摘要
目的比较在有无人工智能支持系统辅助下,放射科医师对儿童左手X线平片的骨龄评估效能。方法回顾性分析280例2~15岁患者的左手X线平片。由两名低年资放射科医师分别在有无人工智能支持系统下对以上X线平片进行骨龄评估,评估方法采用适合中国儿童发育标准的中华05法(TW3-Chinese RUS和TW3-Chinese Carpal骨龄标准),以两位高年资医师评估结果的均值为参考标准,比较在有无人工智能支持系统下低年资医师评估骨龄的准确率、均方根误差(RMSE)、组内相关系数(ICC)及读片时间。结果人工智能支持系统对RUS骨及腕骨评估结果与参考标准相比误差在0.5年内的准确率分别为72.3%、82.4%,RMSE分别为0.584年、0.486年;在人工智能辅助下,两位医师对桡骨、尺骨、掌指骨及腕骨评估误差在0.5年内的准确率升高,且差异具有显著意义(P<0.05),两位医师对手腕部骨龄评估RMSE均减低,但医师2在有无人工智能支持下对腕骨评估的RMSE未见明显统计学差异。相较于医师独立评估,人工智能辅助下,两位医师之间掌指骨及腕骨骨龄评测结果的ICC均高于无人工智能支持结果(0.989 vs 0.981,0.991 vs 0.982)。结论在人工智能支持系统辅助下,低年资医师能提高手腕部骨龄评估准确性及一致性,减少评测时间。
Objective To compare bone age assessment performance of radiologists reading left-hand radiographs unaided versus aided by an artificial intelligence(AI)system.Methods Totally 280 left-hand radiographs of children aged 2-15 years were retrospectively studied.According to China 05(TW3-C RUS and TW3-Carpal)standard,two junior radiologists conducted bone age assessment unaided versus aided by an artificial intelligence(AI)system.The mean value of the bone age assessed by two senior radiologists was the reference standard.Accuracy,root mean squared error(RMSE),intraclass correlation coefficient(ICC),and time efficiency of bone age assessment were compared.Results AI BAA accuracy was 72.3%within 0.5 year of RUS and 82.4%of carpal,and the RMSE were 0.584 years and 0.486 years.The bone age assessment accuracy within 0.5 years was higher and the RMSE was lower with AI aided than unaided reading,while no significant difference of carpal RMSE was found between physician 2 with and without aiding AI.With AI aiding,the ICC of RUS and carpal between Physician1 and Physician2 are higher than unaided by AI(0.989 vs 0.981,0.991 vs0.982).Conclusion With AI aiding,the junior radiologists increased bone age assessment accuracy and consistency,decreased reading time.
作者
李然然
杨芸晓
高剑波
于湛
LI Ranran;YANG Yunxiao;GAO Jianbo(Department of Radiology,the First Affiliated Hospital of Zhengzhou University,Zhengzhou,Henan Province 450052,P.R.China)
出处
《临床放射学杂志》
北大核心
2021年第3期560-563,共4页
Journal of Clinical Radiology
关键词
人工智能
深度学习
骨龄评估
中华05
Artificial intelligence
Deep learning
Bone age assessment
China 05
作者简介
通讯作者:于湛
