Use of a Green Familiar Faces Paradigm Improves P300-Speller Brain-Computer Interface Performance
Use of a Green Familiar Faces Paradigm Improves P300-Speller Brain-Computer Interface Performance
Qi Li 0 1
Shuai Liu 0 1
Jian Li 0 1
Ou Bai 0 1
0 Academic Editor: Nader N. Pouratian , UCLA, UNITED STATES
1 1 School of Computer Science and Technology, Changchun University of Science and Technology , Changchun , China , 2 Hequ Hydrologic Station, Middle Reaches Hydrology and Water Resources Bureau of Yellow River Conservancy Commission , Jinzhong , China , 3 School of Engineering, Virginia Commonwealth University , Richmond , United States of America
Combining the chromatic property (green color) and the FF spelling paradigm led to better
classification accuracy and an increased information transfer rate. These findings
demonstrate a promising new approach for improving the performance of the P300-speller.
Competing Interests: The authors have declared
that no competing interests exist.
The brain-computer interface (BCI) provides an alternative communication channel that is
independent of muscular control [1–4]. Non-invasive BCI systems commonly utilize
electroencephalography (EEG) signals from the scalp to control external computers or machines, which
have been found to be particularly useful for patients with amyotrophic lateral sclerosis (ALS)
and locked-in state (LIS) [5–6].
The P300-speller system is one of the most commonly used non-invasive BCI systems; its
name derives from the fact that it mainly relies on P300 and other event-related potentials
(ERPs) [7–8]. Through the P300-speller system, users can communicate a specific character by
attending to the cell of the matrix that contains the desired character (target character) and
counting the number of times it is intensified (or flashed) . However, performance of the
P300-speller BCI system is not yet satisfactory due to its low classification accuracy and
information transfer rate (ITR) .
A large amount of research has been conducted to improve the performance of the
P300-speller system by optimizing the signal-processing algorithm and developing novel
classification techniques, such as step-wise discriminant analysis, wavelets, and support vector
machines [11–13]. In addition, some researchers have tried to improve the performance of this
system by optimizing the physical properties of the spelling paradigm, such as the matrix size
, the stimulation frequency , the inter-stimulus interval (ISI) [9, 15], the stimulation
intensity , and other factors [17–18].
Recent research has focused on the manipulation of spelling paradigm stimuli to increase
other ERP components that occur before or after the P300 potential for the purpose of
enhancing the difference between the attended and ignored characters . For example,
Kaufmann et al. (2011) superimposed a row or column of the P300-speller with translucent
pictures of familiar faces (Albert Einstein or Ernesto “Che” Guevara) (FF spelling paradigm)
and found its performance was markedly superior to the conventional P300-speller system
and the 'flash only' P300-speller system, as it evoked additional N170 and N400f ERPs [10,
20]. Some studies showed that the N170 occurs between 130 and 200 ms post-stimulus and
has been associated with the pre-categorical structural encoding of faces [21–22]. The N400f
occurs between 300 and 500 ms post-stimulus at the parietal and central electrode sites, and
is related to the familiarity component of face recognition [19, 23]. As this technology
advanced, many researchers attempted to further optimize the FF spelling paradigm to
improve the performance of the P300-speller system. For example, a facial expression
changes paradigm was developed to decrease adjacent interference , and a multi-faces
paradigm was used to decrease repetition . The FF spelling paradigm was also tested in
patients and demonstrated good performance . In addition, Takano et al. (2009) found
that the color of the stimuli could also influence P300-speller system performance. They
replaced the white/gray flicker matrix with a green/blue flicker matrix and found that this
chromatic stimulus improved the performance of the P300-speller system . Therefore, we
hypothesized that combining the chromatic property and the FF spelling paradigm may lead
to a better classification and ITR.
In the present study, we proposed a new spelling paradigm in which the flashing row or
column is overlaid with translucent green pictures of familiar faces (GFF spelling paradigm). We
analyzed the elicited ERP waveforms induced by the FF spelling paradigm and by the proposed
GFF spelling paradigm, and compared P300-speller BCI system performance between the two
The study comprised one offline and one online experiment. Seventeen university students (6
female students; age, 21–26 years; mean age: 24.6 years) participated in the offline experiment,
and 12 of these also participated in the online experiment. The participants did not have any
known neurological disorders, and had normal or corrected-to-normal vision. After receiving
a full explanation of the purpose and risks of the study, participants signed a written informed
consent and were paid 50 RMB per experiment. The individual whose facial photograph is
shown in Fig 1 provided written informed consent (as outlined in PLOS consent form) for
publication of the photograph. The study was approved by the ethics committee of Changchun
University of Science and Technology (CUST). All participants were native Chinese speakers,
but were familiar with the Western characters used in the display.
The spelling paradigms
We designed two P300-speller spelling paradigms based on the conventional P300-speller
spelling paradigm. In each paradigm, 36 spelling characters were presented in a 6 × 6 matrix
subtended 13.4°×19.4° (24 × 16.5 cm) visual angle on a 19-in screen at a refresh rate of 60 Hz
(Fig 1). The size of each character was 1.2° × 1.2° (1.5 × 1.5 cm). The distance between each
character in was 3.7° × 2.5° (4.5 × 3 cm). The rows and columns of the matrix were flashed
consecutively in pseudorandom order. In the first paradigm, the rows or columns of the characters
were covered with translucent pictures of a familiar face (David Beckham) while they were
flashed (FF spelling paradigm, Fig 1a). The ISI was set to 250 ms, in which each character was
changed to the face picture for 200 ms, and then reverted to gray characters for 50 ms. The
second spelling paradigm was similar to the first, but the translucent pictures of the familiar faces
were painted green (GFF spelling paradigm, Fig 1b). We used the same brightness (20 cd/cm)
and contrast to prevent these parameters from affecting the results.
Each subject sat in a comfortable chair approximately 70 cm from the front of the computer
monitor. Subjects were asked to focus on the target character, avoid blinking during stimulus
presentation, and silently count the number of target character flashes. In the offline
experiment, each spelling paradigm was conducted six times with different five-character words, and
each was considered a separate session. Each session consisted of five runs, each of which
involved a different target character. One flash of a row or column was referred to as a trial.
The flash of a row or column including the target character was defined as a target trial, and
the flash of a row or column without the target character was defined as a non-target trial. A
sequence consisted of 12 flashes (trails), six from the rows and six from the columns. In each
run, the sequence was repeated 15 times (Fig 2a). Thus, each run consisted of 180 flashes of
row or column to output a target character. Participants never received feedback. The sessions
of the two paradigms were conducted alternately to control for potential habituation effects.
Participants were allowed to take a 5-min break between sessions. To avoid novelty effects, the
stimulus image in each spelling paradigm was presented to subjects for 20 s prior to each
The online experiment was implemented in a different day. Each spelling paradigm was
conducted one time as a separate session. Each session consisted of 30 runs, each of which
involved a different target character. The number of sequences per run was two. Feedback
regarding spelling correctness was provided to the subjects after each of the runs (Fig 2b).
Fig 1. Two different spelling paradigms were designed and employed in this study. Translucent pictures
of a familiar face (David Beckham) covered the characters in one row or column while it was intensified. (a) In
the FF spelling paradigm, the characters were covered with flashing familiar faces. (b) In the GFF spelling
paradigm, the characters were covered with flashing green familiar faces. The facial photographs of David
Beckham are replaced by those of a subject in the figure because of the lack of a print license. The individual in
the photograph has given permission for his photograph to be published.
Fourteen-channel (Fz, F3, F4, FC1, FC2, Cz, C3, C4, Pz, P3, P4, Oz, O1, and O2; Fig 3)
electroencephalogram (EEG) data were recorded with the left mastoid as the ground and the right
mastoid as the reference. Horizontal eye movements were measured by deriving the
electrooculogram (EOG) from a pair of horizontal EOG (HEOG) electrodes placed at the outer canthi of
the left and right eyes. Vertical eye movements and eye blinks were detected by deriving an
EOG signal from a pair of vertical EOG (VEOG) electrodes placed approximately one
centimeter above and below the subject’s left eye. The impedance was maintained below 5 kO. All
signals were band-pass filtered at 0.1–100 Hz, amplified with a NeuroScan amplifier (SynAmps 2,
Fig 2. Experimental arrangement of each spelling paradigm in the offline (a) and online (b) experiments.
NeuroScan Inc., Abbotsford, Australia), and digitized at a rate of 250 Hz. Stimulus presentation
was controlled by a personal computer running Presentation 0.71 software (Neurobehavioral
Systems Inc., Albany, NY, USA). Data acquisition was conducted using Scan4.5 software
Event-related potentials (ERP) processing
Before the offline classification, we compared the difference waveforms of ERPs elicited by the
target and non-target trials in the FF spelling paradigm with those of the GFF spelling
EEG data were digitally filtered using a band-pass filter of 0.01–30 Hz and were corrected
for ocular artifacts using a regression analysis algorithm . EEG signals were divided into
epochs from 100 ms before the onset of each trial to 800 ms after the onset, and baseline
corrections were made against -100–0 ms. The ERP data were averaged for each trial type (target,
non-target trials). The grand-averaged ERP data were obtained from all participants for each
trial type in the two spelling paradigms. The difference waveform
(ERPTarget—ERPNon-target) was computed by subtracting ERP waveforms elicited by non-target trials from those
elicited by target trails in both FF and GFF spelling paradigms.
