计算机工程与应用 ›› 2018, Vol. 54 ›› Issue (1): 42-47.DOI: 10.3778/j.issn.1002-8331.1608-0250

• 理论与研发 • 上一篇    下一篇

基于t-SNE的脑网络状态观测矩阵降维方法研究

董迎朝1,王  彬1,马洒洒1,刘  辉1,熊  新1,薛  洁2   

  1. 1.昆明理工大学 信息工程与自动化学院,昆明 650500
    2.云南警官学院 信息网络安全学院,昆明 650223
  • 出版日期:2018-01-01 发布日期:2018-01-15

Dimension reduction method research of brain network status observation matrix based on t-SNE

DONG Yingzhao1, WANG Bin1, MA Sasa1, LIU Hui1, XIONG Xin1, XUE Jie2   

  1. 1.Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunming 650500, China
    2.Faculty of Information Network Security, Yunnan Police Officer Academy, Kunming 650223, China
  • Online:2018-01-01 Published:2018-01-15

摘要: 针对基于功能核磁共振重构的脑网络状态观测矩阵维数过高和无特征的特点,对其降维方法展开研究,给出了基于t-SNE的脑网络状态观测矩阵降维算法,并且利用Python实现了降维及可视化平台。实验结果表明,与目前主流的其他降维算法相比较,使用该方法得到的脑网络状态观测矩阵低维空间的映射点有明显的聚类表现,并且在多个样本上的降维结果显现出一定的规律性,从而证明了该算法的有效性和普适性。

关键词: 高维数据降维, 脑功能网络, 脑网络状态观测矩阵, t-SNE算法

Abstract: The brain network state observation matrix based on fMRI reconstruction technology is in high dimension and characterless. A dimension reduction method based on t-distributed Stochastic Neighbor Embedding algorithm for this kind of matrix is presented and a platform for the dimension reduction and visualization is built with Python. The experimental results show that compared with popular dimension reduction methods, the low dimension embedding of brain network state observation matrix with this method demonstrates distinct clustering, and the dimension reduction results of different brain network state observation matrix show up some common regularity, which supports the validity and universality of this method.

Key words: high dimension reduction, functional brain network, brain network state observation matrix, t-SNE algorithm