不同人工智能骨龄评测软件的阅片结果比较研究  

Comparative study on the reading results of different artificial intelligence bone age evaluation software

作　　者：乐奖 范小芳 李少维 韩梅敏 何金水 Le Jiang;Fan Xiaofang;Li Shaowei;Han Meimin;He Jinshui(Department of Imaging,Huli District Maternal and Child Health Hospital,Fujian 361001,China;不详)

机构地区：[1]福建省厦门市湖里区妇幼保健院影像科,361001 [2]福建省厦门市湖里区妇幼保健院儿保科,361001 [3]福建医科大学附属漳州市医院儿科

出　　处：《实用医学影像杂志》2025年第1期5-8,共4页Journal of Practical Medical Imaging

基　　金：漳州市医院博士工作室攀登课题(PDA202105)。

摘　　要：目的 通过比较不同的自动骨龄评测软件的阅片结果,分别观察其临床准确性。方法 选择厦门某妇幼保健院的145张不同年龄段的儿童左手腕部直接数字化X线摄影(DR)平片(男性93张,女性52张),统一使用TW3法做为骨龄计算方法,分别使用自主建立的人工智能骨龄评测(AIBAA)-TW3算法模型(以下简称AIBAA-TW3法)、联影骨龄软件(以下简称联影TW3法)、医准骨龄软件(以下简称医准TW3法)、1名资深影像科医师(参考骨龄)和1名经过阅片培训的儿科初级医师(普通医生)进行骨龄评测,比较结果的平均绝对误差(MAE)和准确率。结果 145张不同年龄段的儿童左手腕部DR平片测评结果显示,AIBAA-TW3算法模型的预测骨龄与金标准参考骨龄的平均绝对误差为0.84岁,均高于联影、医准和普通医生,差异具有统计学意义(P<0.05)。当误差范围在±2.0岁、±1.0岁、±0.7岁及±0.5岁以内,AIBAA-TW3算法模型预测骨龄准确率分别为91.7%、64.8%、51.0%和40.7%,与联影、医准和普通医师之间的准确率差异均无统计学意义(P>0.05)。结论AIBAA-TW3算法模型的骨龄预测误差高于普通医生、医准和联影骨龄评测系统,但在预测准确率方面,AIBAATW3算法模型与普通医生和不同骨龄评测系统之间差异无统计学意义。Objective To compare the reading results of different automatic bone age assessment software and observe their clinical accuracy separately.Methods A total of 145 plain DR films(93 for men and 52 for women)of the children′s left wrist of different ages from a maternal and child health care hospital in Xiamen were selected,and TW3 was used as the bone age calculation method,artificial intelligence bone age assessment(AIBAA)-TW3 algorithm model(hereinafter referred to as Xiamen TW3 method),Lianying bone age software(hereinafter referred to as Lianying TW3 method),medical quasi-bone age software(hereinafter referred to as medical quasi-TW3 method)were applied respectively,One senior radiologist(referring to bone age)and one pediatric junior physician trained in reading(general practitioner)performed bone age assessment and compared the mean absolute error(MAE)and accuracy of the results.Results The results of 145 plain DR radiographs of the left wrist of children of different ages showed that the average absolute error between the predicted bone age and the gold standard reference bone age of the AIBAA-TW3 algorithm model was 0.84 years old,which was higher than that of lianying,medical standard and general doctor.The difference was statistically significant(P<0.05).When the error range was within±2.0 years old,±1.0 years old,±0.7 years old,and±0.5 years old,the accuracy of the AIBAA-TW3 algorithm model in predicting bone age was 91.7%,64.8%,51.0%,and 40.7%,respectively,there were no significant differrence between lianying,medical standard,and general physicians.Conclusion The bone age prediction error of the AIBAA-TW3 algorithm model is higher than that of the general doctor,medical standard,and continuous bone age evaluation system.Still,there is no obvious difference between the AIBAA-TW3 algorithm model and the general doctor and different bone age evaluation systems in terms of prediction accuracy.

关 键 词：骨骼年龄测定 软件 算法 人工智能 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术] R318[医药卫生—生物医学工程]