Please wait a minute...
Tsinghua Science and Technology  2021, Vol. 26 Issue (4): 440-451    doi: 10.26599/TST.2019.9010079
    
Exploiting Sparse Representation in the P300 Speller Paradigm
Hongma Liu*(),Yali Li(),Shengjin Wang()
Department of Electronic Engineering, Tsinghua University, Beijing 100084, China.
Download: PDF (2812 KB)      HTML
Export: BibTeX | EndNote (RIS)      

Abstract  

A Brain-Computer Interface (BCI) aims to produce a new way for people to communicate with computers. Brain signal classification is a challenging issue owing to the high-dimensional data and low Signal-to-Noise Ratio (SNR). In this paper, a novel method is proposed to cope with this problem through sparse representation for the P300 speller paradigm. This work is distinguished using two key contributions. First, we investigate sparse coding and its feasibility for brain signal classification. Training signals are used to learn the dictionaries and test signals are classified according to their sparse representation and reconstruction errors. Second, sample selection and a channel-aware dictionary are proposed to reduce the effect of noise, which can improve performance and enhance the computing efficiency simultaneously. A novel classification method from the sample set perspective is proposed to exploit channel correlations. Specifically, the brain signal of each channel is classified jointly using its spatially neighboring channels and a novel weighted regulation strategy is proposed to overcome outliers in the group. Experimental results have demonstrated that our methods are highly effective. We achieve a state-of-the-art recognition rate of 72.5%, 88.5%, and 98.5% at 5, 10, and 15 epochs, respectively, on BCI Competition III Dataset II.



Key wordssparse representation      sample selection      channel-aware dictionary      P300 speller     
Received: 01 December 2019      Published: 12 January 2021
Fund:  National High Technology Research and Development (863) Program of China(2012AA011004);National Science and Technology Support Program(2013BAK02B04)
Corresponding Authors: Hongma Liu     E-mail: liuhongma.thu@hotmail.com;liyali13@tsinghua.edu.cn;wgsgj@tsinghua.edu.cn
About author: Hongma Liu received the BEng degree from Southeast University, Nanjing, China in 2012. He is currently pursuing the PhD degree at the Department of Electronic Engineering, Tsinghua University, Beijing, China. His current research interests include pattern recognition, computer vision, BCIs, and object recognition.|Yali Li received the BEng degree with Excellent Graduates Award from Nanjing University, China in 2007 and the PhD degree from Tsinghua University, Beijing, China in 2013. Currently she is a research assistant in the Department of Electronic Engineering, Tsinghua University. Her research interests include image processing, pattern recognition, computer vision, and video analysis.|Shengjin Wang received the BEng degree from Tsinghua University, China in 1985 and the PhD degree from the Tokyo Institute of Technology, Tokyo, Japan in 1997. From May 1997 to August 2003, he was a member of the research staff in the Internet System Research Laboratories, NEC Corporation, Japan. Since September 2003, he has been a professor with the Department of Electronic Engineering, Tsinghua University, where he is currently also a director of the Research Institute of Image and Graphics. He has published more than 50 papers on image processing. He is the holder of ten patents. His current research interests include image processing, computer vision, video surveillance, and pattern recognition.
Cite this article:

Hongma Liu,Yali Li,Shengjin Wang. Exploiting Sparse Representation in the P300 Speller Paradigm. Tsinghua Science and Technology, 2021, 26(4): 440-451.

URL:

http://tst.tsinghuajournals.com/10.26599/TST.2019.9010079     OR     http://tst.tsinghuajournals.com/Y2021/V26/I4/440

