32nd Annual International Conference of the IEEE EMBS
Buenos Aires, Argentina, August 31 - September 4, 2010
Toward BCI Wizard - best BCI approach for each user
Ivan Volosyak, Christoph Guger and Axel Gräser
Abstract— Modern brain-computer interface (BCI) systems
use different types of neural activity for control. Most BCI
systems only allow the customization of very few parameters
and focus only on one type of BCI approach. Many articles
reported that a certain BCI did not work for some users
(so called BCI illiteracy). We are introducing the BCI wizard
as a system that automatically identifies key parameters to
customize the best BCI paradigm for each user. With a BCI
wizard it is possible to develop an interface that relies on the
best mental strategy for each user and therefore makes the
difference between an ineffective system and a working BCI.
This work presents a preliminary study that aims to develop a
BCI wizard exploring the two most effective BCI approaches
(SSVEP and P300). These types of non-invasive BCIs were
tested and evaluated in a group of 14 healthy subjects. During
online tests all subjects were asked to spell three words with
two spelling applications and at the end of the experiment they
chose their preferred approach. Results showed that all subjects
could communicate with the P300-based BCI with an accuracy
above 69% (5 reached 100% accuracy), 10 out of 14 subjects
could effectively use the SSVEP-based BCI (2 reached 100%
accuracy). These promising results confirm that BCI wizard will
enable BCIs customized to each user with considerably greater
flexibility and independence than present systems allow.
I. I NTRODUCTION
A brain-computer interface (BCI) allows communication
and control without requiring limb articulation or any move-
ment from the user. The normal communication channels
such as speech and movement are not used, but instead brain
activity is directly recorded and transformed into control
signals [1]. There exist different neural activities that could
be used as features for a BCI. Slow cortical potentials (SCPs)
constitute one of the first non-invasive BCI approaches,
which has been used until very recently [2]. Motion-onset
visual evoked potentials (mVEPs) [3], auditory evoked po-
tentials (AEP) [4], [5], and steady-state movement related
potentials (ssMRPs) [6] are modern examples of sources
also used in BCIs. The commonly used and effective ap-
proaches among non-invasive BCIs are the steady state visual
evoked potential (SSVEP) [7], [8], the 300-ms component
of an evoked potential (P300) [9], and sensorimotor rhythms
(SMR) [10]. These approaches have been studied extensively
in healthy participants and subjects with disabilities, and have
been validated as input signals that can be used in a BCI.
I. Volosyak* and A. Gräser are with the Institute of Automation (IAT),
University of Bremen, 28359 Bremen, Germany, www.iat.uni-bremen.de
C. Guger is with g.tec medical engineering GmbH, Guger Technologies
OEG, 8020 Graz, Austria, www.gtec.at
This document is a collaborative effort of Institute of Automation, Uni-
versity of Bremen, Germany and g.tec Medical Engineering GmbH, Austria.
This research was supported within the 7th European Community Frame-
work Program by a Marie Curie European Re-Integration Grant RehaBCI,
and by the EU ICT grants BRAIN, SM4ll and Presenccia. *Corresponding
author is Ivan Volosyak, volosyak@iat.uni-bremen.de
978-1-4244-4124-2/10/$25.00 ©2010 IEEE 4201
However, each lab tends to focus on only one BCI
approach and it still remains unclear which approach is best
suited for the individual user. To date, the first and only study
that explored different mental strategies with the same users
was done by Birbaumer and colleagues in collaboration with
the Wadsworth Center in Albany, New York [11]. The cited
work compared three non-invasive approaches (SCP, P300,
and SMR) with each user in their home environment. Eight
severely paralyzed patients with amyotrophic lateral sclerosis
(ALS) participated in the study. All subjects received 20
sessions of SMR training, six had ten sessions of P300
training and in addition, seven participated in 20 sessions
of SCP training. Results showed that SCP performance was
not sufficient for communication after 20 sessions for any of
the seven subjects. In contrast, SMR training was successful
for half of the subjects. P300 was the best with most subjects.
These results suggest that the P300-based BCI might be the
most efficient BCI for ALS patients [12]. This study also
confirms that for each BCI approach, some subjects could
not produce the neuronal activity needed for control. Other
studies reveal that about 10-25% of users are unable to attain
effective control [13]. The motor imagery BCI system was
also tested with 99 subjects at an exhibition in Graz [14].
The subjects were trained for 6 minutes to imagine left or
right hand movement for a few seconds (20 times each) to
produce SMR changes. The BCI system was then trained
on the individual EEG data for a subsequent session with
visual feedback of cursor movement. The subjects were
able to control the cursor to the left or right side of the
computer screen. 6.2% of the subjects were able to learn
this control with 100% accuracy in this short training session.
About 93.3% showed a control above 59% accuracy (50%
corresponds to random classification). The P300 speller was
tested with 100 subjects with a row/column (RC) and a single
character (SC) speller [15]. 72.8% (81 subjects) were able
to spell with 100% accuracy in the RC paradigm and 55.3%
(N=38) spelled with 100% accuracy in the SC paradigm.
Less than 3% of the subjects did not spell any character
correctly. The SSVEP based Bremen-BCI system was tested
with 37 subjects at an exhibition in Düsseldorf [16]. 86% (32
subjects) were able to complete spelling tasks successfully.
Therefore, flexible and reliable BCIs systems that can
automatically identify and optimize important parameters not
only for some specific BCI paradigm, but across different
BCI approaches are needed. The building of a system that
performs a series of tests with the user with parameter
adaptation within each paradigm and the incorporation of
advanced and flexible signal processing tools may help to
develop a BCI best suited for the individual user.
 II. M ETHODS AND M ATERIALS
The present work constitutes the first effort in building
an expert system that walks the user through a series of
tests to determine optimal parameters across different BCI
approaches. We called this system the BCI wizard. For each
test subject, the wizard will determine the appropriate spatial
filters, electrode sites, and further optimal parameters. For
example, for the P300 paradigm the wizard will determine
the best inter-stimulus timing, stimulus duration, matrix size
and number of sequences. Different stimulus frequencies,
time segment lengths, and display parameters will be as-
sessed for SSVEP BCIs. Since these two approaches are
very reliable and exhibited higher information throughput
than other approaches, there was no reason to pursue other
approaches in this preliminary study.
Additionally, the user may also choose to answer questions
about training, application parameters and other personal
preferences. Based on this information, an expert system will
then determine the best BCI approach and parameters for
each subject. A BCI wizard will enable BCIs customized to
each user with considerably greater flexibility and indepen-
dence than present systems allow.
The principal goal of the present study was to compare and
evaluate the most effective BCI approaches and to identify
key parameters that can be adjusted for each user to improve
BCI performance. The comparison between the different BCI
approaches was done by measuring the information transfer
rate (ITR). This measure is a standard method for measuring
communication and control that incorporates speed and ac-
curacy in a single value [17]. This study also assessed with
brief questionnaires subjective information such as which
paradigm was preferred by the user. This is very important,
because when a user e.g. dislikes stimuli that flash more
quickly than twice per second, the BCI wizard rules out
an SSVEP approach, and instead recommends the next best
approach, for example, a P300 BCI. Or e.g. if the user had
either been willing to undergo training, or shown very strong
SMR activity, an SMR BCI might have been recommended
instead.
A. Subjects
A total of 14 subjects participated in the study. Subjects
mean age was 30.29 years, range 22-41 with standard devia-
tion of 4.78. Five subjects were female and nine male. None
of the subjects have had any prior experience with P300 BCI.
All subjects completed six spelling tasks across the P300 and
SSVEP approaches during one day (P300 first, followed by
SSVEP). Subjects were seated approximately 60cm from the
LCD screen (15.4” with resolution of 1280 x 800 pixels) of
the notebook running the BCI software. The entire session
took on average about 40 minutes per subject. Before each
session started, the experimenter explained to the subject the
task to perform.
B. Hardware
After completing a brief questionnaire each subject was
prepared for EEG recording. In order to reduce preparation
4202
FPZ
Fz
Cz
A2
P3 Pz P4
PO 7 PO8
Oz
Fig. 1. The electrode montage used in the current study. The 10 used
electrode sites are indicated by the dotted circles; the FP Z was used for
ground (circled in black) and right ear lobe A2 for the reference electrode
according to the system of electrode placement [18].
times at the beginning of the experiment, the minimum
necessary number of electrodes was used. Data were ref-
erenced to the right earlobe with a ground at site F P z. This
configuration was used for both approaches, but channels
used were utilized depending on the specific mental strategy
as follows: Fz , Cz , P3 , Pz , P4 , P O7 , Oz and P O8 for P300
and P3 , Pz , P4 , P O7 , Oz and P O8 for SSVEP. Figure 1
shows the electrode arrangement used in this study. Standard
abrasive electrolytic electrode gel was applied between the
electrodes and the skin in order to bring impedances below
5kΩ. An EEG amplifier g.USBamp (Guger Technologies,
Graz, Austria) was used and the sampling frequency was
128Hz for SSVEP and 256Hz for P300 approach. During
the EEG acquisition, an analog bandpass filter between 2
and 30Hz, and a notch filter around 50Hz (mains frequency
in Europe) were applied directly in the amplifier.