The mean amplitudes were calculated for all electrodes at consecutive 20 ms windows
between stimulus onset and 800 ms after stimulus presentation, and the data were then
analyzed using ANOVA with the within-subjects factors of spelling paradigm (FF, GFF spelling
paradigm), time window (40 levels), and electrodes (14 levels). The Greenhouse–Geisser
Epsilon correction was applied to adjust the degrees of freedom of the F ratios, if necessary. In
order to determine the electrodes and time periods in which there was a significant difference
between the two spelling paradigms, a multiple comparison was conducted with the
withinsubjects factors of 2 spelling paradigms (FF and GFF spelling paradigms) × 40 time
windows × 14 electrodes. All statistical analyses were conducted using the SPSS version 19.0
software package (SPSS Inc., Chicago, IL, USA).
Bayesian linear discriminant analysis (BLDA) was used to classify the EEG data. BLDA is an
extension of Fisher's linear discriminant analysis (FLDA) that avoids over fitting, which has be
demonstrated to obtain very good classification performance in familiar faces P300-speller BCI
applications [8, 19, 24]. The details of the algorithm can be found in . We used six-fold
cross-validation to calculate the individual accuracy in the offline experiment (i.e., we
sequentially chose one of the six sessions as the test session and obtained six different training and test
session groups; the accuracy of each of the six groups was computed; the individual accuracy of
each participant was obtained by averaging the six results). Data acquired offline were used to
train the classifier using BLDA and obtain the classifier model. This model was then used in
the online experiment. If there was a tie between multiple characters, the classifier would
automatically select the last output as the target character.
Information transfer evaluation
Information transfer rate (ITR) is generally used to evaluate the communication performance
of a BCI system and is a standard measure that accounts for accuracy, the number of possible
selections, and the time required to make each selection [4, 20, 29]. For a sequence with N
possible choices in which each choice has an equal probability of selection by the user, the
probability (P) that the desired choice will indeed be selected remains invariant, and each error
choice has the same probability of selection, the ITR (bits min-1) can be calculated as
where M denotes the number of commands per minute.
Fig 4 displays the superimposition of grand-averaged ERP waveforms elicited by non-target
trials and target trials in the FF and GFF spelling paradigms.
In both paradigms, a negative ERP component was observed between 150 and 250 ms at the
temporal occipital area for target trials, and its amplitude peaked at the O2 electrode at around
196 ms (-1.974μV) in the FF spelling paradigm, and at around 192 ms (-2.659 μV) in the GFF
spelling paradigm. The distribution was slightly asymmetric insofar as the right was larger than
the left. A positive ERP component was observed between 180 and 380 ms in the frontal area
for target trials, and its amplitude peaked at the Fz electrode at around 236 ms (4.185 μV) and
232 ms (5.263 μV) in the FF and GFF spelling paradigm, respectively. The third ERP
component was found between 300 and 450 ms for target trials and was a positive ERP component.
The amplitude peaked at the Pz electrode at around 372 ms (2.951 μV) in the FF spelling
paradigm and at around 352 ms (3.858 μV) in the GFF spelling paradigm.
A greater difference between target and non-target trials would make their classification
easier. Therefore, the ERP waveforms elicited by the non-target trials were subtracted from those
elicited by the target trials (ERPTarget—ERPNon-target) for the FF and GFF spelling
paradigms (Fig 5a). Although the difference waveforms (ERPTarget—ERPNon-target) were similar
between the two paradigms, differences could be observed. The statistical differences in the
amplitudes measured at 14 electrode sites from 0 to 800 ms during the FF and GFF spelling
paradigms were determined using a multiple comparison analysis. Statistically significant
differences were found during the following four time periods: (1) 160–220 ms at the left occipital
Fig 4. Superimposed grand-averaged ERP waveforms elicited by non-target and target trials in the FF and GFF spelling paradigms. The epochs are
from 100 ms before stimulus onset to 800 ms after onset.
area, (2) 160–260 ms at the frontal-central area, (3) 300–400 ms at the frontal-central area, and
(4) from 640–680 ms at the frontal-central area (Fig 5a). The amplitudes of
(ERPTarget—ERPNon-target) were higher for the GFF spelling paradigm than for the FF spelling paradigm. Fig
5b depicts the scalp topographies for double-difference waveforms obtained by subtracting the
(ERPTarget—ERPNon-target) waveforms in the FF spelling paradigm from those in the GFF
spelling paradigm for the four time periods showing significant differences.
The grand-averaged ERP waveforms in Fig 5 did not adequately reflect the difference
between the spelling paradigms within individual subjects. As individual differences are crucial
in the BCI system, we compared the averaged amplitudes of (ERPTarget—ERPNon-target) in
the four significant time periods (160–220 ms at O1, 160–260 ms at Fz, 300–400 ms at Cz, and
640–680 ms at Fz) for 17 individual subjects (Fig 6). In most subjects, the mean amplitudes in
the four significant time periods were significantly larger in the GFF spelling paradigm
compared to the FF spelling paradigm.
Offline classification results
As shown in Fig 5, highly significant differences for the ERPTarget—ERPNon-target
waveforms between the GFF and FF spelling paradigms were found during the 160–220 ms, 160–
260 ms, 300–400 ms, and 640–680 ms time periods. Therefore, we used the 160–688 ms time
window at 14 electrodes as the classification epoch in order to reduce the computational time.
Fig 5. (a) Superimposed difference waveforms of ERPs elicited by the target and non-target trials
(ERPTarget—ERPNon-target) in the FF and GFF spelling paradigms. The gray square areas indicate the time
periods during which the difference waveforms of ERPs elicited by the target and non-target trials (ERPTarget—
ERPNon-target) were significantly different (p < 0.01) between the FF and GFF spelling paradigms. (b) Scalp
topographies for double-difference waveforms obtained by subtracting the (ERPTarget—ERPNon-target)
waveforms for the FF spelling paradigm from those for the GFF spelling paradigm for the time periods showing
significant differences (160–220 ms, 160–260 ms, 300–400 ms, and 640–680 ms).
First, the original EEG data were filtered between 0.1 and 30 Hz using a third-order
Butterworth band pass filter. The EEG was then down-sampled from 250 Hz to 62.5 Hz by selecting
every four samples from the epoch. Because we used 14 channels, the size of the feature vector
was 14 × 33 (14 channels by 33 time points).
Results of individual and average accuracies of the P300-speller for 17 subjects in both
spelling paradigms are shown in Fig 7. The analysis of the 17 subjects indicated that accuracy
increased with sequence number in both paradigms. The average classification accuracy of the
P300-speller was greater in the GFF spelling paradigm than in the FF spelling paradigm for 1–9
sequences. Fig 8 illustrates the individual and average ITRs of the P300-speller for 17 subjects
Fig 6. The comparison of averaged amplitudes of (ERPTarget—ERPNon-target) in four significant time periods (160–220 ms at O1, 160–260 ms at Fz,
300–400 ms at Cz, and 640–680 ms at Fz) for 17 individual subjects.
in the FF and GFF spelling paradigms. The average ITR of the P300-speller was higher in the
GFF spelling paradigm than in the FF spelling paradigm for 1–9 sequences. In addition, we
counted the number of sequences needed for subjects to achieve a 100% and 70% accuracy
level in both spelling paradigms. A level of 70% may be regarded as a minimum level of
communication [30–31]. Regardless of whether the accuracy level was 100% (p < 0.005) or
70% (p < 0.01), the results of the paired t-test indicated that the number of sequences was
significantly reduced in the GFF spelling paradigm. The subjects required 2.82 ± 0.27
(mean ± standard deviation) sequences to achieve the goal of 100% classification accuracy in
the GFF spelling paradigm, whereas 4.65 ± 0.61 sequences were needed to achieve the same
goal in the FF paradigm. A level of 70% was achieved with 1.71 ± 0.21 stimulus sequences
in the GFF spelling paradigm, whereas the subjects achieved this goal with 2.53 ± 0.31
stimulus sequences in the FF spelling paradigm.
Online classification results
Table 1 shows the online classification accuracy and ITR for each subject by using two
sequences. The best performance, with an accuracy of 96.7% and an ITR of 48.2 bits min-1,
was yielded by the GFF paradigm. The results of the paired t-tests showed that the accuracy
(p < 0.05) and ITR (p < 0.01) was significantly different between the two spelling paradigms.
The mean classification accuracy in the GFF paradigm was 10.5% higher than that of the FF
Fig 7. Individual and average accuracies of the P300-speller for 17 subjects in the FF and GFF spelling paradigms.
In the present study, we assessed grand-averaged ERP waveforms elicited by target trials in
both the FF and GFF spelling paradigms. In addition, we analyzed the difference waveforms of
ERPs elicited by the target and non-target trials (ERPTarget—ERPNon-target), and compared
the offline and online classification performance of the P300-speller in the FF and proposed
GFF spelling paradigms.
In both paradigms, a negative ERP component elicited by the target trials was found at
around 150–250 ms at the temporal occipital area (Fig 4). This ERP component is similar to
N170, which is involved in face recognition [22, 32–34]. Consistent with a previous study, the
Fig 8. Individual and average ITRs of the P300-speller for 17 subjects in the FF and GFF spelling paradigms.
mean amplitude of the N170 at the O2 electrode was stronger than that at the O1 electrode,
which was attributed to the right hemisphere advantage . In addition, a positive ERP
component was observed at around 180–380 ms at the frontal area (Fig 4), which may represent the
vertex positive potential (VPP), a potential implicated in face-sensitive brain responses
reflecting the neural processing of faces . Another positive ERP component between 300 and 450
ms was found at the parietal area (Fig 4), which may well represent the expected P300 .
The performance of the P300-speller system could be improved by enhancing the difference
between target trials and non-target trials . Therefore, the (ERPTarget—ERPNon-target)
difference waveform was computed by subtracting ERP waveforms elicited by non-target trials
from ERP waveforms elicited by target trails in the FF and GFF spelling paradigms (Fig 5a).
Our results indicated four statistically significant differences between the two spelling
paradigms, and the amplitudes of (ERPTarget—ERPNon-target) in the GFF spelling paradigm
31.6 ± 8.1
39.0 ± 5.0
Avg. ± SD
75.6 ± 12.6
86.1 ± 6.3
t = -2.969; p < 0.05
t = -3.139; p < 0.01
were significantly larger than those in the FF spelling paradigm. The first three time periods
with significant differences were respectively consistent with those of the N170, VPP, and
P300. Previous studies showed that the presentation of novel faces, such as inverted faces,
could elicit a larger N170 and VPP, thereby improving the performance of the P300-speller
system [19–20]. The green familiar faces shown in the GFF spelling paradigm were more novel for
the participants, which increased the amplitude of the N170 and VPP, enabling improved
classification relative to the FF spelling paradigm. In addition, many previous studies have reported
that different colors are associated with different arousal levels. Compared to low-arousal
stimuli, high-arousal stimuli produced larger P300 amplitudes [35–37]. The arousal level
associated with green is relatively high , which may have accounted for the increased P300
amplitude in the GFF spelling paradigm compared to the FF spelling paradigm. The fourth
significant difference with the time period of 640–680 ms was likely to originate from the
influence of the P600f, an ERP component demonstrated to be related to the processes involved in
the recollection of familiar faces [22, 39]. Our results indicated that the GFF spelling paradigm
required more recollection of faces, which led to the amplitude difference in P600f between the
two spelling paradigms.
Contrary to previous findings, our results indicated that familiar faces did not elicit the
N400f, a strong negative component observed at 300–500 ms post-stimuli [10, 19]. One
possible explanation for this finding is that the faces (i.e., David Beckham) were not familiar enough
to the participants. A recent study demonstrated that the N400f was best elicited by pictures of
family members . In addition, as the N400f overlaps with the P300, the N400f may have
also been canceled out by the P300.
Based on the analysis of the ERP components, we selected the 160–688 ms time window as
the classification epoch. As expected, both offline and online results indicated that the GFF
spelling paradigm obtained significantly higher classification accuracies and ITRs than the FF
spelling paradigm. The offline and online results testified that the GFF spelling paradigm
induced larger ERP components, resulting in the improvement of P300-speller performance.
