法医学杂志

法医学杂志 ›› 2019, Vol. 35 ›› Issue (2): 194-199.DOI: 10.12116/j.issn.1004-5619.2019.02.012

• 论著 • 上一篇    下一篇

基于主成分分析和支持向量机实现膝关节骨龄评估回归算法

雷义洋1， 申玉姝2， 王亚辉3， 赵虎1   

  1. 1. 中山大学中山医学院法医学系，广东 广州 510080； 2. 上海大学材料科学与工程学院高分子材料系，上海 200444； 3. 司法鉴定科学研究院 上海市法医学重点实验室 上海市司法鉴定专业技术服务平台，上海 200063
  • 发布日期:2019-04-25 出版日期:2019-04-28
  • 通讯作者: 王亚辉，男，副研究员，主要从事法医临床学和法医人类学研究；E-mail：wangyh@ssfjd.cn 赵虎，男，博士，教授，博士研究生导师，主要从事法医精神病学与神经科学研究；E-mail：zhaohu3@mail.sysu.edu.cn
  • 作者简介:雷义洋（1993—），男，硕士研究生，主要从事法医临床学研究；E-mail：leiyiyang2018@163.com
  • 基金资助:
    国家自然科学基金面上资助项目（81571859，81273350，81471829）；上海市法医学重点实验室资助项目（17DZ2273200）；上海市司法鉴定专业技术服务平台资助项目（16DZ2290900）；上海市法医学重点实验室（司法鉴定科学研究院）开放基金资助项目（KF1706）

Regression Algorithm of Bone Age Estimation of Knee-joint Based on Principal Component Analysis and Support Vector Machine

LEI Yi-yang1, SHEN Yu-shu2, WANG Ya-hui3, ZHAO Hu1   

  1. 1. Department of Forensic Medicine, Zhongshan School of Medicine, Sun Yat-sen University, Guangzhou 510080, China; 2. Department of Polymer Materials, School of Materials Science and Engineering, Shanghai University, Shanghai 200444, China; 3. Shanghai Key Laboratory of Forensic Medicine, Shanghai Forensic Service Platform, Academy of Forensic Science, Shanghai 200063, China
  • Online:2019-04-25 Published:2019-04-28

PDF (PC)

623

摘要： 目的 通过方向梯度直方图（histogram of oriented gradient，HOG）、局部二值模式（local binary patterns，LBP）、支持向量机（support vector machine，SVM）以及主成分分析（principal component analysis，PCA）等机器学习方法构建适用于我国维吾尔族青少年骨龄评估的回归算法模型。 方法 采集维吾尔族12.0~<19.0岁青少年的膝关节DR摄片图像，其中男性样本275例、女性样本225例，采用PCA法对提取的HOG与LBP特征图像进行降维，再以支持向量回归（support vector regression，SVR）算法构建膝关节骨龄评估算法模型。采用随机分层抽样法分别选取男性样本215例、女性样本180例作为SVR模型训练集，并用K折交叉验证法优化模型参数。剩余样本作为独立测试集，将模型预报年龄与样本真实年龄相比，统计误差范围分别在±0.8岁、±1.0岁的准确率，同时计算平均绝对误差（mean absolute error，MAE）与均方根误差（root mean square error，RMSE）。 结果 男性年龄误差范围在±0.8岁及±1.0岁的准确率分别为80.67%和89.33%，MAE为0.486岁，RMSE为0.606岁；女性年龄误差范围在±0.8岁及±1.0岁的准确率分别为80.19%和90.45%，MAE为0.485岁，RMSE为0.590岁。 结论 基于PCA与SVM对膝关节DR摄片图像HOG及LBP特征降维建立骨龄的预报模型，具有较高的准确性。

关键词: 法医人类学；年龄测定, 骨骼；膝关节；支持向量机；主成分分析；方向梯度直方图；局部二值模式；维吾尔族；青少年

Abstract: Objective To establish a regression algorithm model that applies to bone age estimation of Xinjiang Uygur adolescents with machine learning methods such as histogram of oriented gradient （HOG）, local binary patterns （LBP）, support vector machine （SVM）, principal component analysis （PCA）. Methods DR images of knee-joints from 275 male and 225 female subjects aged 12.0-<19.0 years old were collected, PCA method was used to reduce the dimensionality of the HOG and LBP features, and support vector regression （SVR） was used to establish a knee-joint bone age estimation algorithm model. Stratified random sampling method was used to select 215 male samples, 180 female samples for the model training set. K-fold cross validation method was used to optimize parameters of the model. The remaining samples as the independent test set was used to compare the sample’s true age and model estimated age, and had an accuracy rate in the statistical error range of ±0.8 and ±1.0 years, respectively. Then the mean absolute error （MAE） and root mean square error （RMSE） were calculated. Results The accuracy rate of male in the statistical error range of ±0.8 and ±1.0 year was 80.67%, 89.33%, respectively. The MAE and RMSE were 0.486 and 0.606 years, respectively. The accuracy rate of female in the statistical error range of ±0.8 and ±1.0 years was 80.19%, 90.45%, respectively. The MAE and RMSE were 0.485 and 0.590 years, respectively. Conclusion Establishment of prediction model for bone age estimation by feature dimension reduction of HOG and LBP in DR images of knee-joint based on PCA and SVM has relatively high accuracy.

Key words: forensic anthropology, age determination by skeleton, knee joint, support vector machine, principal component analysis, histogram of oriented gradient, local binary patterns, Uygur nationality, adolescent