中国儿童保健杂志 ›› 2023, Vol. 31 ›› Issue (6): 595-600.DOI: 10.11852/zgetbjzz2022-0691

• 科研论著 • 上一篇    下一篇

基于机器学习探索数字划消测验用于学习障碍快速筛查的研究

王洪安1, 禹东川1, 刘福临1, 池霞2,3   

  1. 1.东南大学生物科学与医学工程学院,江苏 南京 210096;
    2.南京医科大学儿科学院;
    3.南京医科大学附属妇产医院
  • 收稿日期:2022-06-02 修回日期:2022-09-26 发布日期:2023-06-02 出版日期:2023-06-10
  • 通讯作者: 禹东川,E-mail:dcyu@seu.edu.cn
  • 作者简介:王洪安(1995-),男,四川人,博士研究生,主要研究方向为学习障碍的早期筛查与诊断。
  • 基金资助:
    国家自然科学基金(62073077、61673113)

Feasibility study on rapid screening of learning disabilities by number cancellation test in combination with machine learning algorithm

WANG Hongan1, YU Dongchuan1, LIU Fulin1, CHI Xia2,3   

  1. 1. School of Biological Science and Medical Engineering, Southeast University, Nanjing, Jiangsu 210096, China;
    2. School of Pediatrics, Nanjing Medical University;
    3. Department of Child Health Care, Women's Hospital of Nanjing Medical University, Nanjing Maternity and Child Health Care Hospital
  • Received:2022-06-02 Revised:2022-09-26 Online:2023-06-10 Published:2023-06-02
  • Contact: YU Dongchuan, E-mail:dcyu@seu.edu.cn

摘要: 目的 基于机器学习探索数字划消测验用于学习障碍(LD)快速筛查的可行性研究,为探索LD的快速筛查提供新的研究思路和工具。方法 利用数字划消测验,2020年9月—2021年3月对南京市区随机选取的414名7~12岁儿童进行调查,首先获得数字划消测验的18个观察指标;将这些观察指标作为特征,探索利用7种经典的机器学习算法(包括线性判别分析、朴素贝叶斯、K近邻、神经网络、支持向量机、决策树和随机森林),实现对学习障碍3种亚型的分类;比较不同机器学习算法的分类性能,以从中选出推荐算法。结果 在7种经典的机器学习算法中,随机森林的优势突出,其准确率和曲线下面积(AUC)值分别达0.83和0.92。结论 借助随机森林算法,利用数字划消测验(只需2min就可完成)可以实现对学习障碍的快速筛查,进一步证实机器学习应用于疾病预测的可行性。

关键词: 学习障碍, 机器学习, 快速筛查, 数字划消, 学龄儿童

Abstract: Objective To explore the feasibility of a rapid screening method using number cancellation test (NCT) in combination with machine learning algorithm for learning disabilities (LD), in order to provide ideas for rapid screening of LD. Methods A total of 414 children aged 7-12 years were randomly recruited in Nanjing, and were asked to complete the NCT, in which 18 parameters were measured to evaluate individual's abilities during NCT. Then, these parameters were considered as classification features to explore the classification of LD by seven machine learning algorithms, including linear discriminant analysis, naive Bayes, K-nearest neighbor, neural network, support vector machine, decision tree and random forest. Finally, the classification performance among different machine learning algorithms was compared. Result Among the seven machine learning algorithms, random forest was outstanding for screening LD, and the accuracy and AUC were 0.83 and 0.92, respectively. Conclusion NCT in combination with random forest algorithm can be applied as a rapid screening method (completed within 2 minutes) for LD,which further proves the feasibility of machine learing in disease prediction.

Key words: learning disabilities, machine learning, rapid screening, number cancellation test, school-age children

中图分类号: