基于膝关节MRI T_(1)WI深度学习模型的构建和活体年龄的推断

DEEP LEARNING MODELING USING T_(1)-WEIGHTED IMAGES IN MAGNETIC RESONANCE IMAGING OF THE KNEE JOINTS AND ITS USE IN AGE ESTIMATION OF LIVING BODIES

目的探讨基于膝关节MRI T_(1)加权成像(T_(1)WI)深度学习模型的构建方法,并应用该模型推断青少年的年龄。方法收集2015年1月—2021年12月青岛大学附属医院1212例(内部数据集)及青岛市市立医院341例(外部数据集)10~18岁男性膝关节MRI T_(1)WI图像,经过对股骨远端和胫骨近端骨骺骺板进行标记和图像分割后,采用随机数字表法将内部数据集各年龄组按照8∶2分为训练组(971例)和验证组(241例)用于模型的建立,外部数据集(测试组)用于模型的评价。通过准确率、精准率、召回率、灵敏度、特异度等指标来测试和验证模型的性能。结果验证组的准确率为85.713%,精准率为84.732%,召回率为85.713%,特异度为97.729%,灵敏度为85.713%;而测试组的准确率为82.578%,精准率为83.145%,召回率为82.578%,特异度为97.442%,灵敏度为82.578%,验证集和测试组的各项指标比较差异均无显著性(P>0.05)。结论本研究成功建立了基于膝关节MRI T_(1)WI的深度学习模型,可应用于10~18岁青少年年龄的推断。

Objective To discuss deep learning modeling using T_(1)-weighted images(T_(1)WI)in magnetic resonance imaging(MRI)of the knee joints and its use in age estimation of adolescents.Methods The T_(1)WI of the knee joints were collected from 1212 male patients aged 10-18 years who were admitted to The Affiliated Hospital of Qingdao University from January 2015 to December 2021(internal data set)and 341 male patients of the same ages who were admitted to Qingdao Municipal Hospital during the same period(external data set).After labeling and image segmentation of the epiphyseal plates of the distal femurs and proximal tibiae,the internal data set was divided into training group(971 cases)and validation group(241 cases)at a ratio of 8∶2 according to their age groups using a random number table for modeling,and the external data set(test group)was used for model evaluation.The performance of the model was tested and validated based on accuracy,precision,recall rate,sensitivity,and specificity.Results The accuracy,precision,recall rate,specificity,and sensitivity of the validation group were 85.713%,84.732%,85.713%,97.729%,and 85.713%,respectively;the same indicators of the test group were 82.578%,83.145%,82.578%,97.442%,and 82.578%,respectively.There were no significant differences in the above indicators between the validation group and the test group(P>0.05).Conclusion A deep learning model based on the T_(1)WI of the knee joints is successfully constructed,and it can be used for age estimation of adolescents aged 10-18 years.