Generally, ITR is used as an important statistical metric for a BCI system [4, 20, 29]. The
ITR depends on both classification accuracy and speed of character selection. Moreover, the
speed is based on the number of stimulus sequence used for averaging and ISI [15, 40]. The
reduction in the number of stimulus sequences could shorten the speed of character selection,
but this reduction inevitably decreases the signal-to-noise ratio and thus typically entails a
decrease in classification accuracy. Our individual results indicated that the ITRs were maximal
in some subjects when the stimulus sequence was repeated twice. In addition, decreasing ISI
would result in less time for character selection, but a decrease in ISI leads to smaller P300
amplitudes and larger latencies, which would decrease classification accuracy. Therefore,
classification accuracy and speed of character selection must be weighed for obtaining higher ITR in
the design of the BCI system. A recent study reported that different spelling paradigms might
require different ISIs . Future studies should examine further the effects of ISI in the GFF
This study investigated whether the GFF spelling paradigm would lead to better P300-speller
performance compared to the FF spelling paradigm. Our results indicated a highly significant
improvement in the GFF spelling paradigm. This optimization may have a significant impact
on increasing the communication speed and accuracy of the P300-speller. The performance of
the P300-speller does not depend solely on one attribute; rather, multiple factors are at play.
Therefore, further research is necessary to determine the influence of multiple attributes on
The authors would like to thank all individuals who participated in our study.
Conceived and designed the experiments: QL SL. Performed the experiments: SL JL. Analyzed
the data: QL SL JL. Contributed reagents/materials/analysis tools: SL JL. Wrote the paper: QL
9. Farwell LA, Donchin E. Talking off the top of your head: toward a mental prosthesis utilizing event
related brain potentials. Electroencephalogr Clin Neurophysiol. 1988; 70(6):510–23. PMID: 2461285.
1. Birbaumer N , Ghanayim N , Hinterberger T , Iversen I , Kotchoubey B , Kubler A , et al. A spelling device for the paralysed . Nature . 1999 ; 398 ( 6725 ): 297 - 8 . doi: 10.1038/18581 PMID: 10192330
2. Kubler A , Kotchoubey B , Kaiser J , Wolpaw JR , Birbaumer N. Brain-computer communication: unlocking the locked in . Psychol Bull . 2001 ; 127 ( 3 ): 358 - 75 . PMID: 11393301 .
3. Scherer R , Muller GR , Neuper C , Graimann B , Pfurtscheller G. An asynchronously controlled EEGbased virtual keyboard: improvement of the spelling rate . IEEE Trans Biomed Eng . 2004 ; 51 ( 6 ): 979 - 84 . doi: 10.1109/TBME. 2004 .827062 PMID: 15188868.
4. Wolpaw JR , Birbaumer N , McFarland DJ , Pfurtscheller G , Vaughan TM . Brain-computer interfaces for communication and control . Clin Neurophysiol . 2002 ; 113 ( 6 ): 767 - 91 . PMID: 12048038 .
5. Kubler A , Neumann N. Brain-computer interfaces-the key for the conscious brain locked into a paralyzed body . Prog Brain Res . 2005 ; 150 : 513 - 25 . doi: 10.1016/ S0079-6123(05)50035-9 PMID: 16186045.
6. Nijboer F , Sellers EW , Mellinger J , Jordan MA , Matuz T , Furdea A , et al. A P300-based brain-computer interface for people with amyotrophic lateral sclerosis . Clin Neurophysiol . 2008 ; 119 ( 8 ): 1909 - 16 . doi: 10.1016/j.clinph. 2008 .03.034 PMID: 18571984; PubMed Central PMCID: PMC2853977.
7. Bernat E , Shevrin H , Snodgrass M. Subliminal visual oddball stimuli evoke a P300 component . Clin Neurophysiol . 2001 ; 112 ( 1 ): 159 - 71 . PMID: 11137675 .
8. Jin J , Allison BZ , Zhang Y , Wang X , Cichocki A. An ERP-based BCI using an oddball paradigm with different faces and reduced errors in critical functions . International journal of neural systems . 2014 ; 24 ( 8 ): 1450027 . doi: 10.1142/S0129065714500270 PMID: 25182191 .
10. Kaufmann T , Schulz SM , Grunzinger C , Kubler A. Flashing characters with famous faces improves ERP-based brain-computer interface performance . J Neural Eng . 2011 ; 8 ( 5 ): 056016 . doi: 10.1088/ 1741 - 2560 /8/5/056016 PMID: 21934188.
11. Krusienski DJ , Sellers EW , Cabestaing F , Bayoudh S , McFarland DJ , Vaughan TM , et al. A comparison of classification techniques for the P300 Speller . J Neural Eng . 2006 ; 3 ( 4 ): 299 - 305 . doi: 10.1088/ 1741 - 2560 /3/4/007 PMID: 17124334.
12. Krusienski DJ , Sellers EW , McFarland DJ , Vaughan TM , Wolpaw JR . Toward enhanced P300 speller performance . J Neurosci Methods . 2008 ; 167 ( 1 ): 15 - 21 . doi: 10.1016/j.jneumeth. 2007 .07.017 PMID: 17822777; PubMed Central PMCID: PMC2349091.
