Motor Priming as a Brain-Computer Interface

  • Tom Stewart
  • Kiyoshi Hoshino
  • Andrzej Cichocki
  • Tomasz M. Rutkowski
Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 9948)


This paper reports on a project to overcome a difficulty associated with motor imagery (MI) in a brain–computer interface (BCI), in which user training relies on discovering how to best carry out the MI given only open-ended instructions. To address this challenge we investigate the use of a motor priming (MP), a similar mental task but one linked to a tangible behavioural goal. To investigate the efficacy of this approach in creating the changes in brain activity necessary to drive a BCI, an experiment is carried out in which the user is required to prepare and execute predefined movements. Significant lateralisations of alpha activity are discussed and significant classification accuracies of movement preparation versus no preparation are also reported; indicating that this method is promising alternative to motor imagery in driving a BCI.


Brain-computer interface (BCI) Motor priming EEG Neurophysiological information processing and classification 

1 Introduction

A brain–computer interface (BCI) offers a unique mode of communication between a user and an environment by interpreting and acting upon signals arising directly from the brain [10]. In doing so, BCI bypasses the body’s usual output pathway which requires motor signals to be sent from the brain, through the spinal cord and peripheral nerves to reach, and contract skeletal muscles [10]. Through a number of conditions, most notably spinal cord injury and ALS, this pathway can be interrupted and in extreme cases, the body can lose control of all of its skeletal musculature. In those cases, BCI is the only means available for a patient to interact with his/her environment. As such, one critical application for BCI is to control assistive devices such as wheelchairs and neuroprosthesis.

1.1 Motor Imagery

Current BCI’s aimed at neuroprosthesis control are usually based around a mental task known as motor imagery [10]. To perform motor imagery, a subject is instructed to imagine moving a particular part of their body such as opening and closing one of their hands. The act of imagining a movement is thought to engage the same neural circuits as those that would be involved in planning an actual movement [5] and has been shown to cause patterns in the amplitude of \(8-12\) Hz brainwave EEG activity. The nature of these patterns is consistent with a topographical map of the body found in the brain’s sensorimotor cortex known as the homunculus. In general, motor imagery relating to a particular body part will cause a suppression of \(\mu -\)rhythms in the corresponding region of the homunculus. To harness this phenomenon and create a BCI, machine learning algorithms are used to classify these brainwave patterns and to output an estimate of which body part’s movement, if any, was being imagined based on a user’s on–going EEG readings. The result of this classification can then be used to create a command which may operate a computer or some form of assistive device [5]. As a mental task, motor imagery (MI) is well suited to brain computer interface for a number of reasons. Most importantly, it can be used to communicate different commands depending on the part of the body which the subject chooses to imagine moving. Secondly, MI represents a mental state that the brain can enter into entirely of it’s own volition and without the need for external stimulus. This means that subjects can potentially drive the BCI at their own pace [10]. Finally, MI has been shown to be possible even when the subject has no control over the body part with which the movement is being imagined meaning that it can be utilised even by subjects with severe paralysis [10]. To date, MI–based BCI have been successfully applied to a number of applications including navigating through virtual worlds [6] flying quadrocopters [3] and even controlling the exoskeleton that was used to deliver the opening kick at the 2014 FIFA World Cup.

However, despite it’s potential, this type of BCI has yet to reach the level of robustness and reliability necessary to be applied practically outside of laboratory conditions [8]. One widely recognized challenge associated with this type of BCI is training users in effectively utilising MI to create strong and reliable patterns in their brainwave activity. Motor Imagery is an abstract mental visualisation task without a tangible behavioural goal and it’s therefore difficult to give explicit instructions on how it should be properly carried out. The instructions that are given are quite open to interpretation and usually in the form of “imagine opening and closing your left hand”. This allows for the possibility that different subjects may interpret these instructions differently and arrive at varying strategies on how to carry out the MI task. If these differing strategies bring with them differing levels of efficacy then this might account for the reported variability in subject performance of MI [2].

1.2 Motor Priming

This research aims to address this challenge by investigating the use of a motor priming (MP) as an alternative to MI in controlling a brain computer interface. MP, also referred to as a motor attention [7], is the mental process of covertly preparing to execute a movement. This concept has been studied extensively in the field of cognitive neuroscience to elucidate the neural mechanisms involved in the preparation and execution of voluntary movement [1]. It has been found that preparation to make a voluntary hand movement gives rise to an effect known as a motor related amplitude asymmetry in the alpha band (\(8-12\) Hz) of the EEG spectrum [1]. The orientation of this asymmetry is dependent on which hand is being prepared for the movement. Left hand movement preparation is accompanied by a left hemespheric increase in alpha activity and a right hemispheric decrease. During right hand movement preparation the reverse can be observed. This effect is most prominent over the CP3 / CP4 electrode pair, which are situated over the motor cortex. This lateralisation of alpha activity closely resembles effects of MI. Specifically, hand motor imagery has also been found to cause similar contralateral reductions in alpha band activity [2]. These too can be best measured over motor cortex electrodes, typically C3 and C4.
Fig. 1.

