基于机器学习和锁骨胸骨端薄层CT的年龄推断研究  

Age estimation based on machine learning and thin-layer CT of sternal end of clavicle

作　　者：孙雨潇 王欣怡 热发提江·克然木 徐震 倪海元 占梦军 邓振华[1] Sun Yuxiao;Wang Xinyi;Refatijiang·Keranmu;Xu Zhen;Ni Haiyuan;Zhan Mengjun;Deng Zhenhua(West China School of Basic Medical Sciences&Forensic Medicine,Sichuan University,Chengdu,Sichuan 610041,China)

机构地区：[1]四川大学华西基础医学与法医学院,四川成都610041

出　　处：《中国法医学杂志》2023年第6期623-627,632,共6页Chinese Journal of Forensic Medicine

基　　金：国家自然科学基金面上资助项目(81971801);四川大学专职博士后研发基金项目(2023SCU12037);四川大学大学生创新创业训练计划资助项目(C2023123066)。

摘　　要：目的 采用Kellinghaus分级法对锁骨胸骨端薄层CT进行人工阅片分级,运用多种传统统计学方法以及机器学习方法构建青少年及成人早期年龄推断模型,探索机器学习技术在四川汉族人群年龄推断研究中的应用价值。方法 回顾性收集491例10~30岁个体的胸部薄层CT影像,参照Kellinghaus分级法对所收集样本进行阅片分级赋分。随机选取10%的数据作为测试集,其余数据作为训练集,综合构建多种青少年及成人早期年龄推断的传统统计学回归模型与机器学习模型,采用平均绝对误差值(Mean Absolute Error,MAE)对模型的性能进行评估。结果统计回归模型中效能最好的模型为三次回归模型,男性MAE值为1.34,女性MAE值为1.57;三种机器学习模型中,随机森林模型对男性的预测效能最好,MAE值为1.39;支持向量模型对女性的预测效能最好,MAE值为1.51。结论在锁骨胸骨端年龄推断模型的构建中,机器学习模型在年龄推断中的准确性有一定提升,但与传统统计学回归模型相比并无明显优势,机器学习方法在锁骨胸骨端年龄推断中的应用价值仍有待进一步探索研究。Objective The Kellinghaus grading method was used to manually read and grade the thin-layer CT of sternal end of clavicle,and a variety of traditional statistical methods as well as machine learning methods were used to construct age estimation models for adolescents and adults in early adulthood,to explore the value of the application of machine learning technology in the study of age estimation of the Han Chinese population in Sichuan.Methods Thinsection CT images of the chest were retrospectively collected from 491 individuals aged 10~30 years,and the collected samples were assigned a reading grade with reference to the Kellinghaus grading method.10%of the xases were randomly selected as the test set,and the remaining data were used as the training set to construct a variety of traditional statistical regression models and machine learning models for estimating the age of adolescents and adults in early adulthood,and the performance of the models was evaluated by using the mean absolute error(MAE).Results The statistical regression model with the best efficacy was the cubic regression model,with an MAE value of 1.34 for males and 1.57 for females;of the three machine learning models,the Random Forest model had the best predictive efficacy for males,with an MAE value of 1.39,and the Support Vector model had the best predictive efficacy for females,with an MAE value of 1.51.Conclusion In the construction of age estimation models for sternal end of clavicle,the machine learning model has a certain improvement in the accuracy of age prediction,but there is no obvious advantage compared with the traditional statistical regression model,and the use of the machine learning method in age estimation based on sternal end of clavicle still needs further exploration.

关 键 词：法医影像学 年龄推断 机器学习 锁骨胸骨端 

分 类 号：D919.6[医药卫生—法医学]