摘要
目的研究对比证明解决人工判读骨龄存在耗时长、人为主观影响大、结果一致性稳定性差等问题。方法使用G-P图谱法、TW计分法、中华05等方法对骨龄X光影像进行对照,开展人工智能和人工判读、以及人工智能辅助人工判读的研究,并进行多阅片者间差异性研究。结果基于TW3标准,250份儿童骨龄片由人工智能系统与医生对比,TW3-AI模型的平均处理时间为1.5±0.2 s,明显短于医生的525.6±55.5 s,准确性与可靠性上TW3-AI模型与专家判读结果的均方根为0.50年,表明两者高度一致;基于G-P标准,745份生长发育异常病例骨龄,医生平均判读耗时约2 min,AI模型仅需要1~2 s,准确性上,AI系统与金标准相差1岁以内的平均比例为84.60%;基于中华05标准,人工组阅片平均耗时明显高于AI一致性辅助评估。结论儿童骨龄智能检测系统,可在秒级完成儿童骨龄影像分析并给出骨化中心评级、骨龄等量化结果,从而辅助医生快速进行疾病诊断与疗效评价,为儿童内分泌疾病诊疗提供决策依据。
Objective To study and compare the results to prove that it takes a long time to solve the problems of manual bone age interpretation,such as high human subjective influence,poor consistency and stability of results,etc.Methods G-P map,TW score method,Zhonghua 05 and other methods were used to compare bone age X-ray images.Artificial intelligence,artificial interpretation and artificial intelligence-assisted artificial interpretation were studied,and the differences among multiple readers were studied.Results Based on the standard TW3,250 bone age images of children were compared by artificial intelligence system and doctor,TW3-AI model interpretation efficiency on the average processing time was 1.5±0.2 s,significantly shorter than the doctor’s 525.6±55.5 s.In terms of accuracy and reliability,the root mean square of TW3-AI model and expert interpretation results was 0.50 years,indicating a high degree of consistency between the two.Based on G-P standard,the bone age of 745 patients with abnormal growth and development was estimated.The average time of doctors’interpretation was about 2 min,and the AI model only needed 1~2 s.In terms of accuracy,the average proportion of the AI system was less than one year from the gold standard,84.60%.Based on the Chinese 05 standard,the average time of manual group reading was significantly higher than that of AI conformance assisted assessment.Conclusion The intelligent detection system of children’s bone age can complete the imaging analysis of children’s bone age at the second level and provide quantitative results such as ossification center rating and bone age,so as to assist doctors in rapid disease diagnosis and efficacy evaluation,and provide decision-making basis for the diagnosis and treatment of children’s endocrine diseases.
作者
孙梦莎
丁永红
颜子夜
苏晓鸣
SUN Mengsha;DING Yonghong;YAN Ziye;SU Xiaoming(Hangzhou Yitu Healthcare Technology Co.,Ltd.,Hangzhou Zhejiang 310012,China;Shanghai Key Laboratory of Artificial Intelligence for Medical Image and Knowledge Graph,Shanghai 200051,China)
出处
《中国医疗设备》
2021年第3期28-32,共5页
China Medical Devices
基金
国家重点研发计划(2019YFB1404805)。
关键词
儿童内分泌疾病
儿童骨龄
辅助诊断
人工智能
endocrine diseases in children
bone age of children
computer aided diagnosis
artificial intelligence