C. Experimental protocol
1) P300: The experiment started with the P300 session,
which consisted of the presentation of a 6 × 6 row-column
(RC) matrix containing letters and numbers (see Figure 2).
The individual rows and columns flashed rapidly in a random
sequence, such that a row or column was never followed by
the same row and column. In one sequence of intensifications
the target character flashed twice, once in a column and once
in a row. One character epoch consisted of 15 flashes of each
row and column. Between two characters there was an in-
terval of 6 seconds where the next character was highlighted
in the matrix so that the person was able to find it quickly.
This is an advantage for new users who are unfamiliar with
the P300 matrix. The rows and columns were intensified for
100ms with 60ms breaks between intensifications.
Subjects were instructed to silently count flashes that
included the letter or symbol that they wanted to select, while
ignoring other flashes. Only the row and column that contain
the desired letter evoked large P300 potentials. An initial
run consisted of spelling the word WATER. These data were
recorded and used to calibrate the system. The P300 system
 Start

Cursor: N

SNR
Target: N

A B C D E F
Threshold 170

SubjectNr 1
Accuracy: 100.00
Please spell
BRAIN Next word Word OK
ITR: 50.63 bpm

G H I J K L Clr L Del
F T J
M N O P Q R Q V M I O K .
, U A N E S D C
S T U V W X G P W R B Y -
X H Z
Y Z 0 1 2 3

4 5 6 7 8 9
BRAIN

Fig. 2. P300 paradigm presented on a LCD screen. Fig. 3. SSVEP speller application presented on a LCD screen.

TABLE I
D ETAILED RESULTS OVER 14 SUBJECTS .
collects EEG data for each flashing character and extracts an
interval 100 ms before the flash and 700 ms after the flash. Subject SSVEP paradigm (A) P300 paradigm (B) Preference
Then for each character a buffer is filled up with each flash. # Trials ACC ITR Trials ACC ITR Paradigm
After all characters were shown the EEG data is classified. A [-] [-] [%] [bpm] [-] [%] [bpm] [A, B]
linear discriminant analysis was applied to derive the EEG 1 34 100.00 57.98 13 100.00 10.77 A
weighting parameters for classification. The individualized 2 - - - 11 100.00 10.77 B
classifier was used for the next three runs where the user 3 - - - 12 83.33 7.64 B
was instructed to spell the words BCI and BRAIN (copy 4 88 77.27 12.72 15 93.33 9.32 A
5 - - - 13 84.62 7.84 B
spelling), and the third word was free and chosen by the
6 47 95.74 23.66 12 91.67 9.02 A
user (free spelling). Misspellings should be ignored, as far
7 35 100.00 31.09 13 69.23 5.63 A
as no correcting option is available in the current P300 matrix 8 - - - 14 100.00 10.77 B
(i.e. all letters of the word should appear in the spelled text). 9 54 87.04 15.22 12 83.33 7.64 B
The complete P300 software was provided by g.tec Medical 10 46 95.65 26.91 13 92.31 9.13 A
Engineering GmbH, Austria. This RC-system was already 11 55 87.27 23.95 14 85.71 8.01 A
tested within 81 subjects, 72.8% were able to spell with 12 52 96.15 20.43 14 92.86 9.23 A
100% accuracy (for more details about this study see [15]). 13 43 88.37 11.93 12 100.00 10.77 A
2) SSVEP: SSVEP patterns are elicited in a visual at- 14 37 91.89 24.36 12 100.00 10.77 A
Mean 49.10 91.94 24.83 12.86 91.17 9.09
tention task in which users focus their attention on light
S.D. 15.65 7.07 13.21 1.10 9.10 1.60
sources that flicker with frequencies above 5Hz. Figure 3
shows the display used for the evaluation of this paradigm.
structed to complete the second session following the P300
The centre of the display contained 32 characters (letters
session. The task was to spell exactly the same three words
and other special characters). At the beginning of each run,
as in the previous session. Misspellings should be corrected
a cursor was presented over the “E” character in the middle.
with the ‘Del’ option located at the top-right of the matrix.
