Proceedings of TOBI Workshop IV
Sion, Switzerland, 2013
The Importance of User-centred Design in BCI
Development: A Case Study with a Locked-in Patient
T. Kaufmann1, E. Holz1,2, A. Kübler1,2
1
University of Würzburg, Würzburg, Germany; 2University of Tübingen, Tübingen, Germany
Correspondence: T. Kaufmann, Department of Psychology I, University of Würzburg, Marcusstr. 9-11, 97070 Würzburg, Germany.
E-mail: [email protected]
Abstract. Although Brain Computer Interfaces have been proposed as communication devices for those with
severest motor impairment, research is rarely performed with this target population, i.e. people in the locked-in
state. Usually, developments are tested in healthy samples and assumptions are made regarding generalization of
results to patient samples. Herein we report a case study with a user in the locked-in state. Different paradigms
on different modalities were applied and far best performance was achieved in the tactile modality, usually
regarded as inferior to the visual and auditory modality. Although she displayed distinct ERPs in a visual
oddball, the visual channel could not be utilized for communication – even a so-called gaze independent speller
failed. Our results thus highly encourage BCI development in the frame of a user-centered approach.
Generalization from healthy participant data to patient samples should be treated with great caution and cannot
replace actual end-user testings. BCIs that may be regarded less effective or less practical, may be the only
possibility for a specific end-user. Adjusting BCI development specifically to end-users’ needs and requirements
is mandatory and will thereby potentially allow for a transfer of BCI technology out of the lab into end-users’
daily lifes.
Keywords: brain computer interface (BCI), event related potentials (ERP), end-users, user-centered design, visual, auditory, tactile
1. Introduction
Brain Computer Interface (BCI) developments are often tested in healthy samples and generalization to
functionality in those with severest motor impairment is often assumed yet less often confirmed. It is well known
that performance deteriorates when bringing this technology to potential end-users and that some users even lack
sufficient control (BCI inefficiency). Retained abilities may greatly vary across potential end-users. Thus, there
is an extensive need for testing BCI developments in patient samples. Their needs and requirements may well
differ from those of healthy participants, care-givers or family members [Zickler et al., 2011]. Herein we present
data from a case study which emphasizes the need of a user-centered design in BCI development.
2. Material and Methods
2.1. The case
The herein presented potential BCI end-user is a 46 year old woman who has been in the locked-in state for
7 years after a brainstem stroke in the pons. She has no reliable muscle control other than vertical eye
movements. During the last year, control of the left thumb rehabilitated but is not yet fully reliable. She
communicates by means of binary partner scanning (yes = eye lift; no = looking down). It is assumed that the
lesion in the brain stem barely affected her cortical abilities (as confirmed by CT) and she was fully attentive
during all sessions.
2.2. Experimental Design
Data was collected on five consecutive days and different setups tested for finding a reliable communication
channel: (1) Oddballs in 3 modalities, i.e. visual (count Einstein face, ignore red squares), auditory (count high
pitched tone, ignore low pitched tone) and tactile (count stimulus on one location, ignore the other; locations
were switched between runs to account for sensitivity differences). All oddballs shared the same parameters
except for the modality and stimulus duration (two stimuli, target to non-targets ratio: 1:5, 1000ms inter-stimulus
interval; 2 runs per modality). (2) Matrix based visual ERP-BCI paradigms in different settings (6x6, 4x4;
different sizes on screen) (3) A so-called gaze-independent visual ERP-BCI paradigm with characters displayed
consecutively in the middle of the screen [Acqualagna et al., 2010]. Only 6 characters were used in this case to
align the paradigm with the parameters (incl. timing) of the visual oddball. (4) Tactile paradigm in which letters
were grouped into four categories that could be selected by focusing attention on one of four tactile stimulation
units placed on the left arm (which was sensory-sensitive in her case). This paradigm was specifically designed
to copy her partner scanning approach, thus allowing for communication in a well-known, long established
setting.
