法医学杂志

法医学杂志 ›› 2014, Vol. 30 ›› Issue (6): 422-426.DOI: 10.3969/j.issn.1004-5619.2014.06.005

• 论著 • 上一篇    下一篇

基于支持向量机实现骨骺发育分级的自动化评估

王亚辉1，王子慎2，魏  华1，3，万  雷1，应充亮1，朱广友1   

  1. （1. 司法部司法鉴定科学技术研究所 上海市法医学重点实验室，上海 200063； 2. 合肥市公安局刑事警察支队，安徽 合肥 230001； 3. 华东政法大学研究生教育学院，上海 200042）
  • 发布日期:2014-12-25 出版日期:2014-12-28
  • 通讯作者: 朱广友，男，研究员，硕士研究生导师，主要从事法医临床学研究；E-mail：zhugy@ssfjd.cn
  • 作者简介:王亚辉（1982—），男，陕西榆林人，硕士，助理研究员，主要从事法医临床学科研和鉴定；E-mail：wangyh@ssfjd.cn
  • 基金资助:

    国家自然科学基金青年科学基金资助项目（811023 05）；“十二五”国家科技支撑计划项目（2012BAK16B01-3）；上海市法医学重点实验室资助项目（14DZ2270800）

Automated Assessment of Developmental Levels of Epiphysis by Support Vector Machine

WANG YA-HUI1, WANG ZI-SHEN2, WEI HUA1,3, WAN LEI1, YING CHONG-LIANG1, ZHU GUANG-YOU1   

  1. （1. Shanghai Key Laboratory of Forensic Medicine, Institute of Forensic Science, Ministry of Justice, P.R.China, Shanghai 200063, China; 2. Criminal Police Detachment of Hefei Public Security Bureau, Hefei 230001, China; 3. Graduate School of East China University of Political Science and Law, Shanghai 200042, China）
  • Online:2014-12-25 Published:2014-12-28

被引次数

6

摘要： 目的 运用支持向量机（support vector machine，SVM）实现尺、桡骨远端骨骺发育分级的自动化评估。 方法 收集我国140例11~19周岁青少年左侧腕关节X线正位片作为训练样本。将尺、桡骨远端骨骺分为五个发育分级，每个分级均包含28例样本。另选35例作为独立校验样本。建立尺、桡骨远端骨骺五个发育分级的SVM分类模型，用留一交叉验证法（leave one out cross validation，LOOCV）进行模型内部交叉验证以及梯度方向直方图（histogram of oriented gradient，HOG）进行模型外部验证，分别计算其准确率（PA）。 结果 桡骨远端骨骺分级SVM建模、LOOCV和HOG的PA分别为100.0%、78.6%和82.8%。尺骨远端骨骺分级SVM建模、LOOCV和HOG的PA分别为100.0%、80.0%和88.6%。 结论 运用SVM建立的尺、桡骨远端骨骺发育分级的自动化模型具有一定的可行性，为法医学骨龄评估软件的开发奠定基础。

关键词: 法医人类学, 骨骺, 尺骨, 桡骨, 支持向量机

Abstract: Objective To realize the automated assessment of the levels of epiphysis of distal radius and ulna by support vector machine （SVM）. Methods The X-ray films of the left wrist joints were taken from 140 teenagers aged from 11 to 19 years old as training samples. The levels of epiphysis of distal radius and ulna were divided into five developmental levels. Each level contained 28 samples. Another 35 cases were selected as independent verifying samples. SVM classification models of the five developmental levels of epiphysis of distal radius and ulna were established. The internal cross validation was made by leave one out cross validation （LOOCV）, while the external validation was made by histogram of oriented gradient （HOG）, and then the accuracy （PA） of testing results was calculated, respectively. Results The PA of SVM, LOOCV and HOG of distal radius epiphyseal level were 100%, 78.6%, and 82.8%, respectively; whereas the PA of SVM, LOOCV and HOG of distal ulna epiphyseal level were 100.0%, 80.0% and 88.6%, respectively. Conclusion The SVM-based automatic models of the growth stage of distal radius and ulna appear to have certain feasibility, and may provide a foundation for software development of bone age assessment by forensic medicine.

Key words: forensic anthropology, epiphyses, ulna, radius, support vector machine

中图分类号: