See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/6413426

A survey of signal processing algorithms in brain-computer interfaces based

on electrical brain signals

Article in Journal of Neural Engineering · July 2007

DOI: 10.1088/1741-2560/4/2/R03 · Source: PubMed

CITATIONS READS

901 3,090

4 authors, including:

Ali Bashashati Rabab K. Ward

BC Cancer Agency University of British Columbia
241 PUBLICATIONS 18,290 CITATIONS 434 PUBLICATIONS 14,509 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Rabab K. Ward on 07 January 2015.

The user has requested enhancement of the downloaded file.

IOP PUBLISHING JOURNAL OF NEURAL ENGINEERING
J. Neural Eng. 4 (2007) R32–R57 doi:10.1088/1741-2560/4/2/R03

TOPICAL REVIEW

A survey of signal processing algorithms

in brain–computer interfaces based on
electrical brain signals
Ali Bashashati1, Mehrdad Fatourechi1, Rabab K Ward1,2
and Gary E Birch1,2,3
1
Department of Electrical and Computer Engineering, The University of British Columbia,
2356 Main Mall, Vancouver, V6T 1Z4, Canada
2
Institute for Computing, Information and Cognitive Systems, 289-2366 Main Mall, Vancouver,
BC V6T 1Z4, Canada
3
Neil Squire Society, 220-2250 Boundary Rd, Burnaby, BC, V5M 4L9, Canada
E-mail: alibs@ece.ubc.ca

Received 11 October 2006

Accepted for publication 16 February 2007
Published 27 March 2007
Online at stacks.iop.org/JNE/4/R32

Abstract
Brain–computer interfaces (BCIs) aim at providing a non-muscular channel for sending
commands to the external world using the electroencephalographic activity or other
electrophysiological measures of the brain function. An essential factor in the successful
operation of BCI systems is the methods used to process the brain signals. In the BCI
literature, however, there is no comprehensive review of the signal processing techniques used.
This work presents the first such comprehensive survey of all BCI designs using electrical
signal recordings published prior to January 2006. Detailed results from this survey are
presented and discussed. The following key research questions are addressed: (1) what are the
key signal processing components of a BCI, (2) what signal processing algorithms have been
used in BCIs and (3) which signal processing techniques have received more attention?

S This article has associated online supplementary data files

1. Introduction Signal processing forms an important part of a BCI design,

since it is needed in extracting the meaningful information
The ultimate purpose of a direct brain–computer interface from the brain signal.
(BCI) is to allow an individual with severe motor disabilities to This paper summarizes the results of a comprehensive
have effective control over devices such as computers, speech survey of different signal processing schemes that have
synthesizers, assistive appliances and neural prostheses. Such been used in BCI systems. It specifically focuses on the
an interface would increase an individual’s independence, following signal processing components of a BCI: the pre-
leading to an improved quality of life and reduced social processing (which we refer to as the signal enhancement),
costs. feature extraction, feature selection/dimensionality reduction,
A BCI system detects the presence of specific patterns in feature classification and post-processing blocks. To address
a person’s ongoing brain activity that relates to the person’s all related BCI research, we include all the approaches that
intention to initiate control. The BCI system translates these use standard scalp-recorded EEG as well as those that use
patterns into meaningful control commands. To detect these epidural, subdural or intracortical recordings. The aims of
patterns, various signal processing algorithms are employed. this study are (a) to make it easy to identify the signal

1741-2560/07/020032+26$30.00 © 2007 IOP Publishing Ltd Printed in the UK R32

 Topical Review

electrodes

state feedback

Device
BCI Transducer Control
Display
User
Artifact Feature Feature Control Device
amp Processor Generator Translator Interface Controller

Signal Feature Feature Feature Post-

Enhancement Extraction Selection/ Classification processing
Dimensionality
Reduction

Figure 1. Functional model of a BCI system (Mason and Birch 2003). Note that the control display is optional. This review focuses on the
shaded components of BCI systems.

processing methods employed in different BCI systems, and would continually generate features relating to the power-
consequently to identify the methods that have not yet been spectral estimates of the user’s mu and beta rhythms.
explored, (b) to form a historical reference for new The feature generator generally can be a concatenation
researchers in this field and (c) to introduce a possible of three components—the ‘signal enhancement’, the
taxonomy of signal processing methods in brain–computer ‘feature extraction’ and the ‘feature selection/dimensionality
interfaces. reduction’ components, as shown in figure 1.
The organization of the paper is as follows. In section 2, In some BCI designs, pre-processing is performed on the
the general structure of a BCI system and the current brain signal prior to the extraction of features so as to increase
neuromechanisms4 in BCI systems are presented. Section 3 the signal-to-noise ratio of the signal. In this paper, we use the
details the procedure we followed to conduct this study. term ‘signal enhancement’ to refer to the pre-processing stage.
Results, discussion and conclusion are in sections 4–6, A feature selection/dimensionality reduction component is
respectively. sometimes added to the BCI system after the feature extraction
stage. The aim of this component is to reduce the number of
2. General structure of a BCI system features and/or channels used so that very high dimensional
and noisy data are excluded. Ideally, the features that are
Figure 1 shows the functional model of a BCI system (Mason meaningful or useful in the classification stage are identified
and Birch 2003). The figure depicts a generic BCI system in and chosen, while others (including outliers and artifacts) are
which a person controls a device in an operating environment omitted.
(e.g., a powered wheelchair in a house) through a series of The ‘feature translator’ translates the features into logical
functional components. In this context, the user’s brain activity (device-independent) control signals, such as a two-state
is used to generate the control signals that operate the BCI discrete output. The translation algorithm uses linear
system. The user monitors the state of the device to determine classification methods (e.g., classical statistical analyses) or
the result of his/her control efforts. In some systems, the user nonlinear ones (e.g., neural networks). According to the
may also be presented with a control display, which displays definition in Mason and Birch (2003), the resultant logical
the control signals generated by the BCI system from his/her output states are independent of any semantic knowledge
brain activity. about the device or how it is controlled. As shown in
The electrodes placed on the head of the user record the figure 1, a feature translator may consist of two components:
brain signal from the scalp, or the surface of the brain, or ‘feature classification’ and ‘post-processing’. The main aim
from the neural activity within the brain, and convert this of the feature classification component is to classify the
brain activity to electrical signals. The ‘artifact processor’ features into logical control signals. Post-processing methods
block shown in figure 1 removes the artifacts from the such as a moving average block may be used after feature
electrical signal after it has been amplified. Note that many classification to reduce the number of error activations of the
transducer designs do not include artifact processing. The system. The components between the user and control
‘feature generator’ block transforms the resultant signals into interface can be treated as a single component, a BCI
feature values that correspond to the underlying neurological transducer, which functions in a manner similar to physical
mechanism employed by the user for control. For example, transducers like a dial or switch. The role of the BCI transducer
if the user is to control the power of his/her mu (8–12 Hz) is to translate the user’s brain activity into logical (or device-
and beta (13–30 Hz) rhythms, the feature generator independent) control signals.
4 According to the Merriam-Webster Medical Dictionary, a bodily regulatory The control interface translates the logical control signals
mechanism based in the structure and functioning of the nervous system is (from the feature translator) into semantic control signals
called a neuromechanism. that are appropriate for the particular type of device used.

R33
Topical Review

Table 1. Taxonomy for BCI transducer designs.

Item Terms Definition
BCI transducer Artifact processor Removes artifact from the input signal
Feature generator Signal enhancement (1) Enhances signal-to-noise ratio of the brain signal
(2) The output of this block is a signal with the same nature of the input
(i.e. the output like the input is in the temporal domain).
Feature extraction Generates the feature vectors
Feature selection/ Selects a subset or reduces the dimensionality of features
dimensionality reduction
Feature translator Feature classification Classifies the features into logical control signals
Post-processing Increases the performance after feature classification, e.g., by blocking
activations with low certainty

Finally, the device controller translates the semantic control (1) One or more of the keywords BCI, BMI, DBI appeared in
signals into physical control signals that are used by the device. its title, abstract or keyword list.
The device controller also controls the overall behavior of the (2) The work described one or more BCI designs (the
device. For more detail, refer to Mason and Birch (2003). minimum design content that met the criteria was a BCI
Table 1 provides a simplified description of the BCI transducer as described in section 2). There were a
transducer components. few papers that only reported pre-processing techniques
specifically designed for brain–computer interfaces that
use neural cortical recordings. These papers were
2.1. Electrophysiological sources of control in current BCIs reported in the pre-processing techniques. Papers that
In BCI systems, electrophysiological sources refer to the presented tutorials, descriptions of electrode technology,
neurological mechanisms or processes employed by a BCI neuroanatomy, and neurophysiology discussions that
user to generate control signals. Current BCIs fall into seven might serve as the basis for a BCI were not included.
(3) Only papers published in English and in refereed
main categories, based on the neuromechanisms and recording
international journals and conference proceedings were
technology they use. In Wolpaw et al (2002) BCI systems
included.
are categorized as five major groups. These categories are
(4) Designs that use functional magnetic resonance imaging
sensorimotor activity, P300, VEP, SCP and activity of neural
(fMRI) (Weiskopf et al 2003, 2004, Yoo et al 2004),
cell (ANC). In this paper, two other categories were added:
magneto-encephalography (MEG) signals (Georgopoulos
‘response to mental tasks’ and ‘multiple neuromechanisms’.
et al 2005), near-infra-red spectrum (NIRS), auditory
BCI systems that use non-movement mental tasks to control
evoked potentials (Hill et al 2004, Su Ryu et al 1999)
a BCI (e.g. Anderson et al (1995b) and Millan et al (1998))
and somatosensory evoked potentials (Yan et al 2004)
assume that different mental tasks (e.g. solving a multiplication were not included in this paper.
problem, imagining a 3D object, or mental counting) lead (5) Papers were published prior to January 2006.
to distinct, task-specific EEG patterns and aim to detect the
patterns associated with these mental tasks from the EEG. BCI Although no paper meeting the five criteria explained above
systems based on multiple neuromechanisms (e.g. Gysels et al was omitted from the analysis, some papers may have been
(2005)) use a combination of two or more of the above- missed unintentionally. The current work should thus be
mentioned neuromechanisms in a single design of a BCI regarded as an initial step to build a public database that can
system. be updated and evolved with time.
In tables 3–8, we categorized the papers according to
Table 2 shows these categories with a short description
the signal processing methods used. For each of the design
of each. Note that although the designs that use direct
blocks of a BCI system shown in figure 1 (signal enhancement,
cortical recordings are included as a separate group, direct
feature extraction, feature selection/dimensionality reduction,
cortical recording is a recording technology and not a
feature classification, and post-processing), we created a table
neuromechanism. As shown in table 2, BCI designs that
that reports the signal processing techniques used in that
use sensorimotor activity as the neural source of control can
block. Since most of the BCI designs that use neural cortical
be further divided into three sub-categories: those based on
recordings do not contain a feature extraction component, we
changes in brain rhythms (e.g. the mu and beta rhythms), those generated a separate table (table 7) to report the translation
based on movement-related potentials (MRPs) and those based schemes used in these designs.
on other sensorimotor activity. Feature extraction methods used in BCI systems are
closely related to the specific neuromechanism(s) used by a
BCI. For example, the feature extraction algorithms employed
3. Methods in VEP-based BCIs are used to detect the visual evoked
potentials in the ongoing EEG. In BCI systems that operate
The BCI designs selected for this review include every journal on slow cortical potentials (SCP), the extracted features
and conference paper that met the following criteria: are mostly used for the purpose of identifying this specific

R34
 Topical Review

Table 2. Electrophysiological activities used in BCI designs and their definitions.

Neuromechanism Short description
Sensorimotor Changes in brain Mu rhythms in the range of 8–12 Hz and beta rhythms in the range of 13–30 Hz both originate in
activity rhythms (mu, Beta, the sensorimotor cortex and are displayed when a person is not engaged in processing sensorimotor
and gamma)a inputs or in producing motor outputs (Jasper and Penfield 1949). They are mostly prominent in
frontal and parietal locations (Kozelka and Pedley 1990, Kubler et al 2001a, Niedermeyer and Lopes
da Silva 1998). A voluntary movement results in a circumscribed desynchronization in the mu and
lower beta bands (Pfurtscheller and Aranibar 1977). This desynchronization is called event-related
desynchronization (ERD) and begins in the contralateral rolandic region about 2 s prior to the onset
of a movement and becomes bilaterally symmetrical immediately before execution of movement
(Pfurtscheller and Lopes da Silva 1999). After a voluntary movement, the power in the brain
rhythms increases. This phenomenon, called event-related synchronization (ERS), is dominant over
the contralateral sensorimotor area and reaches a maximum around 600 ms after movement offset
(Pfurtscheller and Lopes da Silva 1999). Gamma rhythm is a high-frequency rhythm in the EEG.
Upon the occurrence of a movement, the amplitude of gamma rhythm increases. Gamma rhythms
are usually more prominent in the primary sensory area.
Movement-related MRPs are low-frequency potentials that start about 1–1.5 s before a movement. They have
potentials (MRPs) bilateral distribution and present maximum amplitude at the vertex. Close to the movement, they
become contralaterally preponderant (Babiloni et al 2004, Deecke and Kornhuber 1976, Hallett
1994).
Other sensorimotor The sensorimotor activities that do not belong to any of the preceding categories are categorized
activities as other sensorimotor activities. These activities are usually not restricted to a particular frequency
band or scalp location and usually cover different frequency ranges. An example would be features
extracted from an EEG signal filtered to frequencies below 30 Hz. Such a range covers different
event-related potentials (ERPs) but no specific neuromechanism is used.
Slow cortical potentials (SCPs) SCPs are slow, non-movement potential changes generated by the subject. They reflect changes in
cortical polarization of the EEG lasting from 300 ms up to several seconds. Functionally, a SCP
reflects a threshold regularization mechanism for local excitory mobilization (Neumann et al 2003,
Wolpaw et al 2002).
P300 Infrequent or particularly significant auditory, visual, or somatosensory stimuli, when interspersed
with frequent or routine stimuli, typically evoke in the EEG over the parietal cortex a positive peak
at about 300 ms after the stimulus is received. This peak is called P300 (Allison and Pineda 2003,
Kubler et al 2001a).
Visual evoked potentials (VEPs) VEPs are small changes in the ongoing brain signal. They are generated in response to a visual
stimulus such as flashing lights and their properties depend on the type of the visual stimulus (Kubler
et al 2001a). These potentials are more prominent in the occipital area.
If a visual stimulus is presented repetitively at a rate of 5–6 Hz or greater, a continuous oscillatory
electrical response is elicited in the visual pathways. Such a response is termed steady-state visual
evoked potentials (SSVEP). The distinction between VEP and SSVEP depends on the repetition rate
of the stimulation (Gao et al 2003b).
Response to mental tasks BCI systems based on non-movement mental tasks assume that different mental tasks (e.g., solving
a multiplication problem, imagining a 3D object, and mental counting) lead to distinct, task-specific
distributions of EEG frequency patterns over the scalp (Kubler et al 2001a).
Activity of neural cells (ANC) It has been shown that the firing rates of neurons in the motor cortex are increased when movements
are executed in the preferred direction of neurons. Once the movements are away from the preferred
direction of neurons, the firing rate is decreased (Donoghue 2002, Olson et al 2005).
Multiple neuromechanisms (MNs) BCI systems based on multiple neuromechanisms use a combination of two or more of the above-
mentioned neuromechanisms.
a
In Ramachandran and Histein (1998), references regarding the similarity between attempted movements and real movements in the ERD
of mu patterns are provided. An attempted movement occurs when a subject attempts to move some part of his/her body, but because of
either a disability or the experiment control, the actual movement does not happen. Similarly, Gevins et al (1989) have shown that
imaginary (attempted) movements generate movement-related potentials (MRPs) similar to those generated by actual movements. Thus,
neuromechanisms corresponding to attempted movements are grouped in the same category of real movement.

phenomenon in the brain signal. Thus in table 5, we algorithms are also categorized based on the feature extraction
categorize the feature extraction algorithms based on the seven methods. This can be found in the supplementary data
neurological sources described in table 2. For example, the at stacks.iop.org/JNE/4/R32, which also contains a detailed
methods used in VEP-based BCIs are assembled under a version of table 7, where the classification algorithms for BCIs
different category from those used in SCP-based BCIs. A that use cortical neural recordings are shown.
more detailed version of table 5 can be found in appendix B at Categorizing the feature classification methods based on
stacks.iop.org/JNE/4/R32. the feature extraction methods used does not necessarily limit
The different feature classification algorithms used the use of a specific feature classification to a specific feature
in BCI systems are shown in table 6. As feature extraction method. The same applies to the categorization
classification algorithms are also closely related to the type of the feature extraction methods based on neuromechanisms
of the features that they classify, the feature classification used in BCI systems. The aim here is to provide as specific

R35
Topical Review

information as possible about signal processing in current BCI designs that incorporated signal enhancement algorithms other
designs and the researchers can combine any feature extraction than the general band-pass filtering of the EEG, the power-line-
method and/or feature classification method from different effect rejection and the traditional normalization of the signal
categories if necessary. were reported in the signal enhancement section of this paper.
Each table includes major classes corresponding to each
design block. These classes were initially determined by 4. Results
our team and then refined after an initial pass through
the selected papers. In some cases, each major class The detailed classification results of the survey are
was further divided into more specific categories. The summarized in tables 3–8 (refer to supplementary data
full classification template with all the major classes and at stacks.iop.org/JNE/4/R32 for more detailed versions of
sub-classes of each design component is listed in the left tables 5–7). As mentioned in section 3, these six tables address
column of tables 3, 4 and 8, and in the left two major columns the signal enhancement, feature selection/dimensionality
of tables 5–7. Note that in this paper, major classes are written reduction, feature extraction, feature classification and post-
in bold type and sub-classes are represented in bold-italic processing methods used. The references listed for each sub-
type. For example, in table 5 or appendix B (in supplementary class entry represent all the papers that reported on designs
data), VEP-based BCI designs that use some type of power- related to that sub-class. As such, one can find all the
spectral parameters of the EEG are categorized under the VEP- designs that have specific attributes of interest. For example,
spectral parameters class, while a BCI design that is based if one is interested in all BCI technology designs that have
on the movement-related potentials (MRP) and that uses the used parametric modeling (and specifically extracted AR
same method is categorized under the sensorimotor activity- parameters of the signal) to detect the sensorimotor activity,
spectral parameters and sensorimotor activity-MRP- then all the references to the relevant papers can be found in
spectral parameters classes in table 5 and appendix B, table 5 under sensorimotor activity—parametric modeling
respectively. As an example from table 6, BCI designs that use (AR, AAR and ARX parameters). Alternatively, if one is
linear discriminant analysis (LDA) classifiers are categorized looking for designs that do not have a feature extraction
under LDA. Supplementary data at stacks.iop.org/JNE/4/R32, block but directly apply the support vector machine (SVM)
which has a more detailed version of table 6, categorizes BCI classification method on the brain signal, then these papers
designs that use PSD features and LDA classifier under PSD- can be easily located in the none-SVM class in supplementary
LDA class. data at stacks.iop.org/JNE/4/R32. Similarly, the designs that
The category for each BCI design was determined by use the SVM classification method, regardless of the feature
selecting the closest sub-class in the classification template. extraction technique used, are categorized under the SVM
For the papers that reported multiple designs multiple class in table 6.
classifications were recorded. The designs were categorized To enhance the clarity of tables 3–8, the following
based only on what was reported in each paper. No personal notations are used:
knowledge of an authors’ related work was used in the (a) BCI designs based on multiple neuromechanisms (as
classification. defined in table 2) are presented in separate categories
In some cases, it was difficult to differentiate between the such as MN: sensorimotor activity + response to mental
signal enhancement, feature selection and feature extraction tasks, which show that the design is based on the
design components of a brain–computer interface. Based sensorimotor activity and the response to mental tasks
on the definitions in table 1, the methods that satisfied neuromechanisms.
the following four criteria were considered to be signal (b) Two or more methods that are consecutively used in a
enhancement methods: design block are separated by ‘−’. As an example, CSP–
log transformation denotes a design that first applies
(1) The method was implemented to improve the signal-to-
common spatial patterns (CSP) on the signal and then
noise ratio of the brain signal.
applies a logarithmic function on the resulting time-
(2) The output of the block had the same nature as the
series.
input brain signal (i.e. the output stayed in the temporal
(c) Two methods that are applied simultaneously in a design
domain).
component are separated by ‘+’. For example, AR
(3) The algorithm was directly performed on the brain signal
parameters + PSD parameters corresponds to a design
and not on the features extracted from the brain signal.
that uses both autoregressive (AR) and power-spectral-
(4) The method did not handle artifacts.
density (PSD) features in the feature extraction block.
The common spatial patterns (CSP) method is an example (d) LRP + {CSP − log-transformation} denotes designs that
of a method that satisfies the four above-mentioned criteria use the two kinds of feature extraction methods separated
and was thus categorized as a signal enhancement method. by ‘+’. The first method is based on the extraction of
The principle component analysis (PCA) method is another lateralized readiness potentials (LRP), and the second
example that sometimes satisfies the above four criteria and feature extraction method is based on consecutively
was categorized as a signal-enhancement method. In the applying CSP followed by a logarithm function on the
cases where PCA is applied after feature extraction to reduce signals that are grouped in ‘{ }’.
the dimensionality of the extracted features, it is categorized (e) To facilitate readability, we have provided an index of
as feature selection/dimensionality reduction method. Only terms in appendix A.

R36
 Topical Review

Table 3. Pre-processing (signal enhancement) methods in BCI designs.

Pre-processing method Reference ID
Common average referencing (CAR) (Cheng et al 2004, Fabiani et al 2004, Kubler et al 2005, Li et al 2004b, McFarland
and Wolpaw 1998, McFarland et al 1997, 2003, Peters et al 2001, Ramoser et al 2000,
Schalk et al 2000, Wolpaw et al 1997)
Surface Laplacian (SL) (Babiloni et al 2000, 2001b, Cincotti et al 2001, 2003a, Dornhege et al 2004, Fabiani
et al 2004, Gysels and Celka 2004, McFarland and Wolpaw 1998, McFarland et al 1997,
2003, 2005, Millan et al 1998, 2000a, 2002a, 2002b, 2003a, 2004a, 2004b, Millan
2004, Millan and Mourino 2003b, Muller et al 2003b, Peters et al 2001, Qin et al
2004a, 2004b, 2005, Qin and He 2005, Ramoser et al 2000, Schalk et al 2000, Wang
et al 2004a, 2004b, Wolpaw and McFarland 2004)
Independent component analysis (ICA) (Bayliss and Ballard 1999, 2000a, 2000b, Erfanian and Erfani 2004, Gao et al 2004,
Peterson et al 2005, Wu et al 2004, Serby et al 2005, Xu et al 2004a, Wang et al 2004c,
Li et al 2004a)
Common spatial patterns (CSP) (Blanchard and Blankertz 2004, Dornhege et al 2003, 2004, Guger et al 2000b, Krauledat
et al 2004, Muller et al 2003b, Pfurtscheller et al 2000, Pfurtscheller and Neuper 2001,
Ramoser et al 2000, Townsend et al 2004, Xu et al 2004b)
Principal component analysis (PCA) (Chapin et al 1999, Guan et al 2004, Hu et al 2004, Isaacs et al 2000, Lee and Choi
2002, 2003, Thulasidas et al 2004, Xu et al 2004a, Yoon et al 2005, Li et al 2004a)
Combined CSP and PCA (Xu et al 2004b)
Singular value decomposition (SVD) (Trejo et al 2003)
Common spatio-spatial patterns (CSSP) (Lemm et al 2005)
Frequency normalization (Freq-Norm) (Bashashati et al 2005, Borisoff et al 2004, Fatourechi et al 2004, 2005, Yu et al
2002)
Local averaging technique (LAT) (Peters et al 2001)
Robust Kalman filtering (Bayliss and Ballard 1999, 2000a)
Common spatial subspace decomposition (Cheng et al 2004, Li et al 2004b, Liu et al 2003, Wang et al 2004d)
(CSSD)
Wiener filtering (Vidal 1977)
Sparse component analysis (Li et al 2004a)
Maximum noise fraction (MNF) (Peterson et al 2005)
Spike detection methods (Obeid and Wolf 2004)
Neuron ranking methods (Sanchez et al 2004)

Table 4. Feature selection/dimensionality reduction methods in BCI designs.

Feature selection/dimensionality reduction method Reference ID
Genetic algorithm (GA) (Flotzinger et al 1994, Garrett et al 2003, Graimann et al 2003a,
2004, Peterson et al 2005, Scherer et al 2004, Schroder et al 2003,
Tavakolian et al 2004, Yom-Tov and Inbar 2001, 2002)
Principal component analysis (PCA) (Anderson et al 1995a, Bashashati et al 2005, Borisoff et al 2004,
Fatourechi et al 2004, 2005)
Distinctive sensitive learning vector quantization (DSLVQ) (Flotzinger et al 1994, Neuper et al 2005, Pfurtscheller et al 1996,
1997, 1998, 2000, Pfurtscheller and Neuper 2001, Pregenzer and
Pfurtscheller 1995, 1999)
Sequential forward feature selection (SFFS) (Fabiani et al 2004, Keirn and Aunon 1990)
Grid search method (Glassman 2005)
Relief method (Kira and Rendell 1992) (Millan et al 2002a)
Recursive feature/channel elimination (RFE) (Lal et al 2004, Schröder et al 2005)
Support vector machine (SVM)-based recursive feature (Gysels et al 2005)
elimination
Stepwise discriminant procedure (Vidal 1977)
Linear discriminant analysis (LDA) (Graimann et al 2003b)
Fisher discriminant analysis (dimensionality reduction) (Wang et al 2004d)
Fisher discriminant-based criterion (feature selection) (Cheng et al 2004, Lal et al 2004)
Zero-norm optimization (l0-opt) (Lal et al 2004)
Orthogonal least square (OLS1) based on radial basis function (Xu et al 2004b)
(RBF)

5. Discussion enhancement, feature selection/dimensionality reduction and

post-processing algorithms in BCI designs. Specifically, this
Several points raised in the previous section deserve figure shows the number of BCI designs that use specific signal
further comment. Figure 2 summarizes the information enhancement, feature selection/dimensionality reduction and
in tables 3, 4 and 8, which respectively address signal post-processing techniques.

R37
Topical Review

Table 5. Feature extraction methods in BCI designs. Refer to appendix B in supplementary data for a more detailed version of this table.
Neuro-
mechanism Feature extraction method Reference ID
Sensorimotor Spectral parameters (Babiloni et al 2000, 2001a, 2001b, Blanchard and Blankertz 2004, Boostani and Moradi
activity 2004, Cho et al 2004, Cincotti et al 2001, 2003a, 2003b, Coyle et al 2005, Fabiani et al
2004, Flotzinger et al 1994, Garcia et al 2003b, Garrett et al 2003, Guger et al 2000b, 2003a,
Ivanova et al 1995, Kalcher et al 1992, 1993, Kelly et al 2002b, Krauledat et al 2004, Krausz
et al 2003, Kubler et al 2005, Lal et al 2004, Leeb and Pfurtscheller 2004, Lemm et al
2005, Mahmoudi and Erfanian 2002, Mason and Birch 2000, McFarland and Wolpaw 1998,
McFarland et al 1997, 2003, 2005, Millan et al 2002a, 2002b, Muller et al 2003c, Muller-Putz
et al 2005b, Neuper et al 2003, 2005, Pfurtscheller et al 1993, 1994, 1996, 1997, 1998, 2000,
2005, Pfurtscheller and Neuper 2001, Pfurtscheller et al 2003a, 2003b, Pineda et al 2003,
Pregenzer and Pfurtscheller 1995, 1999, Ramoser et al 2000, Schalk et al 2000, Scherer et al
2004, Sheikh et al 2003, Townsend et al 2004, Trejo et al 2003, Jia et al 2004, Wolpaw et al
1991, 1997, 2000, 2003, Wolpaw and McFarland 1994, 2004, Li et al 2004a)
Parametric modeling (Burke et al 2002, 2005, Graimann et al 2003b, Guger et al 1999, 2000a, 2003a, 2003b,
(AR, AAR & ARX Haselsteiner and Pfurtscheller 2000, Huggins et al 2003, Kelly et al 2002a, 2002b, Lal et al
parameters) 2004, Neuper et al 1999, Obermaier et al 2001a, 2001b, Peters et al 2001, Pfurtscheller et al
1998, 2000, Pfurtscheller and Guger 1999, Pfurtscheller and Neuper 2001, Schloegl et al
1997a, 1997b, Schlogl et al 2003, Schröder et al 2005, Sykacek et al 2003, Yoon et al 2005)
TFR method (Bashashati et al 2005, Birch et al 2002, 2003, Borisoff et al 2004, Bozorgzadeh et al 2000,
Costa and Cabral 2000, Fatourechi et al 2004, 2005, Garcia et al 2003a, 2003b, Glassman
2005, Graimann et al 2003a, 2004, Huggins et al 2003, Lemm et al 2004, Lisogurski and
Birch 1998, Mason and Birch 2000, Mason et al 2004, Pineda et al 2000, Qin et al 2004b,
2005, Qin and He 2005, Yom-Tov and Inbar 2003)
CCTM (Balbale et al 1999, Graimann et al 2003b, 2004, Huggins et al 1999, 2003, Levine et al 1999,
2000)
Signal envelope − cross- (Wang et al 2004a, 2004b)
correlation
Hjorth parameters (Boostani and Moradi 2004, Lee and Choi 2002, Obermaier et al 2001a, 2001c, Pfurtscheller
and Neuper 2001)
Signal complexity (Boostani and Moradi 2004, Roberts et al 1999, Trejo et al 2003)
Combination of different (Cheng et al 2004, Dornhege et al 2003, 2004, Krauledat et al 2004, Mahmoudi and Erfanian
feature extraction 2002, Muller et al 2003b, Yom-Tov and Inbar 2001, 2002)
methods
LRP features (Blankertz et al 2002a, 2003, Krauledat et al 2004)
Other (Coyle et al 2004, Huggins et al 2003, Hung et al 2005, LaCourse and Wilson 2003, Li et al
2004b, Liu et al 2003, Mason and Birch 2000, Pineda et al 2000, Qin et al 2004a, 2005, Wang
et al 2004d, Xu et al 2004b, Yom-Tov and Inbar 2003)
None (Barreto et al 1996a, 1996b, Blankertz et al 2002a, Lee and Choi 2002, 2003, Mahmoudi and
Erfanian 2002, Parra et al 2002, 2003a, Schroder et al 2003, Trejo et al 2003)
SCP Calculation of SCP (Birbaumer et al 1999, 2000, Hinterberger et al 2003, 2004a, 2004b, 2005a, 2005b, Kaiser
amplitude et al 2001, 2002, Kubler et al 1998, 1999, 2001b, Neumann et al 2003, 2004)
TFR method (Bostanov 2004, Hinterberger et al 2003)
Mixed filter (Hinterberger et al 2003)
None (Hinterberger et al 2003, Schroder et al 2003)
P300 Cross-correlation (Bayliss and Ballard 1999, 2000a, 2000b, Farwell and Donchin 1988)
Stepwise (Donchin et al 2000, Farwell and Donchin 1988)
discriminant
analysis
Matched filtering (Serby et al 2005)
PPM (Jansen et al 2004)
TFR method (Bostanov 2004, Donchin et al 2000, Fukada et al 1998, Glassman 2005, Jansen et al 2004,
Kawakami et al 1996)
Peak picking (Allison and Pineda 2003, 2005, Bayliss et al 2004, Farwell and Donchin 1988)
Area calculation (Farwell and Donchin 1988)
Area and peak picking (Kaper and Ritter 2004b, Xu et al 2004a)
Not mentioned (calcu- (Bayliss 2003, Polikoff et al 1995)
lated P300 but details not
mentioned)
None (Guan et al 2004, Jansen et al 2004, Kaper and Ritter 2004a, 2004b, Kaper et al 2004,
Thulasidas et al 2004)
VEP Spectral parameters (Cheng and Gao 1999, Cheng et al 2001, 2002, 2005, Gao et al 2003b, Kelly et al 2004, 2005a,
2005b, 2005c, Lalor et al 2005, Middendorf et al 2000, Muller-Putz et al 2005a, Wang et al
2004c, 2005a)
Lock-in amplifier (Calhoun and McMillan 1996, McMillan and Calhoun 1995, Muller-Putz et al 2005a)
Asymmetry ratio of (Su Ryu et al 1999)
different band powers

R38
 Topical Review

Table 5. (Continued.)
Neuro-
mechanism Feature extraction method Reference ID
Cross-correlation (Sutter 1992)
Amplitude between (Lee et al 2005)
N2 and P2 peaks
None (Guan et al 2005, Vidal 1977)
Response to Spectral parameters (Bashashati et al 2003, Keirn and Aunon 1990, Kostov and Polak 1997, Liu et al 2005, Millan
mental tasksa et al 1998, Palaniappan et al 2002, Palaniappan 2005, Peterson et al 2005, Polak and Kostov
1997, 1998, Wang et al 2005a)
Parametric modeling (Anderson et al 1995b, 1998, Garrett et al 2003, Huan and Palaniappan 2004, Keirn and
(AR & AAR parameters) Aunon 1990, Kostov and Polak 2000, Huan and Palaniappan 2005, Polak and Kostov 1998,
1999, Sykacek et al 2003)
Signal Complexity (Bashashati et al 2003, Tavakolian et al 2004)
Eigenvalues of (Anderson et al 1998)
correlation matrix
LPC using Burg’s (Kostov and Polak 1997)
method
None (Anderson et al 1995a, Panuccio et al 2002)
ANC Cross-covariance − PCA (Isaacs et al 2000)
LBG vector quantization (Darmanjian et al 2003)
(VQ)
Filtering − rectification (Karniel et al 2002, Kositsky et al 2003, Reger et al 2000a, 2000b)
− thresholding
Averaging (Laubach et al 2000, Otto et al 2003, Vetter et al 2003)
TFR methods (Laubach et al 2000, Musallam et al 2004)
None (most of these (Black et al 2003, Byron et al 2005, Carmena et al 2003, 2005, Chapin et al 1999, Gao et al
designs model the 2002, 2003a, Hatsopoulos et al 2004, Hu et al 2004, Karniel et al 2002, Kemere et al 2004,
relationship between Kennedy et al 2000, Kim et al 2005a, 2005b, Lebedev et al 2005, Olson et al 2005, Patil et
neural firing rates and al 2004, Rao et al 2005, Roushe et al 2003, Sanchez et al 2002a, 2002b, 2003, Serruya et al
‘position and/or velocity 2002, 2003, Taylor et al 2002, 2003, Wessberg et al 2000, Wu et al 2002a, 2002b)
and/or acceleration’ of
hand)
MN: Spectral parameters (Gysels and Celka 2004, Gysels et al 2005, Millan et al 2000a, 2000b, 2002b, 2003a, 2004a,
sensorimotor Millan 2004, Millan and Mourino 2003b, Obermaier et al 2001d, Varsta et al 2000)
activity +
response to
mental tasks
Parametric modeling (Curran et al 2004, Penny et al 2000, Roberts and Penny 2003, Sykacek et al 2004, Varsta
et al 2000)
TFR method (Garcia and Ebrahimi 2002, Garcia et al 2002, 2003c, Molina et al 2003, Varsta et al 2000)
Combination of different (Erfanian and Erfani 2004)
features
PLV (Gysels and Celka 2004, Gysels et al 2005)
Mean spectral coherence (Gysels and Celka 2004)
None (Mourino et al 2002, Rezek et al 2003)
MN: SCP + SCP calculation + power (Hinterberger and Baier 2005, Mensh et al 2004)
other brain spectral parameters
rhythms
a
Designs that differentiate between relaxed state and movement tasks are considered in ‘sensorimotor activity + response to mental tasks’
category.

