Discriminating three motor imagery states of the same joint for brain-computer interface

Abstract

The classification of electroencephalography (EEG) induced by the same joint is one of the major challenges for brain-computer interface (BCI) systems. In this paper, we propose a new framework, which includes two parts, feature extraction and classification. Based on local mean decomposition (LMD), cloud model, and common spatial pattern (CSP), a feature extraction method called LMD-CSP is proposed to extract distinguishable features. In order to improve the classification results multi-objective grey wolf optimization twin support vector machine (MOGWO-TWSVM) is applied to discriminate the extracted features. We evaluated the performance of the proposed framework on our laboratory data sets with three motor imagery (MI) tasks of the same joint (shoulder abduction, extension, and flexion), and the average classification accuracy was 91.27%. Further comparison with several widely used methods showed that the proposed method had better performance in feature extraction and pattern classification. Overall, this study can be used for developing high-performance BCI systems, enabling individuals to control external devices intuitively and naturally.

Keywords: Brain-computer interface; Cloud model; Common spatial pattern; Local mean decomposition; Motor imagery; Multi-objective grey wolf optimizer; Twin support vector machine.

Figure 1. Emotive Epoc+ and the placement position of Emotive electrodes.

(A) Emotive Epoc+ EEG signal acquisition instrument. (B) Location of all electrodes and marked electrodes are used in this study.

Figure 2. Experimental paradigm for the motor imagery (MI) tasks.

(A) The experiment included three MI tasks (shoulder abduction, flexion, and extension). (B) The experimental process of one trail.

Figure 3. Distributions of the best two features obtained by the proposed feature extraction method (LMD-CSP).

(A) Distribution of the best two features of abduction and extension. (B) Distribution of the best two features of abduction and flexion. (C) Distribution of the best two features of extension and flexion.

Figure 4. Average classification accuracies for the three categories.

Three motor imagery (MI) tasks classification accuracy of all subjects obtained using the proposed method (LMD-CSP and MOGWO-TWSVM).

Figure 5. The mean confusion matrix of all subjects (Abdu represents abduction, Ext represents extension, Flex represents flexion).

Figure 6. Comparison of classification accuracy obtained by the different feature extraction methods.

Five different feature extraction methods, including time domain parameters (TDP), common spatial pattern (CSP), filter-bank common spatial pattern (FBCSP), common spatial pattern based on empirical mode decomposition (EMD-CSP), and the proposed feature extraction method (LMD-CSP), were employed to extract motor imagery (MI) features on our data sets, respectively. Then, the same multi-objective grey wolf optimization twin support vector machine (MOGWO-TWSVM) classified those MI features to obtain classification accuracy of all subjects.

Figure 7. Averaged channel weight scores for each pair of motor imagery tasks (A/E represents abduction and extension, A/F represents abduction and flexion, and E/F represents extension and flexion).

Figure 8. Comparison of classification accuracy obtained by the different classifiers.

The proposed feature extraction method (LMD-CSP) was employed to extract (motor imagery) MI features on our data sets. Then the same features were classified by six different classifier, including linear discriminant analysis (LDA), extreme learning machine (ELM), k-nearest neighbors (KNN), least squares support vector machine (LS-SVM), and the proposed classifier (MOGWO-TWSVM), to obtain classification accuracy of all subjects.

Figure 9. Classification accuracy comparison by the proposed method with other recent methods.

The temporal filter parameter optimization with CSP (TFPO-CSP), the frequency-based deep learning scheme for recognizing brain wave signals (OPTICAL+), and the proposed method were employed to discriminate two-class EEG data (shoulder abduction and extension) in our data sets to obtain the classification accuracy of all subjects.