Translation of EEG spatial filters from resting to motor imagery using independent component analysis

Abstract

Electroencephalogram (EEG)-based brain-computer interfaces (BCIs) often use spatial filters to improve signal-to-noise ratio of task-related EEG activities. To obtain robust spatial filters, large amounts of labeled data, which are often expensive and labor-intensive to obtain, need to be collected in a training procedure before online BCI control. Several studies have recently developed zero-training methods using a session-to-session scenario in order to alleviate this problem. To our knowledge, a state-to-state translation, which applies spatial filters derived from one state to another, has never been reported. This study proposes a state-to-state, zero-training method to construct spatial filters for extracting EEG changes induced by motor imagery. Independent component analysis (ICA) was separately applied to the multi-channel EEG in the resting and the motor imagery states to obtain motor-related spatial filters. The resultant spatial filters were then applied to single-trial EEG to differentiate left- and right-hand imagery movements. On a motor imagery dataset collected from nine subjects, comparable classification accuracies were obtained by using ICA-based spatial filters derived from the two states (motor imagery: 87.0%, resting: 85.9%), which were both significantly higher than the accuracy achieved by using monopolar scalp EEG data (80.4%). The proposed method considerably increases the practicality of BCI systems in real-world environments because it is less sensitive to electrode misalignment across different sessions or days and does not require annotated pilot data to derive spatial filters.

Figure 1. Experiment paradigm for the motor imagery-based brain-computer interface.

Figure 2. Scalp topographies and PSDs of all ICs from one subject.

(A) Scalp topographies; (B) PSDs. IC5 and IC7, which both show a unilateral spatial distribution over the sensorimotor cortex and a mu/beta-band dominant spectral profile, are highlighted by a black rectangle as the selected motor components (cf. details of the identification process in Table 1).

Figure 3. Group-averaged ERSP and PSD for two motor components.

(A) Group-averaged time-frequency distributions of ERSP for the left motor IC and the right motor IC corresponding to left and right hand movement imaginations; (B) Group-averaged PSD of left and right motor ICs under different conditions (RE: resting state, MI: motor imagery state, MI-L: left-hand motor imagery, MI-R: right-hand motor imagery).

Figure 4. Diagram of translating spatial filters from the resting state to the motor imagery state.

Similar spatial filters and spatial patterns were obtained by ICA on data corresponding to the two conditions separately. Spatial filters obtained from the resting data could be used as estimates of those from the motor imagery data.

Figure 5. Spatial patterns and spatial filters of the motor components for all nine subjects.

(A) spatial patterns of the resting state; (B) spatial patterns of the motor imagery state; (C) spatial filters of the resting state; (D) spatial filters of the motor imagery state. Black dots in each scalp map indicate positions of C3 and C4 electrodes. In each subfigure, the left and right motor ICs for all subjects were grouped on the left and the right panel respectively.

Figure 6. Averaged power spectrum density of EEG signals in motor imagery practice across all subjects.

(A) monopolar scalp data at C3 and C4 electrodes; (B) motor-related independent components extracted by ICA using the motor imagery data; (C) motor-related independent components extracted by ICA using the resting data.

Figure 7. Spatial distributions of EEG power difference and IC spatial pattern difference.

(A) power difference between left- and right-hand motor imagery conditions; (B) difference of spatial patterns between left and right independent motor components obtained from the motor imagery data.

Figure 8. EEG power of motor ICs during resting and motor imagery states.

(A) EEG power of motor ICs during motor imagery; (B) EEG power of motor ICs during resting; (C) Weighted EEG power of motor ICs during motor imagery (original power divided by the mean power of the resting data). (D) Single-trial EEG power of motor ICs during motor imagery on one subject. (E) Single-trial EEG baseline power of motor ICs during resting. (F) Weighted single-trial EEG power of motor ICs during motor imagery. In (A), (B), and (C), each solid line connects left hand and right hand data for a subject. The dash line indicates the line .

Figure 9. Spatial patterns and averaged PSDs of the three motor ICs for Subject 5.

(A) Spatial patterns; (B) Averaged PSDs in motor imagery practice.