13. Blankertz B , Lemm S , Treder M , Haufe S , Muller KR. Single-trial analysis and classification of ERP components-a tutorial . Neuroimage . 2011 ; 56 ( 2 ): 814 - 25 . doi: 10.1016/j.neuroimage. 2010 .06.048 PMID: 20600976.
14. Allison BZ , Pineda JA . ERPs evoked by different matrix sizes: implications for a brain computer interface (BCI) system . IEEE Trans Neural Syst Rehabil Eng . 2003 ; 11 ( 2 ): 110 - 3 . doi: 10.1109/TNSRE. 2003 .814448 PMID: 12899248.
15. Sellers EW , Krusienski DJ , McFarland DJ , Vaughan TM , Wolpaw JR . A P300 event-related potential brain-computer interface (BCI): the effects of matrix size and inter stimulus interval on performance . Biol Psychol . 2006 ; 73 ( 3 ): 242 - 52 . doi: 10.1016/j.biopsycho. 2006 .04.007 PMID: 16860920.
16. Polich J , Ellerson PC , Cohen J. P300, stimulus intensity, modality, and probability. Int J Psychophysiol . 1996 ; 23 ( 1-2 ): 55 - 62 . PMID: 8880366 .
17. Allison BZ , Pineda JA . Effects of SOA and flash pattern manipulations on ERPs, performance, and preference: implications for a BCI system . Int J Psychophysiol . 2006 ; 59 ( 2 ): 127 - 40 . doi: 10.1016/j. ijpsycho. 2005 .02.007 PMID: 16054256.
18. Salvaris M , Sepulveda F. Visual modifications on the P300 speller BCI paradigm . J Neural Eng . 2009 ; 6 ( 4 ): 046011 . doi: 10.1088/ 1741 - 2560 /6/4/046011 PMID: 19602731.
19. Jin J , Allison BZ , Kaufmann T , Kubler A , Zhang Y , Wang X , et al. The changing face of P300 BCIs: a comparison of stimulus changes in a P300 BCI involving faces, emotion, and movement . PLoS One . 2012 ; 7 ( 11 ) :e49688 . doi: 10.1371/journal.pone.0049688 PMID: 23189154; PubMed Central PMCID: PMC3506655.
20. Zhang Y , Zhao Q , Jin J , Wang X , Cichocki A. A novel BCI based on ERP components sensitive to configural processing of human faces . J Neural Eng . 2012 ; 9 ( 2 ): 026018 . doi: 10.1088/ 1741 - 2560 /9/2/ 026018 PMID: 22414683.
21. Bentin S , Allison T , Puce A , Perez E , McCarthy G . Electrophysiological Studies of Face Perception in Humans. J Cogn Neurosci . 1996 ; 8 ( 6 ): 551 - 65 . doi: 10.1162/jocn. 1996 .8.6. 551 PMID: 20740065; PubMed Central PMCID: PMC2927138.
22. Eimer M. Event-related brain potentials distinguish processing stages involved in face perception and recognition . Clin Neurophysiol . 2000 ; 111 ( 4 ): 694 - 705 . PMID: 10727921 .
23. Touryan J , Gibson L , Horne JH , Weber P. Real-time measurement of face recognition in rapid serial visual presentation . Front Psychol . 2011 ; 2 :42. doi: 10.3389/fpsyg.2011.00042 PMID: 21716601; PubMed Central PMCID: PMC3110906.
24. Jin J , Daly I , Zhang Y , Wang X , Cichocki A. An optimized ERP brain-computer interface based on facial expression changes . J Neural Eng . 2014 ; 11 ( 3 ): 036004 . doi: 10.1088/ 1741 - 2560 /11/3/036004 PMID: 24743165.
25. Kaufmann T , Schulz SM , Koblitz A , Renner G , Wessig C , Kubler A. Face stimuli effectively prevent brain-computer interface inefficiency in patients with neurodegenerative disease . Clin Neurophysiol . 2013 ; 124 ( 5 ): 893 - 900 . doi: 10.1016/j.clinph. 2012 .11.006 PMID: 23246415.
26. Takano K , Komatsu T , Hata N , Nakajima Y , Kansaku K. Visual stimuli for the P300 brain-computer interface: a comparison of white/gray and green/blue flicker matrices . Clin Neurophysiol . 2009 ; 120 ( 8 ): 1562 - 6 . doi: 10.1016/j.clinph. 2009 .06.002 PMID: 19560965.