Fig. 1 Typical electrode layout of a 64-channel EEG recorder.
Fig. 2 Averaged P300 responses over Channel 11 (Cz). P300 is a positive deflection in EEG approximately 300 ms after the stimulus.
Fig. 3 Character matrix used in the P300 speller paradigm with the third row highlighted.
Training dataset sizeTest dataset size
Subject AP30025503000
No P30012 75015 000
Subject BP30025503000
No P30012 75015 000
Table 1 BCI Competition III Dataset II size.
Fig. 4 Framework of sparse representation used in the P300 speller paradigm. Training samples are used to build the dictionary after sample selection. Test samples are then sparse represented and classified according to the sparse code and reconstruction error.
Fig. 5 Scores of all 15 300 training samples plotted against the sample number.
EpochWithout sample selectionDictionary sizeRand 2000
1000150020002500
547.545.049.050.548.546.0
1581.073.085.588.585.077.5
Table 2 Mean accuracy of two subjects using different dictionary sizes for sample selection. (%)
ChannelEpoch
1510131415
Cz3.320.028.345.048.345.0
Tp75.01.71.71.73.38.3
Table 3 Recognition rate of Channels Cz and Tp7. (%)
Fig. 6 (a) Performance for using each channel independently for Subject A. (b) Corresponding accuracy topography for Subject A. (c) Performance for using each channel independently for Subject B. (d) Corresponding accuracy topography for Subject B.
SubjectElectrode ranking
12345678
ACzFc4FczPzC1Fc2Fc1Cpz
BFc2C2Fc4CzCp2F1C1Cp1
Table 4 8 top-ranked channels for Subjects A and B.
MethodSubjectEpoch
123456789101112131415
BaselineA182328314345464858656668727876
B283139445252575964717679798586
Mean23.027.033.537.547.548.551.553.561.068.071.073.575.581.581.0
BL+SSA212536404350556368707173758786
B364149545861626670747576849191
Mean28.533.042.547.050.555.558.564.569.072.073.074.579.589.088.5
BL+SS+NGSCA232644576669727378858988909898
B424666707982828588929293959799
Mean32.536.055.063.572.575.577.079.083.088.590.590.592.597.598.5
Table 5 Recognition rate on the test dataset for the proposed methods. (%)
SubjectEpochMethod
LDA[26]ESVM[7]CNN[14]DFEDL[18]ADDL[19]LC-PDL[20]JP-LRDL[22]Ours
A54572615557624966
107883827981817685
158897979493969198
B57675777270776979
109291928486898192
159696948990929099
Mean560.573.569.063.563.569.559.072.5
1085.087.087.081.583.585.078.588.5
1592.096.595.591.591.594.090.598.5
Table 6 Recognition rates comparison with other methods in the literature. (%)
MethodTraining time (s)Test time (s)
ESVM[7]20.482.75
CNN[14]11683.49
LDA[26]14.271.38
DFEDL[18]45.501.17
ADDL[19]52.310.94
LC-PDL[20]20.240.66
JP-LRDL[22]189.002.45
Ours18.570.62
Table 7 Efficiency comparison with other methods.
[1]   Alonso L. F. N. and Gil J. G., BCIs, a review, Sensors, vol. 12, no. 2, pp. 1211-1279, 2012.
[2]   Bashashati A., Fatourechi M., Ward R. K., and Birch G. E., A survey of signal processing algorithms in BCIs based on electrical brain signals, Journal of Neural Engineering, vol. 4, no. 2, p. 32, 2007.
[3]   Polich J. and Kok A., Cognitive and biological determinants of P300: An integrative review, Biological Psychology, vol. 41, no. 2, pp. 103-146, 1995.
[4]   Luck S. J., An introduction to the event-related potential technique. Cambridge, MA, USA: MIT Press, 2014.
[5]   Blankertz B., Muller K. R., Curio G., Vaughan T. M., Schalk G., Wolpaw J. R., Schlogl A., Neuper C., Pfurtscheller G., Hinterberger T., et al., The BCI competition 2003: Progress and perspectives in detection and discrimination of EEG single trials, IEEE Transactions on Biomedical Engineering, vol. 51, no. 6, pp. 1044-1051, 2004.
[6]   Xu N., Gao X., Hong B., Miao X., Gao S., and Yang F., BCI competition 2003-dataset IIb: Enhancing P300 wave detection using ICA-based subspace projections for bci applications, IEEE Transactions on Biomedical Engineering, vol. 51, no. 6, pp. 1067-1072, 2004.
[7]   Rakotomamonjy A. and Guigue V., BCI competition III: Dataset II-ensemble of SVMs for BCI P300 speller, IEEE Transactions on Biomedical Engineering, vol. 55, no. 3, pp. 1147-1154, 2008.
[8]   Gerking J. M., Pfurtscheller G., and Flyvbjerg H., Designing optimal spatial filters for single-trial EEG classification in a movement task, Clinical Neurophysiology, vol. 110, no. 5, pp. 787-798, 1999.