In the remainder of this section we highlight the top three sensitive learning vector quantization (DSLVQ), and 13% use
or four methods that have been used in the signal processing PCA.
blocks of BCI systems (as introduced in section 2). Of the 30 BCI designs that use post-processing algorithms
Of the 96 BCI designs that employ signal enhancement to reduce the amount of error in the output of the BCI
techniques before extracting the features from the signal, system, 57% use averaging techniques and consider rejecting
32% use surface Laplacian (SL), 22% use either principal activations that have low certainty, 27% consider using the
component analysis (PCA) or independent component analysis debounce block (or refractory period) to deactivate the output
(ICA), 14% use common spatial patterns (CSP) and for a short period of time when a false activation is detected,
11% use common average referencing (CAR) techniques. and 16% use event-related negativity (ERN) signals to detect
Thirty-eight of the reported BCI designs employ feature error activations.
selection/dimensionality reduction algorithms; 26% of these Figure 3 summarizes the results presented in table 5,
38 designs use genetic algorithms (GA), 24% use distinctive and shows the number of BCI designs that are based on

R39
Topical Review

Table 6. Feature classification methods in BCI designs that use EEG and ECoG recording technology.
Feature classification method Reference ID
Neural MLP (Anderson et al 1995a, 1995b, 1998, Costa and Cabral 2000, Erfanian and Erfani
networks 2004, Fukada et al 1998, Garrett et al 2003, Haselsteiner and Pfurtscheller 2000,
(NN) Huan and Palaniappan 2004, Hung et al 2005, Ivanova et al 1995, Mahmoudi and
Erfanian 2002, Huan and Palaniappan 2005, Palaniappan 2005, Su Ryu et al 1999,
Tavakolian et al 2004)
Committee of MLP NN (Millan et al 2000b, 2002b, Varsta et al 2000)
FIR- MLP NN (Haselsteiner and Pfurtscheller 2000)
Committee of Plat’s RAN (Millan et al 1998)
algorithm (Platt 1991)
Committee of NNs trained (Boostani and Moradi 2004)
with Adaboost
Committee of single perceptrons (Peters et al 2001)
with no hidden layers
TBNN (Ivanova et al 1995)
TDNN (Barreto et al 1996a, 1996b)
LVQ (Flotzinger et al 1994, Ivanova et al 1995, Kalcher et al 1992, 1993, Pfurtscheller
et al 1993, 1994, 1996, 1997, 1998, 2000, Pfurtscheller and Neuper 2001, Pregenzer
and Pfurtscheller 1999)
kMeans − LVQ (Bashashati et al 2005, Birch et al 2002, 2003, Borisoff et al 2004, Bozorgzadeh
et al 2000, Fatourechi et al 2004, 2005, Lisogurski and Birch 1998, Mason and Birch
2000, Mason et al 2004, Yom-Tov and Inbar 2003)
fART − LVQ (Borisoff et al 2004)
DSLVQ (Muller-Putz et al 2005a, Neuper et al 2005, Pregenzer and Pfurtscheller 1995, 1999)
Growing hierarchical SOM (Liu et al 2005)
ALN (Kostov and Polak 1997, 1998, 1999, 2000)
ANN (Cincotti et al 2003b)
Custom designed local NN (Mourino et al 2002, Millan et al 2000a, 2002b, Millan and Mourino 2003b)
Fuzzy ARTMAP (Palaniappan et al 2002)
Single layer NN (Garcia et al 2002)
RBF-NN (Hung et al 2005)
Static neural classifier (Adaline) (Barreto et al 1996a, 1996b)
Gamma NN (Barreto et al 1996a, 1996b)
(R) LDAa (Boostani and Moradi 2004, Bostanov 2004, Burke et al 2002, 2005, Coyle et al
2004, 2005, Dornhege et al 2003, 2004, Fabiani et al 2004, Fukada et al 1998, Garcia
et al 2003b, Garrett et al 2003, Guger et al 2000b, 2003a, 2003b, Hinterberger et al
2003, Huan and Palaniappan 2004, Huggins et al 2003, Kelly et al 2002a, 2002b,
2004, 2005a, 2005c, Krauledat et al 2004, Krausz et al 2003, Lalor et al 2005,
Leeb and Pfurtscheller 2004, Lemm et al 2005, Mensh et al 2004, Muller et al
2003b, 2003c, Muller-Putz et al 2005a, 2005b, Neuper et al 1999, 2003, Obermaier
et al 2001b, Pfurtscheller et al 1998, 2000, 2003b, Pfurtscheller and Guger 1999,
Pfurtscheller and Neuper 2001, Schloegl et al 1997a, 1997b, Townsend et al 2004,
Jia et al 2004)
(R) FLD (Babiloni et al 2001b, Blanchard and Blankertz 2004, Blankertz et al 2002a, 2003,
Cincotti et al 2001, 2003a, Guger et al 1999, 2000a, Hung et al 2005, Obermaier
et al 2001a, Pfurtscheller et al 2003a, Scherer et al 2004, Li et al 2004a)
Sparse FLD (Blankertz et al 2002a)
MD-based classifier (Babiloni et al 2001a, Cincotti et al 2003a, 2003b, Garcia and Ebrahimi 2002,
Molina et al 2003)
Nonlinear discriminant function (Fabiani et al 2004)
Bayes quadratic classifier (Keirn and Aunon 1990)
Bayesian classifier (linear classifier) (Curran et al 2004, Lemm et al 2004, Penny et al 2000, Roberts and Penny 2003)
Linear Bayian decision rule (Vidal 1977)
Linear classifier based on time-warping (Mason and Birch 2000)
Logistic regression (Parra et al 2002, 2003a)
Linear classifier (no details) (Ramoser et al 2000)
Single layer Perceptron model (a linear classifier) (Li et al 2004b, Wang et al 2004d)
Two-dimensional linear classifier trained by a (Cheng et al 2004)
non-enumerative search procedure
ZDA (Hinterberger et al 2003)
LDS (Lee and Choi 2002)
Gaussian classifier (Millan 2004, Millan et al 2004a, 2004b, 2003a)
SSP (Babiloni et al 2000, 2001a, 2001b, Cincotti et al 2001, 2003a, Millan et al 2000b,
2002b)
SOM-based SSP (Millan et al 2000b, 2002b)