Subjects spelled three words by focusing on one of the
five boxes presented on the screen. Each box contained
either an arrow (indicating left, right, up, or down) or the III. R ESULTS
word “Select”. Each box oscillated at a different constant
frequency: 7.5Hz, 8.57Hz, 10Hz, 12Hz and 6.67Hz to encode Table I shows the results from 14 subjects. The paradigms
the command: left, right, up, down and select, respectively. were compared by measuring number of trials, accuracy and
These frequencies, and the best arrangement of characters in information transfer rate. Trials correspond to the number
the display, were determined through prior work [19]. The of correct and incorrect selections. Accuracy was calculated
complete SSVEP system has been developed at the Institute as the ratio between the number of correct selections or
of Automation, University of Bremen and evaluated under hits and the sum of the number of targets or trials. The
real world conditions [16]. During this prior study, spelling information transfer rate (ITR) represented the amount of
tasks were successfully completed by 32 participants with information communicated per unit time. The number of
a mean accuracy of over 92% and an average information possible choices was different for each BCI approach. In the
transfer rate of 22.6 bits/min. EEG signals are processed P300 paradigm, the number of choices corresponded to the
to find signal-to noise ratios (SNR) every 100ms for each number of elements in the matrix (N = 36). In the SSVEP
frequency. The SSVEP detection uses a time window of 2 paradigm, N was the number of flickering boxes available
seconds of EEG data to classify the incoming signals. If (N = 5). Additionally, the last column of Table I shows the
the SNR at a specific frequency exceeded a predetermined answer to the question about preferred paradigm. Subjects
threshold, the corresponding command was executed. This had to choose one of the paradigms (A or B) based on their
system does not require any calibration. Subjects were in- personal experience.
4203
 IV. C ONCLUSIONS AND F UTURE WORK
A. Conclusions
Direct comparison can be done between paradigms A and
B because the words spelled were exactly the same. By
analyzing the number of trials it can be seen that the number
of commands required to complete the task was different
depending on the approach. For example, for the word “BCI”
the number of commands was 9 using the SSVEP BCI,
while using the P300 BCI the number of commands was
3. Accuracies between 77.27 and 100% were achieved for
the SSVEP paradigm, while with the P300 paradigm the
accuracy ranged between 69.23 and 100%.
The P300 information transfer rate was calculated without
considering the break between the single characters that
were spelled, because it was used to show the person the
position of the next character. Trained users can reduce this
time period significantly. It must also be mentioned that the
P300 speller used here was not optimized for speed and was
not individually adjusted to each person. We used the same
settings as in the study performed by Guger et al. [15] to
be able to compare the results (88.9% reached 80-100%
accuracy and 6.2% reached 60-79%). In this study 92.8%
were able to reach accuracy between 80-100% and 7.14%
reached an accuracy of 60-79%. This shows that the results
of both studies are very similar.
Four subjects exhibited weak SSVEP and therefore could
not communicate using the SSVEP-based BCI, while all
subjects could spell using the P300-based BCI. Subjects that
could not control the SSVEP system consequently chose the
P300 as favored approach, only one subject who was able to
spell with the SSVEP paradigm selected P300 as preferred
approach. ITR values for the SSVEP paradigm were higher
than for the P300 paradigm, which suggest it was the
reason why most subjects chose the SSVEP paradigm as the
preferred paradigm. The other reason could be the feedback
type - in case of SSVEP every command classification was
followed by the visual feedback (highlighting actual letter
background), additionally auditory signals played the name
of the received navigation command or selected character
simultaneously with cursor movements or letter selections.
For the P300 the character selection was performed silently
only after all characters flashed up 30 times (15 repetitions
for each row and column), this type of BCI gave feedback
much more rarely. Therefore, the results of this study need
further investigation with optimized speller applications and
with a larger number of subjects.
B. Future work
The further BCI research shall be extended to the deve-
lopment of friendly, straightforward wizard that will walk
each user through a series of tests to determine optimal
parameters. The comparatively shortened tests with the BCI
wizard will allow a rapid choice of the most suitable BCI
paradigm for each user. Nevertheless, training still plays
an important role for the brain-machine interaction. The
inclusion of the cognitive abilities of the users (subject can
4204
learn the time behavior of specific BCI application and could
react foresighted, what is definitely needed e.g. for play-
ing pinball with non-invasive BCI [20]) could considerably
shorten the response times of the BCI system. Additional
tests should be done to analyze the possible correlations
between different BCI paradigms (e.g. high ITR with SSVEP
BCI could correspond to a good response to the ssMRPs and
vice versa).
R EFERENCES
[1] G. Dornhege, J. del R. Millan, T. Hinterberger, D. J. McFarland, and
K.-R. Müller, Toward Brain-Computer Interfacing. MIT Press, 2007.
[2] N. Birbaumer, “Slow cortical potentials: Plasticity, operant control,
and behavioral effects,” Neuroscientist, vol. 5, no. 2, pp. 74–78, 1999.
[3] F. Guo, B. Hong, X. Gao, and S. Gao, “A Brain-Computer Interface
using motion-onset visual evoked potential,” J. Neural Eng., vol. 5,
no. 4, pp. 477–485, 2008.