Experiment trial timing.

Despite these parallels with MI, MP has received little, if any attention from the BCI community. Never the less, motor priming has several key characteristics which might make it an appealing alternative to motor imagery in voluntarily generating the brain activity patterns necessary to control a BCI. Most importantly, motor priming is tied to a clear behavioural goal, that is, to execute a prepared movement. Therefore in order to explain to a subject how to carry out MP, it’s only necessary to instruct them to make a predefined movement as quickly as possible. It can then be reasoned that the subject will instinctively place themselves in the required mental state in order to minimise their reaction time and in so doing create the necessary changes in brain activity to drive the BCI. This bypasses the need to give subjects instructions on how to carry out a mental task which may be difficult to convey and prone to misinterpretation. Furthermore, since the execution of a covertly prepared movement is common in daily activities, it’s reasoned that subjects might be able to draw on instinct or past experience to engage in motor priming, and might therefore be capable of carrying it out with less training than motor imagery. Finally, since motor priming is linked to a subject’s reaction time in carrying out a movement, that reaction time can be used as an unambiguous metric for subject performance. This allows for clear feedback during user training as well as a means of identifying how engaged the user is in carrying out the mental task.

2 Materials and Methods

To investigate whether different motor priming tasks could be classified on a single trial basis and thus serve as an EEG control strategy, an experiment was carried out in which seven subjects were asked to covertly prepare, and then subsequently execute a movement as quickly as possible upon the presentation of a “go” signal.

The experiments with human subjects reported in this paper were conducted with an approval of RIKEN Brain Science Institute Ethical Committee permission. Subjects were seated comfortably in front of a computer monitor. Subjects were asked to place their hands on two arcade–style buttons and fixate their gaze on a small dot shown in the middle of the computer monitor. The experiment consisted of 300 trials with a rest period after each trial and an extended rest period every 50 trials. Each trial began with a 1500 ms interval to be used as a baseline in the EEG data analysis. Immediately after the baseline period, the fixation dot on the computer monitor was replaced with one of three cues for a duration of 200 ms. The cue informed the subject of which movement to prepare to execute. A left arrow indicated a left button press, a right arrow indicated a right button press and a circle indicated that the subject should not prepare any movement at all. Trials in which no movement was prepared are hereafter referred to as “no–go” trials whilst left and right hand movement trials were collectively referred to as “go” trials. These cues are summarised in Table 1. After random delay of between 500 and 2000 ms a go signal was presented to the subject in the form of an auditory tone. Subjects were asked to respond to the go signal by making the cued movement as quickly as possible. To encourage the subject to minimise their reaction time, feedback was given upon the user’s response in accordance with their reaction time. This feedback was in the form of increasingly congratulatory sounds depending on the speed of the subject reaction. The timing of the experiment is shown in Fig. 1.
Table 1.

Preparatory cues and associated movements

Preparatory cue


\(\rightarrow \)

Press right button with right hand

\(\leftarrow \)

Press left button with left hand

\(\downarrow \)

Press foot pedal with right foot

\(\bigcirc \)

No movement

2.1 Data Collection

Behavioural data in the form of reaction times and EEG activity were collected throughout each experimental session. EEG activity was measured using a g.HIamp amplifier by gtec Medical Instruments GmbH, Austria, which recorded activity from 62 scalp locations conforming to a 10 / 10 electrode montage. EEG measurement was carried out at 512 Hz using a right ear reference and active electrodes.

