A contralateral channel guided model for EEG based motor imagery classification. (March 2018)

• Record Type: Journal Article
• Authors: Sun, Lei; Feng, Zuren; Chen, Badong; Lu, Na
• Is Part Of: Biomedical signal processing and control. Volume 41(2018)
• Page Start: 1
• Page End: 9
• Publication Date: 2018-03
• Subjects: Brain computer interface -- Motor imagery -- EOG artifact; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2017.10.012
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A normalization model for EEG signal standardization is developed which could efficiently reduce the influence from EOG artifact. Contralateral EOG channel is demonstrated to be more effective for EOG artifact removal as compared with the ipsilateral EOG channel. Hjorth feature has been employed for motor imagery classification which has shown superior performance. Abstract: Objective: A novel and effective EOG correction method is proposed to improve the motor imagery (MI) classification performance. Methods: A new normalization model with one contralateral EOG channel is developed to retain the MI-related neural potentials and avoid the redundant influence among the EOG channels. By using the Hjorth features, the sub-optimal weights of our normalization model are learned for the MI classification of evaluation data. Results: The proposed method was applied on BCI Competition IV dataset 2b and 2a, and one dataset collected in our laboratory. As a result, the proposed method obtained an average kappa of 0.72 for the dataset 2b, 0.53 for the dataset 2a and 0.47 for the collected dataset. Conclusions: The proposed method could exclude interference among the EOG channels and the cross-interference between the EOG and EEG channel. The results proved that the EOG signal does have certain useful information for MI classification. The proposed method could emphasize ERD/ERS features, and improve MI classification performance. Significance: Compared to the regression method, the raw data based and the ipsilateral EOG channel based methods, the proposed method has significantly improved the MI classification performance. In addition, compared to other state-of-the-art methods, our approach also has obtained the best performance.