- Short report
- Open access
- Published:
Biased feedback in brain-computer interfaces
Journal of NeuroEngineering and Rehabilitation volume 7, Article number: 34 (2010)
Abstract
Even though feedback is considered to play an important role in learning how to operate a brain-computer interface (BCI), to date no significant influence of feedback design on BCI-performance has been reported in literature. In this work, we adapt a standard motor-imagery BCI-paradigm to study how BCI-performance is affected by biasing the belief subjects have on their level of control over the BCI system. Our findings indicate that subjects already capable of operating a BCI are impeded by inaccurate feedback, while subjects normally performing on or close to chance level may actually benefit from an incorrect belief on their performance level. Our results imply that optimal feedback design in BCIs should take into account a subject's current skill level.
Findings
Brain-computer interfaces (BCIs) enable subjects to communicate without using the peripheral nervous system by recording brain signals and translating these into control commands [1]. To operate a BCI, subjects need to learn how to intentionally modulate certain characteristics of their brain signals in order to express their intention. For example, in motor imagery, one of the most frequently used experimental paradigms in BCIs [2], subjects are instructed to haptically imagine movements of either the left or right hand, which typically induces a decrease in power of the electromagnetic field of the brain over contralateral sensorimotor cortex in the
As in any form of skill acquisition, subjects require feedback on their performance in order to learn how to optimally regulate their brain signals. While the importance of feedback in BCIs has long been recognized [1], surprisingly little is known on how feedback should be designed in BCIs in order to facilitate the skill acquisition process. In [4], the authors investigated whether instantaneous or delayed feedback proved to be more beneficial. While individual differences could be found, on average no significant effect was observed. Recently, the influence of realistic vs. abstract feedback on BCI performance was investigated [5]. However, the authors again found no evidence for a significant influence of the type of feedback on BCI performance. As such, it appears that the specfic feedback design has little influence on BCI performance.
It should be noted, however, that in previous studies only accurate feedback was considered. While it is generally accepted that feedback in skill acquisition should be timely and precise, motivation is also known to play an important role in BCIs (cf. [6]). Accordingly, subjects may benefit from feedback that trades feedback accuracy for motivation, e.g., by artificially biasing the belief subjects have on their success in the skill acquisition process.
In this work, we investigate the influence of such a feedback bias on BCI performance. Subjects participated in a standard BCI experiment, in which they were asked to navigate a falling ball into a basket in either the left or right corner of the screen by performing haptic motor imagery of either the left or right hand. A depiction of the visual interface is shown in Figure 1. Each experimental trial lasted four seconds, and was considered successful if the ball ended up in the correct half-side of the screen. While usually the horizontal position of the ball on the screen reflects the belief of the BCI system on a subject's intention, we artificially distorted this feedback. Specifically, every two milliseconds we coded the classifier's belief on a subject's intention as a value in the range [0-1]. Then, we drew a sample from a Gaussian distribution, and added this to the classifier's belief. The mean of this sample was chosen as a function of the type of bias, and its variance was determined heuristically and identical for all type of feedback to prevent subjects' awareness of the feedback bias (
Eleven healthy subjects with a mean age of 26.18 ± 4.14 years, seven of them male and four female, participated in the study, all except one were naive to BCIs. Every subject initially performed one session. Four subjects attaining a good level of BCI-control were asked to perform two additional sessions each, as we expected effects to be most prominent in well-performing subjects. Each session consisted of nine runs, with each run being composed of 15 trials per condition in pseudo-randomized order. The first three runs of each session, during which no feedback was presented to the subject, were used to train the classification system. During the following six runs, biased feedback was presented as discussed above. For each session, this resulted in a total of 36 trials for each of the five feedback biases. Mean classification accuracy was then computed for each session and feedback bias, using the undistorted classifier output hidden from the subject. Subjects were not informed that the presented feedback was biased until they had completed their last session.
Setup of visual feedback. Arrangement of the elements present in the visual feedback interface of the used BCI system. The subject is told to look at the fixation cross, which is always present on the screen. During each trial an arrow showing the objective basket appears on screen. The position of the baskets is fixed, and the falling ball always starts at the shown position at the beginning of each trial.
The BCI system employed in this study is described in detail in [7]. Briey, classification was performed by logistic regression with l1-regularization, using logarithmic bandpower in frequency bands ranging from 7 to 40
Mean classification accuracies across all subjects and sessions are shown in Table 1. While subject-specific effects of feedback bias could be observed (not shown here), mean classification accuracy was found to be around 68% for each type of feedback bias. In agreement with previous studies, this appears to indicate that the specific type of feedback had no general effect on BCI performance. However, Figure 2 shows the change in classification accuracy within a session due to each type of bias relative to the no-bias condition, with each dot representing one session and different subjects coded by number. Interestingly, for each type of feedback bias a negative correlation between unbiased classification accuracy and change in classification accuracy due to the bias could be observed. This correlation was found to be highly significant for a strong positive or negative bias (p++ = 0.0045, p- - = 0.0057), and only close to or weakly significant for a weak positive or negative bias (p+ = 0.0762, p- = 0.0384). All p-values were computed by random permutation analysis with 10,000 permutations and n = 648. Furthermore, the points of intersection of the regression lines in Figure 2 with zero change in classification accuracy roughly coincide with the unbiased classification accuracy required to reject chance-level classification accuracy (for each session, an accuracy of 63.9% is required to reject the null-hypothesis of chance-level classification accuracy at significance level
Unbiased classification accuracy vs. deviation in accuracy due to feedback bias. Unbiased classification accuracy vs. deviation from this accuracy due to feedback bias. A +10% value in the y-axis represents a 10% improvement in absolute mean accuracy. Each dot corresponds to one session, the numbers identificating the subjects. Least squares regression lines for each type of feedback bias are shown in grey along with their correlation coefficient. The x2 maker denotes overlapping datapoints corresponding to the same subject.