2.2 Data Analysis

Reaction time was taken as the time between the onset of the go signal and the subject’s response. Accuracy was defined as the percentage of correct responses where incorrect movements or failures to respond were taken as errors. EEG data analysis was carried out using Matlab in conjunction with the EEG/MEG data analysis toolbox Fieldtrip [4]. The continuous EEG data was first preprocessed into epochs time–locked to the onset of the preparatory cue. In the present analysis, only trials with a delay period of 2000 ms were considered which constituted \(50\,\%\) of the total. The remaining trials were included to ensure that the subject could not predict the “go” signal and therefore had to focus their attention for the duration of the random delay. Artifact rejection was carried out manually by omitting trials and channels on the basis of EEG signal variance. All analysis was focused on the alpha band of the EEG spectrum (\(8-12\) Hz) which was extracted using a Morlet wavelet transform. Two analysis approaches were used to investigate how effectively a classification might be made between “go” vs. “no–go” trials as well as left hand movement vs. right hand movement trials. To investigate left hand vs. right hand trials an area under the ROC curve (AUC) analysis was carried out using a threshold placed on the log lateralisation index as a binary classifier. To calculate the lateralisation index, the alpha band power from each channel in the left hemisphere \((\alpha _L)\) was matched to counterpart in the right hemisphere \((\alpha _R)\) and the index was calculated for the pair using \(LI = \log (\alpha _l/\alpha _R)\). AUC scores were calculated for each channel pair and time within the period between the cue and the go signal. The statistical significance of the scores was assessed using a permutation test to generate a population of 300 AUC scores against which the original score was compared using a two-tailed t-test. The “go” vs. “no–go” trials weren’t expected to exhibit any lateralisation so instead logistic regression was used to discriminate between left and right hand movement and no movement trials. The classifier was trained for each time in the cue / go signal delay period using the instantaneous alpha band power measured from all channels as a feature vector. To reduce the effects of overfitting and perform dimensionality reduction a regularisation component was also included and set to \(\alpha = 0.2\). The accuracy of the resulting classifiers was evaluated using five–fold cross–validation.
Fig. 2.

Subject \(\#1\) accuracies (a) and electrode importance map (b) results. (a) Left hand / no movement classification accuracy vs. time. (b) Left hand / no movement classifier importance map.

3 Results

The average reaction time across all subjects was \(416 \pm 112\) ms while the accuracy was \(97.9 \pm 2.2\,\%\). Classification accuracy between “go” and “no–go” trials exhibited a high degree of variability depending on the latency at which the classifier was trained. All subjects temporarily reached periods of statistically significant classification however only the subjects \(\#1, \#2\) and \(\#5\) showed sustained periods high classification (\(>100\) ms). By inspecting the topography of weights learned by classifier during these periods it was found that these classifications were made primarily on the basis of an alpha desynchronisations in the occipital region. Figure 2a shows the classification accuracy left hand movement vs. no movement by the subject \(\#1\) with respect to time. Figure 2b shows the importance map of the weights learned by the classifier during the periods of statistically significant classification.

In the AUC analysis, subjects \(\#1\), 3&7 showed statistically significant lateralisations of alpha activity in response to left vs. right hand movement priming. These manifested as an increase in the AUC score almost immediately after the onset of the cue and reaching statistical significance (\(p < 0.05\)) at roughly t = 500 ms. This effect was observed for the duration of the trial and peaked in significance over the \(C3-C4\) and \(CP1-CP2\) electrode pairs.

Figure 3a shows the changes in AUC score with respect to time for the most highly discriminating electrode pairs in subjects \(\#1\), 3&7. Figure 3a shows the statistical significance of these AUC scores in a two tailed t-test. Figure 3b shows the topography of the AUC scores for each electrode pair, note that the right hemisphere scores were set to \(1-AUC\) for the corresponding electrode pair.
Fig. 3.

Subjects \(\#1\), 3 and 7 AUC scores and statistical significances (a) subject \(\#3\) AUC score topography. (b). (a) Statistical significances of AUC scores vs. time. (b) AUC score topography in left hand / right hand discrimination.

4 Discussion

The behavioural results showed similar reaction times and high task accuracy across all subjects suggesting that all seven participants were able to carry out the experiment with similar, satisfactory levels of proficiency. The AUC analysis revealed strong lateralization in alpha band activity over the sensorimotor cortex in three subjects in response to left or right hand movement preparation. The AUC scores reached strong statistical significance \((p < 0.01)\) within 500 ms after the onset of the cue; suggesting that this effect was indeed a result of movement preparation. In the context of BCI control, this results suggests that a user may issue either a left or right command on the basis of alpha band lateralisation over \(C3-C4 / CP1-CP2\) electrode pairs through MP in the same way as is possible through the use of MI.