[4] S. Kanoh, K.-I. Miyamoto, and T. Yoshinobu, “A brain-computer
interface (BCI) system based on auditory stream segregation,” in Proc.
30th Annual International Conference of the IEEE Engineering in
Medicine and Biology Society EMBS 2008, 20–25 Aug. 2008, pp.
642–645.
[5] J. Guo, B. Hong, F. Guo, X. Gao, and S. Gao, “An Auditory BCI Using
Voluntary Mental Response,” in Proc. 4th International IEEE/EMBS
Conference on Neural Engineering NER 09, May 2009, pp. 455–458.
[6] K. Nazarpour, P. Praamstra, R. C. Miall, and S. Sanei, “Steady-state
movement related potentials for brain-computer interfacing,” IEEE
Trans. Biomed. Eng., vol. 56, no. 8, pp. 2104–2113, Aug 2009.
[7] O. Friman, I. Volosyak, and A. Gräser, “Multiple channel detection
of steady-state visual evoked potentials for brain-computer interfaces,”
IEEE Trans. Biomed. Eng., vol. 54, no. 4, pp. 742–750, Apr. 2007.
[8] Y. Wang, X. Gao, B. Hong, C. Jia, and S. Gao, “Brain-computer
interfaces based on visual evoked potentials,” IEEE Eng. Med. Biol.
Mag., vol. 27, no. 5, pp. 64–71, 2008.
[9] U. Hoffmann, J.-M. Vesin, T. Ebrahimi, and K. Diserens, “An effi-
cient P300-based brain-computer interface for disabled subjects,” J.
Neurosci. Methods, vol. 167, no. 1, pp. 115–125, Jan 2008.
[10] G. Pfurtscheller and F. Lopes da Silva, “Event-related EEG/MEG
synchronization and desynchronization: Basic principles,” Clin. Neu-
rophysiol., vol. 110, pp. 1842–1857, 1999.
[11] N. Birbaumer and L. G. Cohen, “Brain-computer interfaces: commu-
nication and restoration of movement in paralysis,” J. Physiol., vol.
579, no. 3, pp. 621–636, 2007.
[12] T. Hinterberger et al., Brain-Computer Interfaces for Communication
in Paralysis: A Clinical Experimental Approach. Cambridge, Mas-
sachusetts: The MIT Press, 2007, ch. 3, pp. 41–64.
[13] A. Nijholt and D. Tan, “Brain-Computer Interfacing for Intelligent
Systems,” IEEE Intell. Syst., vol. 23, no. 3, pp. 72–79, 2008.
[14] C. Guger, G. Edlinger, W. Harkam, I. Niedermayer, and G. Pfurtschel-
ler, “How Many People are Able to Operate an EEG-Based Brain-
Computer Interface (BCI)?” IEEE Trans. Neural Syst. Rehabil. Eng.,
vol. 11, no. 2, pp. 145–147, Jun 2003.
[15] C. Guger, S. Daban, E. Sellers, C. Holzner, G. Krausz, R. Carabalona,
F. Gramatica, and G. Edlinger, “How many people are able to control
a P300-based brain-computer interface (BCI)?” Neurosci. Lett., vol.
462, no. 1, pp. 94–98, Jun 2009.
[16] I. Volosyak, H. Cecotti, D. Valbuena, and A. Gräser, “Evaluation of the
Bremen SSVEP based BCI in real world conditions,” in Proc. IEEE
ICORR’09, Jun. 2009, pp. 322–331.
[17] B. Dal Seno, M. Matteucci, and L. Mainardi, “The Utility Metric: A
Novel Method to Assess the Overall Performance of Discrete Brain-
Computer Interfaces,” IEEE Trans. Neural Syst. Rehabil. Eng., vol. 18,
pp. 20–28, 2010.
[18] R. Oostenveld and P. Praamstra, “The five percent electrode system
for high-resolution EEG and ERP measurements,” Clin. Neurophysiol.,
vol. 112, no. 4, pp. 713–719, Apr 2001.
[19] I. Volosyak, H. Cecotti, and A. Gräser, “Impact of Frequency Selection
on LCD Screens for SSVEP Based Brain-Computer Interfaces,” in
IWANN 2009, Part I, LNCS 5517. Springer, 2009, pp. 706–713.
[20] M. Tangermann, M. Krauledat, K. Grzeska, M. Sagebaum, C. Vi-
daurre, B. Blankertz, and K.-R. Müller, “Playing Pinball with Non-
Invasive BCI,” in Neural Information Processing Systems (NIPS),
2009.