R40
 Topical Review

Table 6. (Continued.)
Feature classification method Reference ID
HMM-based CHMM (Rezek et al 2003)
techniques
AR HMM (Panuccio et al 2002)
HMM + SVM (Lee and Choi 2002, 2003)
HMM (Cincotti et al 2003b, Lee and Choi 2003, Liu et al 2003, Obermaier et al 2001a,
2001c, 2001d, Pfurtscheller and Neuper 2001, Sykacek et al 2003)
SVM (Blankertz et al 2002a, Guan et al 2004, Garcia et al 2003a, 2003b, 2003c, Garrett
et al 2003, Glassman 2005, Gysels and Celka 2004, Gysels et al 2005, Hung et al
2005, Guan et al 2005, Kaper and Ritter 2004a, 2004b, Kaper et al 2004, Lal et al
2004, Peterson et al 2005, Schroder et al 2003, Schröder et al 2005, Thulasidas
et al 2004, Trejo et al 2003, Xu et al 2004b, Yom-Tov and Inbar 2001, 2002, 2003,
Yoon et al 2005)
NID3 (Ivanova et al 1995)
CN2 (Ivanova et al 1995)
C4.5 (Ivanova et al 1995, Millan et al 2002a)
k-NN (Blankertz et al 2002a, Pineda et al 2000, Pregenzer and Pfurtscheller 1999)
Threshold detector (Allison and Pineda 2003, Balbale et al 1999, Bayliss and Ballard 1999, 2000a,
2000b, Bayliss 2003, Bayliss et al 2004, Calhoun and McMillan 1996, Cheng
and Gao 1999, Cheng et al 2001, 2002, 2005, Donchin et al 2000, Farwell and
Donchin 1988, Gao et al 2003b, Graimann et al 2003a, 2003b, 2004, Hinterberger
et al 2003, Huggins et al 1999, 2003, Jansen et al 2004, Kawakami et al
1996, Kelly et al 2005b, Kostov and Polak 1997, Lee et al 2005, Levine et al
1999, Levine et al 2000, McMillan and Calhoun 1995, Middendorf et al 2000,
Pfurtscheller et al 2005, Pineda et al 2003, Polak and Kostov 1997, Polikoff et al
1995, Qin et al 2004a, 2004b, Qin and He 2005, Roberts et al 1999, Serby et al
2005, Sutter 1992, Wang et al 2004a, 2004b, Xu et al 2004a, Yom-Tov and Inbar
2003)
Linear combination − threshold detector (Townsend et al 2004)
Continuous feedback + threshold detector (Birbaumer et al 1999, 2000, Hinterberger et al 2003, 2004a, 2004b, 2005a, 2005b,
Kaiser et al 2001, 2002, Kubler et al 1998, 1999, 2001b, Neumann et al 2003, 2004)
Linear combination − continuous feedback (Fabiani et al 2004, Krausz et al 2003, Kubler et al 2005, McFarland and Wolpaw
1998, McFarland et al 1997, 2003, 2005, Schalk et al 2000, Sheikh et al 2003,
Wolpaw and McFarland 1994, 2004, Wolpaw et al 1997, 2000, 2003)
Continuous feedback (Bashashati et al 2003, Cho et al 2004, LaCourse and Wilson 2003, Middendorf
et al 2000, Trejo et al 2003, Wolpaw et al 1991)
Continuous feedback using MD (Schlogl et al 2003)
Continuous audio feedback (Hinterberger and Baier 2005)
Variational Kalman filter (Sykacek et al 2004)
Static classifier that is inferred with (Curran et al 2004, Sykacek et al 2004)
sequential variational inference
Random forest algorithm (Neuper et al 1999)
a
Regularization may be applied before LDA classification scheme.

sensorimotor activity, SCP, VEP, P300, activity of neural cells, designs that use mental tasks to control a BCI use power-
‘response to mental tasks’ and multiple neuromechanisms and spectral features and 37% use parametric modeling of the input
use different feature extraction techniques. signal. As most of the BCI designs that are based on neural
Based on the results of figure 3, 41% of the BCIs that are cortical recordings mainly try to model the direct relationship
based on the sensorimotor activity use power-spectral-density between the neural cortical recordings and movements, they
features, 16% rely on parametric modeling of the data, 13% use do not use a feature-extraction algorithm. 45% of the
time–frequency representation (TFR) methods and 6% do not BCI designs that are based on multiple neuromechanisms
employ any feature extraction methods. 74% of the SCP-based rely on power-spectral features, 17% use parametric
BCI designs calculate SCP signals using low-pass filtering modeling, and 17% use time–frequency representation (TFR)
methods, and 64% of the VEP-based BCIs use power-spectral methods.
features at specific frequencies. 26% of the BCIs based on Summarizing tables 6 and 7, the number of BCI
P300 calculate the peaks of the signal in a specific time window designs that use different feature classification algorithms
to detect the P300 component of the EEG; 22% use TFR-based are shown in figure 4. About 75% of the BCI designs
methods, 22% use no feature extraction method, and 15% use use classification schemes that are not based on neural
cross-correlation with a specific template. 41% of the BCI networks (NN). These are composed of those methods that

R41
Topical Review

Table 7. Feature classification methods in BCI designs that are based on neural cortical recordings.
Feature
classification Reference ID
Neural networks Recurrent MLP neural (Sanchez et al 2002a, 2002b, 2003)
network (RNN)
MLP (Kim et al 2005b)
Feed-forward ANN (Patil et al 2004)
ANN recurrent dynamic (Chapin et al 1999)
back-propagation
ANN model (Hatsopoulos et al 2004, Wessberg et al 2000)
LVQ (Laubach et al 2000)
Other (Karniel et al 2002)
Support vector machine regression (SVR) model (Kim et al 2005b)
Cosine tuning model (a linear model) (Black et al 2003, Kemere et al 2004, Taylor et al 2002, 2003)
Linear Gaussian models (LGM) implemented by (Black et al 2003, Gao et al 2003a, Patil et al 2004, Sanchez et al 2002a, Wu et al
Kalman filter 2002a, 2002b)
Generalized linear models (GLA) (Black et al 2003, Gao et al 2003a)
Generalized additive models (GAM) (Black et al 2003, Gao et al 2003a)
Weighted linear combination of neuronal activity (Carmena et al 2005, Hatsopoulos et al 2004, Kim et al 2005a, 2005b, Lebedev et al
(Wiener filter: a linear model) 2005, Patil et al 2004, Sanchez et al 2002b, Serruya et al 2003, 2002)
Gamma filter (a linear model) (Sanchez et al 2002b)
Mixture of multiple models based on NMF (Kim et al 2005a)
(non-negative matrix factorization)
Echo state networks (ESN) − optimal sparse (Rao et al 2005)
linear mapping
Linear model (no details mentioned) (Carmena et al 2003, Wessberg et al 2000)
Threshold detector (Otto et al 2003, Roushe et al 2003, Vetter et al 2003)
SVM (Byron et al 2005, Hu et al 2004, Olson et al 2005)
Bayesian classifier (Gao et al 2002, Hu et al 2004, Musallam et al 2004)
Maximum likelihood-based model (Hatsopoulos et al 2004, Kemere et al 2004, Serruya et al 2003)
LPF (continuous signal) (Karniel et al 2002, Kositsky et al 2003, Reger et al 2000a, 2000b)
Direct translation of firing rate to cursor (Kennedy et al 2000)
movement (continuous signal)
k-NN (Isaacs et al 2000)
HMM (Darmanjian et al 2003)

Table 8. Post-processing methods in BCI designs.

