| 王岩,霍爱华,王大为,沈云,彭芸.基于深度学习人工智能骨龄测评系统临床应用[J].中国医学影像技术,2021,37(1):104~107 |
| 基于深度学习人工智能骨龄测评系统临床应用 |
| Clinical application of artificial intelligence bone age measurement system based on deep learning |
| 投稿时间：2019-10-13 修订日期：2020-05-11 |
| DOI：10.13929/j.issn.1003-3289.2021.01.025 |
| 中文关键词: 年龄测定,骨骼 深度学习 人工智能 |
| 英文关键词:age determination by skeleton deep learning artificial intelligence |
| 基金项目:北航-首医大数据精准医疗高精尖创新中心计划（BHME-201908、BHME-201802）。 |
|
| 摘要点击次数: 3402 |
| 全文下载次数: 1175 |
| 中文摘要: |
| 目的 观察基于深度学习的人工智能骨龄测评（BAA）系统临床应用及其对低年资儿科放射医师的辅助作用。方法 收集80名5~12岁儿童手腕骨X线正位片，对掌指骨进行骨龄测评，由高年资儿科放射医师借助BAA系统建立研究所用参考骨龄。根据中华-05法中的TW3-RUS标准，分别以BAA系统、低年资儿科放射医师及BAA系统辅助低年资儿科放射医师测评骨龄，评价BAA系统临床应用价值及其对低年资儿科放射医师的辅助作用。结果 BAA系统骨龄预测值均方根误差（RMSE）和平均绝对误差（MAE）分别为0.62岁和0.47岁，在不同年龄亚组及不同性别亚组预测骨龄均与参考骨龄显著相关（P均<0.05）。BAA系统辅助下，低年资儿科放射医师预测骨龄的RMSE和MAE分别由0.71岁和0.90岁降为0.27岁和0.38岁，用时由4 min 40 s缩短至2 min 15 s。结论 BAA系统评价掌指骨骨骺发育情况及骨龄与金标准的一致性良好，可用于辅助测评骨龄，提高低年资儿科放射医师骨龄测评水平和临床工作效率。 |
| 英文摘要: |
| Objective To explore the clinical application value of artificial intelligence bone age assessment system (BAA) based on deep learning and its auxiliary effect for junior pediatric radiologists. Methods Bone age of metacarpal and phalangeal bones of 80 children aged 5-12 years old were evaluated using hand and wrist bone X-ray films. The ground truth of bone ages (reference bone ages) was determined by a senior pediatric radiologist with the assistance of BAA system. Then, bone age was evaluated by a junior pediatric radiologist alone and by the same radiologist under assistance of BAA system using the TW3-C-RUS standard from the Skeletal Development Standards of Hand and Wrist for Chinese Children-China 05. The clinical application value of BAA system and its auxiliary effect on junior pediatric radiologists were evaluated. Results The root mean square error (RMSE) and mean absolute error (MAE) of BAA system was 0.62 and 0.47 years, respectively. The predicted bone age of different age subgroups and gender subgroups were significantly correlated with the reference bone age (all P<0.05). Under the assistance of BAA system, RMSE and MAE of bone age prediction decreased from 0.71 and 0.90 years to 0.27 and 0.38 years, respectively, and the mean evaluation time shortened from 4 min 40 s to 2 min 15 s. Conclusion BAA system had good consistency with the gold standard in evaluation of metacarpal and phalangeal epiphyseal development and bone age, which could be used to help junior pediatric radiologists for improving the level and efficiency of bone age evaluation. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
