究所, 天津 300050
天津市骨科生物力学与医学工程重点实验室, 天津 300050 | | | 马信龙 | 天津大学天津医院骨科研究所, 天津 300050
天津市骨科生物力学与医学工程重点实验室, 天津 300050 | |
|
| 摘要点击次数: 928 |
| 全文下载次数: 222 |
| 中文摘要: |
| 目的 基于深度学习（DL）方法构建自动测量下肢全长正位X线片关键角度模型，评估其临床应用价值。方法 回顾性选取634幅下肢全长正位X线片，由5名骨科医师分别标注下肢力线关键点，包括髋关节中心、股骨髁间窝顶点、胫骨髁间嵴中点、股骨内侧和外侧髁最低点、胫骨内侧和外侧平台最低点、距骨宽度中点，并建立数据集。采用高分辨率网络（HRNet）进行迁移学习，构建自动检测关键点模型，以5折交叉验证筛选最优模型，确定关键点坐标后，通过余弦定律计算关键角度机械股骨远端外侧角（mLDFA）、胫骨近端内侧角（MPTA）、股骨胫骨关节线夹角（JLCA）及髋-膝-踝角（HKA），实现自动测量关键角度，并以关键点自动检测模型和通过余弦定律计算所得关键角度共同构建自动测量关键角度模型。随机选取50幅图像，由另3名骨科医师手动测量上述关键角度，评估自动测量关键角度模型与医师测量结果的一致性。结果 3名骨科医师所测mLDFA、MPTA、JLCA及HKA的均值分别为（88.50±2.59）°、（86.41±2.25）°、（2.90±2.27）°及（174.62±3.97）°；自动测量关键角度模型所获结果分别为（88.48±2.60）°、（86.52±2.57）°、（3.11±2.41）°及（174.53±3.99）°，与医师测量结果的一致性较好（ICC=0.897、0.888、0.826、0.996）。结论 所构建的自动测量下肢全长正位X线片关键角度模型有助于识别骨科关键点和测量关键角度。 |
| 英文摘要: |
| Objective To construct an automatic measurement model of key angles on anteroposterior full-length lower limb radiograms based on deep learning(DL), and to evaluate its clinical application value. Methods Totally 634 anteroposterior full-length lower limb radiograms were retrospectively selected, and the key points of lower limb alignment were marked by 5 orthopedic surgeons, including the center of hip joint, the apex of femoral intercondylar fossa, the midpoint of tibial intercondylar ridge, the lowest point of medial and lateral femoral condyles, the nadir of medial and lateral tibial plateau and the midpoint of talus width, then the data set was established. High resolution network (HRNet) and transfer learning were used to construct the automatic detection model of key points, and the optimal model was selected with 5-fold cross validation. The coordinates of key points were detected, the key angles mechanical lateral distal femoral angle (mLDFA), medial proximal tibial angle (MPTA), joint line convergence angle (JLCA) and hip-knee-ankle angle (HKA) were calculated by cosine law, so as to realize the automatic measurement of key angles. The key angle automatic measurement model was constructed with automatic detection model of key points together with the key angles calculated by cosine law. On 50 randomly selected radiograms, the above mentioned key angles were measured by another 3 orthopedic doctors, then the consistency of the automatic detection model of key angles and the results measured by the physicians were evaluated. Results The mean values of mLDFA, MPTA, JLCA and HKA measured by 3 orthopedic surgeons was (88.50±2.59)°, (86.41±2.25)°, (2.90±2.27)° and (174.62±3.97)°, respectively, those obtained with the automatic measurement model of key angles was (88.48±2.60)°, (86.52±2.57)°, (3.11±2.41)° and (174.53±3.99)°, respectively, showing good agreement with the physician's measurement results (ICC=0.897, 0.888, 0.826, 0.996). Conclusion The automatic measurement model of key angles on anteroposterior full-length lower limb radiograms based on DL was helpful for orthopedist to identify key points and to measure key angles. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
