Many BCIs are cue-paced, i.e. they signal the start and end of the control phase to the user. Because this gives the BCI a task-like structure, such a synchronous approach is not very useful for assistive applications. Yet, there are several benefits of such a task-structure, which makes this approach very applicable for restorative BCIs or neurofeedback.

First, we know that the subject will try to control the BCI during a certain phase of the task. This allows simple supervised classifier training and adaptation, monitoring of the subjects performance and learning, and easy control of the training dosage. One example is motor imagery driven BCIs for stroke rehabilitation. In that example, the start of the imagination phase would be signalled by a (visual, auditory, haptic or multimodal) cue, the patient starts to imagine the movement, the BCI measures whether the desired brain modulation is sufficient, and delivers (visual, auditory, haptic or multimodal) feedback to the patient. The first couple of runs can be used for training or adaptation of the feature weights; in subsequent runs the patient performs motor imagery training, and its performance can be monitored by the therapist.

Yet, such cue-paced approaches come with one caveat. Obviously, we would already have sufficient information to classify the brain state of the subject with perfection. Whenever the subject is in the control phase, it will perform motor imagery. If the subject is not in the control phase, it does not. If we would inform the classifier about the current phase, i.e. include task phase as a feature, we could easily achieve a high classification accuracy – simply because task phase and class are identical. This is pretty obvious, and therefore, no sound classifier would take this feature into account. Instead, we would only use features which are actually based on the subjects brain activity. But alas, the subject will process the cue, and this will result in some event-related potential or oscillation. The subject acts therefore as a noisy filter for the cue, which will then pop up in the feature space for the classifier. Via this backdoor, the classifier is informed about the current task phase.

The main solution to this issue is to use only features which are not modulated by the cue. This could be achieved e.g. by using only features from a later period in the control-phase, when the sensory processing is already over (e.g. after the P300 occured); or limiting the feature space to electrodes or frequencies which are usually not affected by sensory processing, but indicative of motor imagination (e.g. central beta-power). Yet, the picture is a little bit more complicated.

Formally, even such features (assumed to be unrelated) are linked to the sensory processing of the cue:  if there had been no go cue, the subject would not have performed motor imagination. This appears to be a trivial – but note that there is research on faciliating motor imagery by appropriate cueing.  Cues have been found to increase vividness [1,2], and visual cues appear to increase spatial accuracy, while auditory cues increase temporal accuracy [2]. Additionally, pre-cue brain activity appears to predict the later performance of the subject [3]. Naturally, from a therapists perspective, we would want to faciliate motor imagery with cues. But improving the subjects performance and attention with cues, while removing the processed cue from the feature space can then be considered a non-trivial issue in the research of synchronous BCIs.

Put shortly: Synchronous BCIs exploit cues to improve subject performance and simplify the classification. This comes with one caveat: how to disentangle the processed cue from the features used for classification. 

References

1. Heremans, Elke, Alice Nieuwboer, Peter Feys, Sarah Vercruysse, Wim Vandenberghe, Nikhil Sharma, and Werner F Helsen. “External Cueing Improves Motor Imagery Quality in Patients with Parkinson Disease.” Neurorehabilitation and Neural Repair 26, no. 1 (January 2012): 27–35. doi:10.1177/1545968311411055.
2. Heremans, Elke, Werner F Helsen, Harjo J De Poel, Kaat Alaerts, Pieter Meyns, and Peter Feys. “Facilitation of Motor Imagery through Movement-Related Cueing.” Brain Research 1278 (June 30, 2009): 50–58. doi:10.1016/j.brainres.2009.04.041.
3. Bamdadian, Atieh, Cuntai Guan, Kai Keng Ang, and Jianxin Xu. “The Predictive Role of Pre-Cue EEG Rhythms on MI-Based BCI Classification Performance.” Journal of Neuroscience Methods 235 (September 30, 2014): 138–44. doi:10.1016/j.jneumeth.2014.06.011.

Robert Bauer

Written by Robert Bauer

Agricolab | Descendant of Latin ‚agricola‘, farmer; Lab (colloquial) A laboratory

Schreibe einen Kommentar

Deine E-Mail-Adresse wird nicht veröffentlicht. Erforderliche Felder sind mit * markiert.