41
Proceedings of TOBI Workshop IV
Sion, Switzerland, 2013
EEG was obtained from 15 passive Ag/AgCl electrodes with mastoid ground and reference and sampled at
512Hz. All paradigms were implemented in Python 2.7 and connected to BCI2000 via UDP.
3. Results
Fig. 1 displays the average ERP at electrode Cz for each modality evoked in two runs of an oddball
paradigm. To control for reliability of elicited ERPs we set up a classifier on one run and tested it on the second
run (and vice versa). Tactile modality displayed most reliable ERPs that resulted in an average offline
classification accuracy of M=100%, i.e. all trials were correctly identified (each run comprised 3 trials with 30
odds per trial). Visual modality was moderately accurate (M=66.5%), whereas auditory ERPs resulted in worst
performance (M=33.0%). We thus tested only visual and tactile modalities with BCI systems.
(1) Visual BCI: Although she reported to fully perceive the entire screen, no matrix based BCI communication
could be established. Interestingly, also the gaze-independent central speller paradigm did not yield positive
results. From none of the acquired data sets, reliable classifier weights could be generated.
(2) Tactile BCI: Four tactile stimuli were the target once in a tactile BCI calibration session. Offline performance
was estimated 100% with 8 stimulation sequences. Unfortunately, this system could not be tested online as
the user had to cancel the last testing session due to strain. Online results are thus pending.
Figure 1: Average ERP at electrode Cz
from the oddball sessions in three different
modalities
4. Discussion
Results of our case study manifest the importance of testing BCI systems in the target population and
developing systems specifically adjusted to their needs. To our experience, only such user-centered BCI design
is able to account for the great variance of users’ retained abilities. From the literature we expected visual (gazeindependent central speller) and auditory modalities to work with the locked-in patient. Although she was not
able to fixate, she reported to fully perceive the entire screen. Compared to the visual oddball, in which a salient
white face was used as rare stimulus, it may be much more difficult to identify a target character among others of
same size and color if the user is not able to fixate the stimuli. This result may question the gaze-independence of
gaze-independent spellers. From the auditory oddball results we assume that no auditory BCI would work in her
case, yet no auditory BCI data were recorded for systematic investigation. Importantly, from the literature we
expected tactile stimulation inferior over others [Aloise et al., 2007], as the only study reporting similarity of
P300 amplitudes across modalities placed an odd stimulus on the sensitive belly while all other stimuli were on
less sensitive locations [Brouwer et al., 2010]. In our case, tactile stimulation evoked large and reliable ERPs that
could accurately be classified. Online results are pending but offline classification accuracy for discrimination
between four tactile stimulation units is promising.
Consequently, we highly encourage to not exclude approaches that may not be feasible for effective
communication from a healthy user’s perspective and to test BCI technology in end-user samples.
Acknowledgements
This work was supported by the European ICT Programme Project FP7-224631. This paper only reflects the
authors’ views and funding agencies are not liable for any use that may be made of the information contained herein.
References
Aloise, F, Lasorsa, I., Brouwer, A. M., Mattia, D., Babiloni, F., Salinari, S., Marciani, M. G., et al. Multimodal stimulation for a P300-based
BCI. International Journal of Bioelectromagnetism, 9(3), 128–130. 2007
Acqualagna, L., Treder, M. S., Schreuder, M., & Blankertz, B. (2010). A novel brain-computer interface based on the rapid serial visual
presentation paradigm. Conf Proc IEEE Eng Med Biol Soc. 2010, 2686–2689.
Brouwer, A.-M., van Erp, J. B. F., Aloise, F., & Cincotti, F. Tactile, Visual, and Bimodal P300s: Could Bimodal P300s Boost BCI
Performance? SRX Neuroscience, 2010, 1–9. 2010
Zickler, C., Riccio, A., Leotta, F., Hillian-Tress, S., Halder, S., Holz, E., Staiger-Sälzer, P., et al. A brain-computer interface as input channel
for a standard assistive technology software. Clinical EEG and Neuroscience: 42(4), 236–244. 2011
42