深度学习模型在儿童TW3法骨龄评估的临床效能分析.pdfVIP

浙江临床医学2023年1 25 月第25卷第1期 临床研究 深度学习模型在儿童TW3法骨龄评估的临床效能分析 程梦 钱琦 高铖 田曼曼 励杨晟 林敏* 目的探讨深度学习模型在儿童应用TW3法进行骨龄评估的临床效能。方法 回顾性收集180例儿童左手X线片。评估基于 【摘要】 Tanner-WhitehouseⅢI（TW3） 法的掌指骨和腕骨骨龄。以3位高年资主任医师的评估均值为金标准组，计算并比较深度学习模型（模型 组）及2位低年资医师（医师组，分别记作医师1、医师2）与金标准组评估时间差异，掌指骨骨龄、腕骨骨龄的均方误差（MSE）及平均绝 对误差（MAE）；采用组内相关系数（ICC）分析模型组和医师组与金标准组结果一致性。结果骨龄评估所用时间模型组明显少于医师组 （P0.05）。掌指骨骨龄评估，模型组和医师1、医师2与金标准组相比MSE、MAE差异有统计学意义（P0.05）。腕骨骨龄评估，模型组和 医师1、医师2与金标准组相比MSE、MAE差异有统计学意义（P0.05）。掌指骨骨龄评估，模型组与金标准组ICC为0.988，医师1、医师2与 金标准组的ICC分别为0.986、0.977。腕骨骨龄评估，模型组与金标准组ICC为0.971，医师1、医师2与金标准组ICC分别为0.970、0.953。结 论对于儿童，应用TW3法深度学习模型在评估掌指骨和腕骨骨龄有临床价值。 【关键词】骨龄 TW3法儿童 深度学习 [ Abstract Objective To explore the clinical eficacy of deep learning model in bone age assessment(BAA) of children using TW3 method, based on artificial intelligence(AI) system. Methods Totally 180 left hand and wrist radiographs of children aged 5~13 years in Zhejiang Province were retrospectively studied. Tanner-Whitehouse I (TW3) method based on radius, ulna and short bone(RUS) and carpal bone ages were assessed. The average value of bone age estimated by three chief radiologists was taken as reference standard group, two junior radiologists(denoted as doctor 1,2) as doctor group, and AI system as model group. The BAA time, mean square error(MSE) and mean absolute error(MAE) of BAA relative to reference standard group were calculated and compared among three groups. The intraclass correlation coeficient(ICC) was used to compare the consistency of the results among the model group and the doctor group relative to reference standard group. Results BAA time of the model group was significantly shorter than that of the doctor group(P0.05). There were significant differences of TW3 RUSs BAA MSE and MAE between the model group and doctor 1,2(all P0.05). There were significa