As the study design required trials with different types of feedback to be interleaved as well as subjects remaining ignorant of the feedback distortion, we could not ask subjects to report their experiences regarding different types of feedback. As such, any interpretation of the observed effects currently remains speculative. We hypothesize that subjects already capable of utilizing a BCI for means of communication are able to make use of instantaneous and accurate feedback in order to optimally regulate their SMR. In these subjects, any type of feedback bias appears to interfere with this feedback loop and hence leads to degraded performance. Accurate feedback in incapable subjects, on the other hand, may be perceived as random noise, as the horizontal movement of the falling ball is uncorrelated with the intended movement direction. We hypothesize that this perceived lack of control leads to frustration and demotivation, impeding an effective skill acquisition process. In these subjects, biased feedback may reduce the perceived randomness of the visual feedback. Specifically, our results indicate that a strong positive bias may be particularly helpful for focussing on the intended task.
In terms of feedback design for future BCI systems, our results suggest that a subject's current skill level should be taken into account. Subjects already capable of modulating their sensorimotor rhythm to some extent should receive accurate feedback. Subjects not yet capable of utilizing a BCI, on the other hand, may benefit by designs that aim to induce a beneficial state-of-mind. While further investigations into the behavioral and neural correlates of a beneficial state-of-mind for BCIs are required (cf. [9, 10] for two recent studies on this topic), the results presented here suggest that incapable subjects may particularly benefit if their belief on the level of control over the BCI-system is positively biased.
References
Wolpaw JR, Birbaumer N, Heetderks WJ, McFarland DJ, Peckham PH, Schalk G, Donchin E, Quatrano La, Robinson CJ, Vaughan TM: Brain-computer interface technology: a review of the first international meeting. IEEE Transactions on Rehabilitation Engineering 2000, 8: 164-173. 10.1109/TRE.2000.847807
Mason S, Bashashati A, Fatourechi M, Navarro K, Birch G: A comprehensive survey of brain interface technology designs. Annals of Biomedical Engineering 2007,35(2):137-169. 10.1007/s10439-006-9170-0
Pfurtscheller G, Neuper C: Motor Imagery and Direct Brain-Computer Communication. Proceedings of the IEEE 2001,89(7):1123-1134. 10.1109/5.939829
McFarland D, McCane L, Wolpaw J: EEG-based communication and control: short-term role of feedback. IEEE Transactions on Biomedical Engineering 1998, 6: 7-11.
Neuper C, Scherer R, Wriessnegger S, Pfurtscheller G: Motor imagery and action observation: Modulation of sensorimotor brain rhythms during mental control of a brain-computer interface. Clinical Neurophysiology 2009,120(2):239-247. 10.1016/j.clinph.2008.11.015
Curran EA, Stokes MJ: Learning to control brain activity: A review of the production and control of EEG components for driving brain-computer interface (BCI) systems. Brain and Cognition 2003,51(3):326-336. 10.1016/S0278-2626(03)00036-8
Grosse-Wentrup M, Liefhold C, Gramann K, Buss M: Beamforming in non-invasive Brain-Computer Interfaces. IEEE Transactions in Biomedical Engineering 2009,56(4):1209-1219. 10.1109/TBME.2008.2009768
Ramoser H, Mueller-Gerking J, Pfurtscheller G: Optimal Spatial Filtering of Single Trial EEG During Imagined Hand Movement. IEEE Transactions on Rehabilitation Engineering 2000,8(4):441-446. 10.1109/86.895946
Blankertz B, Sannellia C, Halder S, Hammer E, Kübler A, Müller KR, Curio G, Dickhaus T: Neurophysiological predictor of SMR-based BCI performance. NeuroImage 2010,51(4):1303-1309. 10.1016/j.neuroimage.2010.03.022
Grosse-Wentrup M, Hill J, Schölkopf B: Causal Influence of Gamma Oscillations on the Sensorimotor-Rhythm. NeuroImage 2010, in press.
Acknowledgements
This work was developed at the Max Planck Institute for Biological Cybernetics, under partial support of Spain's TIN 2007-66862 and "Cátedra UAM-IIC en Modelado y Predicción". The first author is supported by the FPU-MEC grant reference AP2006-02285. We would like to acknowledge the support of Bernd Battes for participating in the preparation and execution of the BCI experiments.
Author information
Authors and Affiliations
Corresponding author
Additional information
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
AB carried out the BCI experiments for this study, adapted the BCI system to include the feedback bias, performed the statistical analysis and participated in the writing of the manuscript. MGW conceived and supervised the study, and participated in the data acquisition, statistical analysis and writing of the manuscript. All authors read and approved the final manuscript.
Authors’ original submitted files for images
Below are the links to the authors’ original submitted files for images.
Rights and permissions
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
About this article
Cite this article
Barbero, Á., Grosse-Wentrup, M. Biased feedback in brain-computer interfaces. J NeuroEngineering Rehabil 7, 34 (2010). https://doi.org/10.1186/1743-0003-7-34
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/1743-0003-7-34