| 康敏,王齐艳,蒲杨梅,丁立,邹雪瑾.人工智能骨龄评测系统评估儿童腕骨骨龄[J].中国医学影像技术,2019,35(12):1804~1807 |
| 人工智能骨龄评测系统评估儿童腕骨骨龄 |
| Artificial intelligence system for assessment of children carpal bone age |
| 投稿时间：2019-07-09 修订日期：2019-09-26 |
| DOI：10.13929/j.1003-3289.201907065 |
| 中文关键词: 年龄测定,骨骼 腕骨 深度学习 儿童 体层摄影术,X线计算机 |
| 英文关键词:age determination by skeleton carpal bones deep learning child tomography, X-ray computed |
| 基金项目: |
|
| 摘要点击次数: 2641 |
| 全文下载次数: 752 |
| 中文摘要: |
| 目的 探讨人工智能（AI）系统评测儿童腕骨骨龄的可行性。方法 回顾性分析130幅1~13岁儿童左手骨龄X线片。以3名中高年资放射科医师腕骨骨龄评测结果为参考标准，计算并对比AI系统（简称模型）及3名低年资放射科医师（医师1、2、3，简称医师）与参考标准之间腕骨骨龄和腕骨成熟度分值的均方根误差（RMSE）及平均绝对误差（MAE）；采用组内相关系数（ICC）评价模型、医师与参考标准之间评测骨龄结果的一致性；比较模型与医师间骨龄测评时间。结果 模型与参考标准之间腕骨骨龄的MAE、RMSE与医师1、2与参考标准之间MAE、RMSE差异均有统计学意义（P均<0.05），与医师3的MAE、RMSE差异无统计学意义（P均>0.05）。模型与参考标准之间腕骨成熟度分值的MAE、RMSE与医师1、参考标准之间MAE、RMSE差异均有统计学意义（P均<0.05），与医师2、3的MAE、RMSE差异均无统计学意义（P均>0.05）。模型与参考标准之间腕骨骨龄评测结果的ICC=0.997，医师1、2、3与参考标准之间ICC分别为0.994、0.996、0.997。模型对骨龄的测评时间均小于医师（P均<0.001）。结论 AI骨龄评测系统能够准确、快速评估儿童腕骨骨龄。 |
| 英文摘要: |
| Objective To observe the clinical feasibility of carpal bone age assessment (BAA) using artificial intelligence (AI) system. Methods Totally 130 hand-wrist radiographs of children aged 1-13 years were retrospectively studied. Carpal bone ages estimated by three senior radiologists were taken as reference standards. The root mean square error (RMSE) and mean absolute error (MAE) of carpal bone age estimations and carpal maturity scores relative to the reference standard were calculated and compared between AI system (model) and 3 junior radiologists (physician 1, 2, 3),respectively. The intraclass correlation coefficient (ICC) was used to test the agreement of BAA among the model, physicians and reference standards. BAA time was also compared between model and physicians,respectively. Results There were significant differences of carpal BAA's MAE and RMSE of model and physician 1, 2 (all P<0.001), but not between model and physician 3 (all P>0.05). There were significant differences of carpal maturity score's MAE and RMSE between model and physicians 1 (both P<0.05), while no significant difference was found between model and physician 2, 3 (all P>0.05). ICC between BAA of AI and reference standards was 0.997, between physician 1, 2, 3 and reference standards was 0.994, 0.996 and 0.997, respectively. BAA time of AI was significantly shorter than that of three physicians (all P<0.05). Conclusion Using AI BAA system can fast estimate carpal bone age with high accuracy. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