27. Semlitsch HV , Anderer P , Schuster P , Presslich O. A solution for reliable and valid reduction of ocular artifacts , applied to the P300 ERP. Psychophysiology . 1986 ; 23 ( 6 ): 695 - 703 . PMID: 3823345 .
28. Hoffmann U , Vesin JM , Ebrahimi T , Diserens K. An efficient P300-based brain-computer interface for disabled subjects . J Neurosci Methods . 2008 ; 167 ( 1 ): 115 - 25 . doi: 10.1016/j.jneumeth. 2007 .03.005 PMID: 17445904.
29. McFarland DJ , Sarnacki WA , Wolpaw JR . Brain-computer interface (BCI) operation: optimizing information transfer rates . Biol Psychol . 2003 ; 63 ( 3 ): 237 - 51 . PMID: 12853169 .
30. Kubler A , Birbaumer N. Brain-computer interfaces and communication in paralysis: extinction of goal directed thinking in completely paralysed patients? Clin Neurophysiol . 2008 ; 119 ( 11 ): 2658 - 66 . doi: 10. 1016/j.clinph. 2008 .06.019 PMID: 18824406; PubMed Central PMCID: PMC2644824.
31. Kubler A , Neumann N , Kaiser J , Kotchoubey B , Hinterberger T , Birbaumer NP . Brain-computer communication: self-regulation of slow cortical potentials for verbal communication . Arch Phys Med Rehabil . 2001 ; 82 ( 11 ): 1533 - 9 . PMID: 11689972.
32. Rossion B , Curran T , Gauthier I. A defense of the subordinate-level expertise account for the N170 component . Cognition . 2002 ; 85 ( 2 ): 189 - 96 . PMID: 12127699 .
33. Blau VC , Maurer U , Tottenham N , McCandliss BD . The face-specific N170 component is modulated by emotional facial expression . Behav Brain Funct . 2007 ; 3 :7. doi: 10.1186/ 1744 - 9081 -3- 7 PMID: 17244356; PubMed Central PMCID : PMC1794418 .
34. Freeman JB , Ambady N , Holcomb PJ . The face-sensitive N170 encodes social category information . Neuroreport . 2010 ; 21 ( 1 ): 24 - 8 . doi: 10.1097/WNR.0b013e3283320d54 PMID: 19864961 ; PubMed Central PMCID : PMC3576572 .
35. Cuthbert BN , Schupp HT , Bradley MM , Birbaumer N , Lang PJ . Brain potentials in affective picture processing: covariation with autonomic arousal and affective report . Biol Psychol . 2000 ; 52 ( 2 ): 95 - 111 . PMID: 10699350 .
36. Olofsson JK , Nordin S , Sequeira H , Polich J. Affective picture processing: an integrative review of ERP findings . Biol Psychol . 2008 ; 77 ( 3 ): 247 - 65 . doi: 10.1016/j.biopsycho. 2007 .11.006 PMID: 18164800; PubMed Central PMCID: PMC2443061.
37. Rozenkrants B , Polich J. Affective ERP processing in a visual oddball task: arousal , valence, and gender. Clin Neurophysiol . 2008 ; 119 ( 10 ): 2260 - 5 . doi: 10.1016/j.clinph. 2008 .07.213 PMID: 18783987; PubMed Central PMCID: PMC2605010.
38. Valdez P , Mehrabian A. Effects of color on emotions . J Exp Psychol Gen . 1994 ; 123 ( 4 ): 394 - 409 . PMID: 7996122 .
39. Curran T , Hancock J. The FN400 indexes familiarity-based recognition of faces . Neuroimage . 2007 ; 36 ( 2 ): 464 - 71 . doi: 10.1016/j.neuroimage. 2006 .12.016 PMID: 17258471; PubMed Central PMCID: PMC1948028.
40. Kleih SC , Nijboer F , Halder S , Kubler A. Motivation modulates the P300 amplitude during brain-computer interface use . Clin Neurophysiol . 2010 ; 121 ( 7 ): 1023 - 31 . doi: 10.1016/j.clinph. 2010 .01.034 PMID: 20188627.
41. Jin J , Sellers EW , Wang X. Targeting an efficient target-to-target interval for P300 speller brain-computer interfaces . Medical & biological engineering & computing . 2012 ; 50 ( 3 ): 289 - 96 . doi: 10.1007/ s11517- 012 - 0868 -x PMID: 22350331 ; PubMed Central PMCID : PMC3646326 .