Additionally, the “go” vs “no–go” trials, showed spurious but statistically significant discrimination based on desynchronisations in occipital alpha activity. These desynchronisations did not show any consistent spatial characteristics between the different movement conditions therefore it’s likely that this distinction is a result of a heightened state of attention during the movement trials rather than any motor specific activity. Nevertheless, this result suggests that it might be possible to distinguish intentional left and right BCI commands from ambient lateralisation in rolandic activity. The question remains as to why only three of the seven subjects exhibited significantly elevated AUC scores. The observed reaction times suggested all subjects were actively engaged with the mental task, however a more fine grained analysis may reveal underlying differences to suggest reduced concentration in the non–lateralising subjects. Alternatively, similar research on both motor priming [1] as well as visual spatial attention [9] has found that different subjects show varying degrees of lateralisation which might account for this inter-subject variability. If this is the case then a more general feature extraction method such as a common spatial pattern might yield better separation of left vs. right hand trials.

5 Conclusions

This research demonstrates that by carrying out a simple movement preparation task given only basic instructions, it’s possible for subjects to achieve significant changes in alpha band EEG activity that may be used to drive a brain computer interface. These changes in brain activity have similar characteristics to with those associated with motor imagery in terms of a contralateral decrease and an ipsilateral increase in alpha power relative to the hand being prepared for movement. Unlike MI, MP is linked to a tangible behavioural goal and should therefore be easy and intuitive for new users to acquire. The is hypothesis is supported by the fact that the three subjects who achieved significant lateralisation did so in one experimental session despite having no prior experience with MP tasks.

Further investigation into the reaction time characteristics of the non–laterali-sing subjects along with more general EEG feature selection is expected to account for the observed variability in subject performance.


  1. 1.
    Deiber, M.P., Sallard, E., Ludwig, C., Ghezzi, C., Barral, J., Ibañez, V.: EEG alpha activity reflects motor preparation rather than the mode of action selection. Front. Integrat. Neurosci. 6(59) (2012)Google Scholar
  2. 2.
    LaFleur, K., Cassady, K., Doud, A., Shades, K., Rogin, E., He, B.: Quadcopter control in three-dimensional space using a noninvasive motor imagery-based brain-computer interface. J. Neural Eng. 10(4), 046003 (2013)CrossRefGoogle Scholar
  3. 3.
    Lotte, F., Larrue, F., Mühl, C.: Flaws in current human training protocols for spontaneous brain-computer interfaces: lessons learned from instructional design. Front. Hum. Neurosci. 7, 568 (2013)CrossRefGoogle Scholar
  4. 4.
    Oostenveld, R., Fries, P., Maris, E., Schoffelen, J.M.: Fieldtrip: open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data. Comput. Intell. Neurosci. 2011, Article no. 1 (2010)Google Scholar
  5. 5.
    Pfurtscheller, G., Brunner, C., Schlögl, A., da Silva, L.: Mu rhythm (de) synchronization and EEG single-trial classification of different motor imagery tasks. Neuroimage 31(1), 153–159 (2006)CrossRefGoogle Scholar
  6. 6.
    Pfurtscheller, G., Neuper, C.: Motor imagery and direct brain-computer communication. Proc. IEEE 89(7), 1123–1134 (2001)CrossRefGoogle Scholar
  7. 7.
    Rushworth, M.F., Krams, M., Passingham, R.E.: The attentional role of the left parietal cortex: the distinct lateralization and localization of motor attention in the human brain. J. Cogn. Neurosci. 13(5), 698–710 (2001)CrossRefGoogle Scholar
  8. 8.
    Scherer, R., Lee, F., Schlogl, A., Leeb, R., Bischof, H., Pfurtscheller, G.: Toward self-paced brain-computer communication: navigation through virtual worlds. IEEE Trans. Biomed. Eng. 55(2), 675–682 (2008)CrossRefGoogle Scholar
  9. 9.
    Van Gerven, M., Bahramisharif, A., Heskes, T., Jensen, O.: Selecting features for BCI control based on a covert spatial attention paradigm. Neural Netw. 22(9), 1271–1277 (2009)CrossRefGoogle Scholar
  10. 10.
    Wolpaw, J., Wolpaw, E.W.: Brain-Computer Interfaces: Principles and Practice. Oxford University Press, Oxford (2012)CrossRefGoogle Scholar

Copyright information

© Springer International Publishing AG 2016

Authors and Affiliations

  • Tom Stewart
    • 1
    • 2
  • Kiyoshi Hoshino
    • 2
  • Andrzej Cichocki
    • 1
  • Tomasz M. Rutkowski
    • 1
    • 3
    • 4
  1. 1.RIKEN Brain Science InstituteWako-shiJapan
  2. 2.University of TsukubaTsukubaJapan
  3. 3.The University of TokyoTokyoJapan
  4. 4.Saitama Institute of TechnologySaitamaJapan

Personalised recommendations