4种新型人工智能算法在区域人群骨龄测定中的价值  

The value of novel artificial intelligence in the determination of bone age in regional population

作　　者：彭乐媛 段玉梅[1] 王永芹[1] 彭涛[1] 牛翔科 PENG Le-yuan;DUAN Yu-mei;WANG Yong-qin(Department of radiology,Affiliated Hospital of Chengdu University,Chengdu,Sichuan Province,610081,China)

机构地区：[1]成都大学附属医院放射科,四川成都610081

出　　处：《中国医学装备》2021年第11期113-117,共5页China Medical Equipment

基　　金：四川省卫生和计划生育委员会科研课题(18PJ150)“人工智能技术在磁共振计算机辅助检测前列腺癌的应用研究”。

摘　　要：目的:研究在实际临床工作中适用于区域人群的人工智能骨龄测定法。方法:选取在医院行手腕骨X射线检查的152名儿童及青少年受检者,应用《中国人手腕骨发育标准(CHN)法》对受检者骨龄进行评估,以评估的骨龄平均值为“金标准”;采用人工智能的RUS-CHN法掌指骨骨龄法、TW3-RUS掌指骨骨龄法、TW3-Carpal腕骨骨龄法和Greulich-Pyle图谱法4种不同新型全自动算法阅片,对照医师采用CHN法进行阅片,并与骨龄“金标准”相比得出平均绝对偏差(MAD)和均方根偏差(RSMD)。结果:人工智能中的4种全自动算法阅片结果与“金标准”相比MAD分别为0.56岁、0.61岁、1.09岁和0.90岁,RSMD分别为0.72岁、0.73岁、1.38岁和1.11岁,对照医师组与“金标准”相比MAD和RSMD分别为0.58岁和0.74岁;人工智能中RUS-CHN法最接近金标准,且具有统计学意义(F=0.01,F=0.06,F=5.58,F=0.34,F=0.31;P<0.05)。人工智能软件测评速度为(25.99±13.88)s显著高于医师测评速度(149.72±60.79)s,差异有统计学意义(t=24.312,P<0.05)。结论:人工智能中RUS-CHN掌指骨骨龄法诊断准确性最接近金标准,且显著提高临床工作效率。Objective:To explore the determination method of artificial intelligence for bone age,which was appropriate for regional population in actually clinical work.Methods:152 children and adolescents who underwent X-ray examination on skeleton of hand and wrist were selected.<Chinese skeleton of hand and wrist development standard(CHN)method>was applied to assess bone ages of subjects,and the mean of assessed bone ages was used as“golden standard”.The 4 kinds of different novel full-automatic algorithms included RUS-CHN metacarpophalangeal bone age method,TW3-RUS metacarpophalangeal bone age method,TW3-carpal bone age method and Greulich-Pyle atlas method of artificial intelligence were adopted to read pictures,and physician adopted CHN method to read picture.Compared with golden standard of bone age,the mean absolute deviation(MAD)and root mean square deviation(RSMD)were obtained.Results:Compared with gold standard,MADs of 4 kinds of full-automatic algorithms in artificial intelligence were 0.56 years of age,0.61 years of age,1.09 years of age and 0.90 years of age,respectively.And compared with gold standard,RSMD of them were 0.72 years of age,0.73 years of age,1.38 years of age and 1.11 years of age.And MAD and RSMD of physician were 0.58 years of age and 0.74 years of age,respectively.In artificial intelligence,RUS-CHN method was nearest to gold standard,and the differences of that between them were significant(F=0.01,F=0.06,F=5.58,F=0.34,F=0.31,P<0.05).The speed of testing and evaluation of the artificial intelligence software was 25.99±13.88s,which was significantly higher than that of physician(t=24.312,P<0.05).Conclusion:The RUS-CHN metacarpophalangeal bone age method of artificial intelligence not only can’t reduce accuracy of diagnosis,but also can significantly increase the efficiency of clinical practice.

关 键 词：人工智能 骨龄 RUS-CHN掌指骨骨龄法 诊断准确性 

分 类 号：R179[医药卫生—妇幼卫生保健]