Categorical-Data-Based BCI with Motor Imagery Using Equivalent Current Dipole Source Localization Pages 1-12

Toshimasa Yamazaki1, Kazufumi Tanaka2, Takahiro Shibata3, Hiromi Yamaguchi1, Minami Ouda4, Yukari Sasaguri1, Shun Hirose1, Hiroshi Takayanagi5, Hideyuki Maki6, Takahiro Yamanoi7 and Ken-Ichi Kamijo8

1Department of Bioscience and Bioinformatics, Kyushu Institute of Technology, 680-4 Kawazu, Iizuka, Fukuoka 820-8502, Japan; 2CAROLsystem Ltd., Tokyo 150-0036, Japan; 3Olympus Software Technology Corporation, Tokyo 163-1414, Japan; 4Software Vision Company, Kumamoto 862-0926, Japan; 5Future University Hakodate, Tokyo 101-0021, Japan; 6Japan Technical Software Co., Ltd, Sapporo 001-0021, Japan; 7Hokkai Gakuen University, Sapporo 064-0926 8NEC Corporation, Tokyo 108-8558, Japan

DOI: http://dx.doi.org/10.12970/2308-8354.2014.02.01

Download PDF

Abstract: Purpose: In order to enable data reduction in single-trial-electroencephalograms (EEGs)-based Brain-Computer Interfaces (BCIs), a new algorithm based on categorical data is proposed.

Method: The EEGs were categorized by independent component analysis (ICA) and equivalent current dipole source localization (ECDL), and Hayashi’s second method of quantification (H2MQ) was applied to the categorical data. Ten healthy subjects performed left- and right-hand movement imagery tasks, while EEGs were recorded from 32 electrodes on the scalp. Using the categorical data with respect to the brain sites where dipoles were located by ECDL after ICA, a learning model for the discrimination between the left- and right-hand imageries was conducted by H2MQ.

Results: Using 16 ICs, all the H2MQ models had the correlation ratios of more than 0.90, where the numbers of trials and categories were 60 and 4, respectively. The accuracy averages across all the subjects for the left- and right-hand imageries in each 10-trial validation phase were 94.5 % and 91.5 %, respectively, which was better than the previous common-spatial-pattern (CSP)-based BCIs.

Conclusion: The above results led us to a new paradigm for single-trial-EEG-based BCIs, quite different from continuous-valued EEGs and their spectral analysis, yielding the data reduction.

Keywords: Brain-Computer Interface, independent component analysis, equivalent current dipole source localization, movement imagery. Read more