北京生物医学工程

一种个性化动态脑功能网络的构建与特征提取方法

A personalized dynamic brain functional network and feature extraction

作者：张娜孙炎珺李明爱

单位：北京工业大学信息学部(北京 100124)；计算智能与智能系统北京市重点实验室(北京 100124）

关键词：运动想象脑电信号; 动态脑功能网络; 特征提取; 个性化皮尔逊相关系数; 共空间模式

分类号：R318.04

出版年·卷·期（页码）：2020·39·6（551-560）

摘要：

目的为了研究运动想象过程中脑功能网络(Brain Functional Network, BFN)的时频变化特征及对运动想象任务识别的影响，本文提出一种个性化皮尔逊相关系数(Personalized Pearson Correlation Coefficient, PPCC)及动态BFN的构建与特征提取方法。方法首先，对各受试者运动想象脑电（Motor Imagery EEG, MI-EEG）频带范围进行两级筛选，获得其最优频带；然后，将运动想象时间段进行分割，计算各子时段最优频带MI-EEG的PPCC，并用于构建个性化的动态BFN；进而，计算各个BFN的度作为网络特征，并将多时段的网络特征串行融合获得特征向量；最后，针对BCI Competition III Data Set IIIa和BCI Competition IV Data Set 2a数据集，采用支持向量机检验特征的有效性。结果在两个公共数据集上，本文方法的10×10折交叉验证最高识别率分别为100.00%和68.84%。与基于共空间模式和基于PCC的BFN特征提取方法相比，具有明显的优势，双样本t检验的结果也充分表明了PPCC的优越性。结论与PCC相比，基于PPCC能构建出可客观地展现运动想象个性化特点的动态BFN，反映了不同受试者运动想象时大脑激活的差异性，及其在时域和频域同时呈现的动态变化特点，有效增强了特征提取的自适应性。

Objective In order to study the time-frequency variation characteristics of Brain Functional Network (BFN) in the process of motor imagery (MI) and the impact on the recognition of MI tasks, this paper proposes a Personalized Pearson Correlation Coefficient (PPCC) and dynamic BFN and feature extraction method. Methods First, two-stage screening is performed on each subject's Motor Imagery EEG (MI-EEG) frequency band range to obtain the optimum frequency band; Then，the MI time period is segmented into several sub-periods，the PPCC of the optimum frequency band MI-EEG is calculated for each sub-period, and personalized dynamic BFNs are constructed based on the PPCC values; Furthermore, the degree of each BFN is calculated as network features, which are serially fused to obtain feature vectors；Finally，the feature vectors are evaluated by support vector machine on the BCI Competition III Data Set IIIa and BCI Competition IV Data Set 2a. Results On the two public datasets, the 10×10 fold highest recognition rates achieve 100.00% and 68.84%, which have obvious advantages compared with common spatial pattern-based and PCC-based BFN feature extraction methods. The two-sample t-test clearly shows its superiority as well. Conclusions Compared with PCC, the dynamic BFNs which objectively express the personalized characteristic of MI-EEG can be constructed based on PPCC, reflect the extinctions of activated brain areas caused by MI among different subjects as well as the characteristics of dynamic changes exhibited in time and frequency domains simultaneously, and effectively enhance the adaptivity of feature extraction.

参考文献：

[1] 李明爱, 罗新勇, 崔燕,等. 基于MI-BCI的上肢在线运动康复原型系统[J]. 北京生物医学工程, 2017, 36(3)：273-278.
Li MA，Luo XY，Cui Y, et al. MI-BCI based online prototype system for upper limb rehabilitation[J]. Beijing Biomedical Engineering，2017，36(3)：273-278.
[2] Nicolas-Alonso LF, Corralejo R, Gomez-Pilar J, et al. Adaptive stacked generalization for multiclass motor imagery-based brain computer interfaces[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2015, 23(4): 702-712.
[3] Bassett DS,  Bullmore E. Small-world brain networks[J]. Neuroscientist, 2006, 12(6): 512-523.
[4] 李明爱, 南琳, 孙炎珺. 基于锁相值和图论的脑功能网络特征提取方法[J]. 北京生物医学工程, 2019, 38(1): 15-21.
Li MA，Nan L，Sun YJ．Applying phase locking value and graph theory for feature extraction of brain functional network[J]. Beijing Biomedical Engineering，2019, 38(1): 15-21.
[5] Falzon O, Camilleri KP, Muscat J. Complex-valued spatial filters for task discrimination[J]. Annual International Conference of the IEEE Engineering in Medicine and Biology Society , 2010,2010: 4707-4710.
[6] Zarei A, Ghassemi F, Moradi MH. Information based source number estimation for probabilistic common spatial pattern in motor imagery BCI system[C]//2016 24th Iranian Conference on Electrical Engineering (ICEE). Shiraz, Iran: IEEE Press, 2016: 555-560.
[7] Jin J, Miao YY, Daly I, et al. Correlation-based channel selection and regularized feature optimization for MI-based BCI[J]. Neural Networks, 2019, 118: 262-270.
[8] Stefano Filho CA, Attux R, Castellano G. EEG sensorimotor rhythms’ variation and functional connectivity measures during motor imagery: linear relations and classification approaches[J]. PeerJ, 2017, 5: e3983.
[9] Petti M, Pichiorri F, Toppi J, et al. Bci-assisted training for upper limb motor rehabilitation: estimation of effects on individual brain connectivity and motor functions[C]//Proceedings of the 7th Graz Brain-Computer Interface Conference 2017. Graz, Austria, IEEE Press, 2017: 406-409.
[10] Kong W, Guo X, Zhao X, et al. Spectral analysis of brain function network for the classification of motor imagery tasks[C]//2011 4th International Conference on Biomedical Engineering and Informatics (BMEI). Shanghai: IEEE Press, 2011, 2: 850-853.
[11] Li J, Wang Y, Zhang L, et al. Decoding EEG in cognitive tasks with time-frequency and connectivity masks[J]. IEEE Transactions on Cognitive and Developmental Systems, 2016, 8(4): 298-308.
[12] Bono V, Biswas D, Das S, et al. Classifying human emotional states using wireless EEG based ERP and functional connectivity measures[C]//2016 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI). Las Vegas, NV, USA: IEEE Press, 2016: 200-203.
[13] Rodrigues PG, Stefano Filho CA, Attux R, et al. Space-time recurrences for functional connectivity evaluation and feature extraction in motor imagery brain-computer interfaces[J]. Medical & Biological Engineering & Computing, 2019, 57(8): 1709-1725.
[14] Zhang R, Li X, Wang Y, et al. Using brain network features to increase the classification accuracy of MI-BCI inefficiency subject[J]. IEEE Access, 2019, 7: 74490-74499.
[15] Stefano Filho CA, Attux R, Castellano G. Can graph metrics be used for EEG-BCIs based on hand motor imagery?[J]. Biomedical Signal Processing and Control, 2018, 40: 359-365.
[16] Rubinov M, Sporns O. Complex network measures of brain connectivity: uses and interpretations[J]. Neuroimage, 2010, 52(3): 1059-1069.

服务与反馈：

【文章下载】【加入收藏】

提示：您还未登录，请登录！点此登录