Exploration of computational methods for classification of movement intention during human voluntary movement from single trial EEG

Abstract

Objective: To explore effective combinations of computational methods for the prediction of movement intention preceding the production of self-paced right and left hand movements from single trial scalp electroencephalogram (EEG).

Methods: Twelve naïve subjects performed self-paced movements consisting of three key strokes with either hand. EEG was recorded from 128 channels. The exploration was performed offline on single trial EEG data. We proposed that a successful computational procedure for classification would consist of spatial filtering, temporal filtering, feature selection, and pattern classification. A systematic investigation was performed with combinations of spatial filtering using principal component analysis (PCA), independent component analysis (ICA), common spatial patterns analysis (CSP), and surface Laplacian derivation (SLD); temporal filtering using power spectral density estimation (PSD) and discrete wavelet transform (DWT); pattern classification using linear Mahalanobis distance classifier (LMD), quadratic Mahalanobis distance classifier (QMD), Bayesian classifier (BSC), multi-layer perceptron neural network (MLP), probabilistic neural network (PNN), and support vector machine (SVM). A robust multivariate feature selection strategy using a genetic algorithm was employed.

Results: The combinations of spatial filtering using ICA and SLD, temporal filtering using PSD and DWT, and classification methods using LMD, QMD, BSC and SVM provided higher performance than those of other combinations. Utilizing one of the better combinations of ICA, PSD and SVM, the discrimination accuracy was as high as 75%. Further feature analysis showed that beta band EEG activity of the channels over right sensorimotor cortex was most appropriate for discrimination of right and left hand movement intention.

Conclusions: Effective combinations of computational methods provide possible classification of human movement intention from single trial EEG. Such a method could be the basis for a potential brain-computer interface based on human natural movement, which might reduce the requirement of long-term training.

Significance: Effective combinations of computational methods can classify human movement intention from single trial EEG with reasonable accuracy.

Fig. 1

The complete procedure of offline optimization: dataset generation, training, and testing. The optimization of computational methods was explored using all combinations from spatial filtering, temporal filtering and classification. The optimization experiments were performed five times for each subject.

Fig. 2

Multiple comparison results of significant main effects from three-way ANOVA test on testing accuracy. (a) Spatial filter: ‘ICA’ approach produced significantly higher accuracy than those of ‘None’, ‘PCA’, and ‘CSP’ approaches, but comparable with ‘SLD’ approach. The estimated mean difference between them was about 5–8%. (b) Temporal filter: ‘DWT’ approaches provided significantly higher accuracy than ‘VAR’ approach, but comparable with ‘PSD’ approach. The estimated mean difference between them was about 3%. (c) Classification: linear and quadratic statistical classification methods of ‘LMD’, ‘QMD’, and ‘BSC’, and neural network approach of ‘SVM’ provided significantly higher accuracy than two neural network approaches of ‘MLP’. The estimated mean difference between them was about 1–2%.

Fig. 3

Multiple comparison results of significant interaction between spatial filter and classification method from three-way ANOVA test on testing accuracy. The combinations of spatial filter of ‘ICA’ and ‘SLD’, and classification method of ‘LMD’, ‘QMD’ and ‘SVM’ provided higher classification accuracy.

Fig. 4

Movement-related cortical potentials (MRCP) and event-related desynchronization (ERD) averaged from 11 subjects (excluding subject 2) preceding self-paced right (on the left column) and left (on the right column) hand movements. The waveforms of the MRCP from channel C4 are illustrated in (a) and (b). Peak MRCP amplitude of left hand movement was larger than that of right hand movement. The head topography of MRCP at movement onset is plotted in (c) and (d) for right and left hand, respectively. The MRCP over sensorimotor cortex lateralized to contralateral left hemisphere preceding the right hand movement; the MRCP over sensorimotor cortex lateralized to the contralateral right hemisphere preceding the left hand movement, however, activity on the ipsilateral left hemisphere was also observed before the left hand movement. Time-frequency plots of ERD from channel C4 are shown in (e) and (f). Both alpha and beta band activities were observed over sensorimotor cortex before the movements. The lateralized ERD over left sensorimotor cortex was observed at 500 ms before the onset of right hand movement (g), but bilateral ERD activities on both hemispheres was seen before the left hand movement.

Fig. 5

Feature analysis for the classification of human movement intention. The channel-frequency plots of Bhattacharyya distance, the head topography of Bhattacharyya of alpha band activity (8–12 Hz), and the head topography of Bhattacharyya of beta band activity (16–24 Hz) are illustrated in the first, second and third column, respectively. The average of Bhattacharyya distance from 11 subjects (excluding subject 2) is provided in the fourth row. High separability for intention classification was observed in the beta EEG activity over right sensorimotor cortex, whereas the beta band Bhattacharyya distance was small over left sensorimotor cortex from both individual and average plots. Only subject 3 showed high Bhattacharyya distance in alpha band over contralateral sensorimotor cortex. Bhattacharyya distance of both DC and alpha band components in the other subjects was small.