[9]   Shi L., Li Y., Sun R., and Lu B., A sparse CSP algorithm for BCI, in Proc. of International Conference on Neural Information Processing, Belin, German, 2011, pp. 725-733.
[10]   Koles Z. J., Lazar M. S., and Zhou S. Z., Spatial patterns underlying population differences in the background EEG, Brain Topography, vol. 2, no. 4, pp. 275-284, 1990.
[11]   Kamp S. M., Murphy A. R., and Donchin E., The component structure of ERPs in the P300 speller paradigm, IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 21, no. 6, pp. 897-907, 2013.
[12]   Kaper M., Meinicke P., Grossekathoefer U., Lingner T., and Ritter H., BCI competition 2003-dataset LLb: SVMs for the P300 speller paradigm, IEEE Transactions on Biomedical Engineering, vol. 51, no. 6, pp. 1073-1076, 2004.
[13]   Burke D. P., Kelly S. P., Chazal P., Reilly R. B., and Finucane C., A parametric feature extraction and classification strategy for BCI, IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 13, no. 1, pp.12-17, 2005.
[14]   Cecotti H. and Graser A., CNNs for P300 detection with application to BCIs, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 33, no. 3, pp. 433-445, 2011.
[15]   Liu M., Wu W., Gu Z., Yu Z., Qi F. F., and Li Y. Q., Deep learning based on batch normalization for P300 signal detection, Neurocomputing, vol. 275, pp. 288-297, 2018.
[16]   Stober S., Sternin A., Owen A. M., and Grahn J. A., Deep feature learning for EEG recordings, arXiv preprint arXiv:1511.04306, 2015.
[17]   Shin Y., Lee S., Ahn M., Jun S. C., and Lee H. N., Motor imagery-based BCI classification via sparse representation of EEG signals, in Proc. of the 8th International Symposium on Noninvasive Functional Source Imaging of the Brain and Heart and the 8th International Conference on Bioelectromagnetism, Banff, Canada, 2011, pp. 93-97.
[18]   Li Z. M., Zhang Z., Qin J., Zhang Z., and Shao L., DFEDL algorithm for object recognition, .
doi: 10.1109/TNNLS.2019.2910146
[19]   Zhang Z., Jiang W. M., Qin J., Zhang L., Li F. Z., Zhang M., and Yan S., Jointly learning structured analysis discriminative dictionary and analysis multiclass classifier, IEEE Transactions on Neural Networks and Learning Systems, vol. 29, no. 8, pp. 3798-3814, 2017.
[20]   Zhang Z., Jiang W., and Zhang Z., Scalable block-diagonal locality-constrained projective dictionary learning, arXiv preprint arXiv:1905.10568, 2019.
[21]   Gu S., Zhang L., and Zuo W., Projective dictionary pair learning for pattern classification, Advances in Neural Information Processing Systems, Montreal, Canada, 2014, pp.793-801.
[22]   Foroughi H., Ray N., and Zhang H., Object classification with joint projection and low-rank dictionary learning, IEEE Transactions on Image Processing, vol. 27, no. 2, pp. 806-821, 2017.
[23]   Donoho D. L., Drori I., Tsaig Y., and Starck J. L., Sparse solution of underdetermined linear equations by stagewise orthogonal matching pursuit, IEEE Transactons on Information Theory, vol. 58, no. 2, pp. 1094-1121, 2012.
[24]   Wright J., Yang A. Y., Ganesh A., Sastry S. S., and Ma Y., Robust face recognition via sparse representation, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 31, no. 2, pp. 210-227, 2009.
[25]   Chi Y. T., Ali M., Rushdi M., and Ho J., Affine-constrained group sparse coding and its application to image-based classifications, in Proc. of IEEE Conference on Computer Vision and Pattern Recognition, Potland, OR, USA, 2013, pp. 681-688.
[26]   Liang N. and Bougrain L., Averaging techniques for single-trial analysis of oddball ERPs, in Proceeding of Fourth International BCI Workshop and Training Course, Graz, Australia, 2008, pp. 44-49.
[1] Liuqing Chen, Xiaofeng Zhang, Hongbing Ma. Sparse Representation over Shared Coefficients in Multispectral Pansharpening[J]. Tsinghua Science and Technology, 2018, 23(03): 315-322.
[2] . Robust Hierarchical Framework for Image Classification via Sparse Representation*[J]. Tsinghua Science and Technology, 2011, 16(1): 13-21.
[3] . Ways to Sparse Representation: A Comparative Study[J]. Tsinghua Science and Technology, 2009, 14(4): 434-443.