基于动态功能性脑网络的情感分析-《计算机技术与发展》

文章信息/Info

作者:: 黄义华¹; 4 ; 童玥2 ; 衡霞²; 3 ; 卢忱¹; 4 ; 王忠民²; 3; 1. 中兴通讯股份有限公司企业发展部,广东深圳 518057;
2. 西安邮电大学计算机学院,陕西西安 710121;
3. 西安邮电大学陕西省网络数据分析与智能处理重点实验室,陕西西安 710121;
4. 移动网络和移动多媒体技术国家重点实验室,广东深圳 518055

Author(s):: HUANG Yi-hua1; 4 ; TONG Yue2 ; HENG Xia2; 3 ; LU Chen1; 4 ; WANG Zhong-min2; 3; 1. Enterprise Development Department,ZTE Corporation,Shenzhen 518057,China;
2. School of Computer Science and Technology,Xi’an University of Posts and Telecommunications,Xi’an 710121,China;
3. Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing,Xi’an University of Posts and Telecommunications,Xi’an 710121,China;
4. State Key Laboratory of Mobile Network and Mobile Communication Multimedia Technology,Shenzhen 518055,China

Keywords:: brain functional network; Pearson 爷 s correlation coefficient; functional connectivity; singular value decomposition; brainstate segmentation

摘要:: 人脑活动是在秒级与毫秒级动态变化的,因此采用静态连接方式构建的功能性脑网络,会造成部分与时间相关有效特征的缺失。该文旨在研究情绪变化期间不同大脑区域之间相互作用的时空变化,提出了一个系统的分析框架。该框架包括相关性度量,脑状态分割,代表性时间片段提取以及动态网络构建和分析。首先,利用皮尔逊相关系数量化不同脑区之间的功能连通性。其次,计算两相邻时间点的相关性矩阵之间的奇异值分解(singular value decomposition,SVD) 矢量空间距离,确定情绪转换点并对非平稳脑状态进行时间片分割,提取代表性时间片段。最后,基于相关性和频带功率分布构建不同网络模式,利用滑动窗口法估计动态相关模式和动态功率分布变化,然后提取脑动力学的多变量特征并进行分类识别。在 SEED 数据集上进行的相关实验验证了基于动态功能连接的情感评估方法的可行性,为不同情绪状态下建立脑动态模型开辟了新的途径。

Abstract:: The human brain changes dynamically at the second and millisecond level,so the construction of functional brain network bystatic connection will cause the loss of some time-related effective features. The purpose of this paper is to study the temporal and spatialchanges of the interaction between different brain regions during the emotional change,and to propose a systematic analysis framework.The framework includes correlation measurement,brain state segmentation,representative time segment extraction and dynamic networkmeasurement. First of all, the functional connectivity between different brain regions is measured by correlation size. Secondly, thesingular value decomposition ( SVD) vector space distance between the correlation matrix of two adjacent time points is calculated,the emotional transition point is determined, and the time slice of non - stationary brain state is segmented to extract representative timesegments. Finally,different network modes are constructed based on correlation mode and power distribution in frequency band. Thedynamic correlation mode and power distribution change are estimated by sliding window method,and then the multivariable features ofnetwork level brain dynamics are extracted and classified. Relevant experiments on the SEED data set verify the feasibility of theemotional assessment method based on dynamic functional connection,and open up a new way for establishing brain dynamic modelsunder different emotional states.