Post processing Reference ID
ERN (event-related negativity)-based (Bayliss et al 2004, Blankertz et al 2002b, 2003, Parra et al 2003b, Schalk et al 2000)
error correction

Successive averaging and/or rejection (Anderson et al 1995a, Bashashati et al 2005, Birch et al 2002, Borisoff et al 2004, Fatourechi
option for ‘moderated’ posterior et al 2004, 2005, Gysels and Celka 2004, Millan et al 1998, 2004b, Millan 2004, Millan and
probabilities (choice of ‘unknown’ Mourino 2003b, 2004b, Muller-Putz et al 2005b, Penny et al 2000, Roberts and Penny 2003,
output state) (SA-UK) Townsend et al 2004, Vidal 1977, Millan et al 2003a)

Debounce (considering refractory (Bashashati et al 2005, Borisoff et al 2004, Fatourechi et al 2004, 2005, Muller-Putz et al 2005b,
period) Obeid and Wolf 2004, Pfurtscheller et al 2005, Townsend et al 2004)

use threshold detectors as the feature classifier or as part of During our analysis of the literature, a number of salient
the feature classification scheme (27%), linear discriminant points about the signal processing methods emerged. We
(either LDA or FLD) classifiers (26%), those that show think that some of these points are worth sharing with the
continuous feedback of the extracted features (16%), and BCI research community. In the following sections we
those that use support vector machines (SVM) (11%). 27% summarize some of them. Note that these observations
of the neural-network-based classifiers are based on the are based on comments made by the researchers in their
multi-layer perceptrons (MLP) neural network and 39% are published papers and are also based on the trends in the
based on learning-vector-quantization (LVQ) classification literature; they do not cover all the methods reported in the
scheme. literature.

R42
 Topical Review

For BCIs that use the mu and beta rhythms, the

ERN common average referencing (CAR) and Laplacian methods
SA-UK Post-Processing are superior to the ear reference method. This may be because
Methods
Debounce these methods use high-pass spatial filters and enhance the
focal activity from the local sources (e.g. the mu and the
Other Feature Selection/ beta rhythms) and reduce the widely distributed activity,
DSLVQ Dimensionality including that resulting from distant sources (e.g. EMG, eye
PCA Reduction Methods movements and blinks, visual alpha rhythm). Comparing
GA the two variations of the Laplacian filtering methods (the large
Laplacian and the small Laplacian), it is shown that the large
Laplacian method is superior to the small Laplacian method
Other
in BCI systems that use the mu rhythm (McFarland et al
Freq-Norm
1997).
CSSD
Although an accurate Laplacian estimate from raw
CSP Signal Enhancement
Methods
potentials requires many electrodes, one study showed that
ICA
the recognition rates were increased by using a small number
PCA
of electrodes only (Cincotti et al 2003a). In this case, the
SL linear combination of channels implementing the Laplacian
CAR
estimation was likely to have caused a favorable transformation
0 5 10 15 20 25 30 35 of the signals to recognize different patterns in the ongoing
Number of designs
EEG.
Figure 2. Signal enhancement, feature selection/dimensionality
reduction and post-processing methods in BCI designs.
5.1.2. Common spatial patterns (CSP). CSP is a signal
5.1. Signal enhancement enhancement method that detects patterns in the EEG by
Signal enhancement (pre-processing) algorithms have been incorporating the spatial information of the EEG signal. Some
used for brain–computer interfaces that are based on the of its features and limitations include the following.
EEG and the activity of the neural cells (ANC), but An advantage of the CSP method is that it does not
no signal enhancement algorithms have been applied on require the a priori selection of subject-specific frequency
electrocorticogram (ECoG)-based brain–computer interfaces. bands. Knowledge of these bands, however, is necessary
Given the huge difference in the characteristics of EEG and for the band-power and frequency-estimation methods (Guger
ANC, the signal enhancement algorithms used in EEG-based et al 2000b). One disadvantage of the CSP method is
and ANC-based BCIs have very little overlap; only PCA that it requires the use of many electrodes. However, the
has been used in both groups. While ANC-based BCIs inconvenience of applying more electrodes can be rationalized
mostly aim at spike detection, ranking, and sorting neuron by improved performance (Guger et al 2000b, Pfurtscheller
activities, EEG-based ones mostly transform or select EEG et al 2000). The major problem in the application of CSP is
channels that yield better performance. Overall, the use of its sensitivity to artifacts in the EEG. Since the covariance
a pre-processing stage before feature extraction (if applied) matrices are used as the basis for calculating the spatial
has been proven to be useful. The choice of a suitable filters, and are estimated with a comparatively small number
pre-processing technique however is dependent on several of examples, a single trial contaminated with artifacts can
factors such as the recording technology, number of electrodes, unfortunately cause extreme changes to the filters (Guger
and neuromechanism of the BCI. Next, we discuss some of et al 2000b, Ramoser et al 2000). Since the CSP method
the techniques used in signal enhancement in EEG-based detects spatial patterns in the EEG, any change in the electrode
BCI systems. Specifically, a discussion of spatial filtering positions may render the improvements in the classification
including referencing methods and common spatial patterns accuracy gained by this method useless. Therefore, this
(CSP) is presented. These methods are among the most
method requires almost identical electrode positions for all
used techniques that have become increasingly popular in BCI
trials and sessions which may be difficult to accomplish
studies.
(Ramoser et al 2000).

5.1.1. Referencing methods. Referencing methods are

considered as spatial filters. The proper selection of a spatial 5.2. Feature extraction
filter for any BCI is determined by the location and extent of the
control signal (e.g. the mu rhythm) and of the various sources In this section we discuss some of the feature extraction
of EEG or non-EEG noise. The latter two are not completely techniques that have received more attention in BCI systems.
defined and presumably vary considerably between different Specifically, time and/or frequency representation methods,
studies and both across and within individuals (McFarland parametric modeling and specific techniques of modeling
et al 1997). neural cortical recordings are discussed.

R43
Topical Review

70

60
Number of designs

50

40

30

20
Multiple Neuromechanisms
10 Activity of Neural Cells
0 Sensorimotor Activity
Response to Mental Tasks
D
PS

VEP
g
in
ds
el

P300
od

ho

n
ar latio

lc.
m

et

SCP
n
m

ca
ric

io
lcu

s
R

t
et

ea

er
la

es
TF

Ca
m

ne
et
rre

ur
ra

er
No
P

at
d
Pa

ra

th
-c
SC

an

fe
pa

O
X

P
ak

LR
th
Pe

or
Hj

Figure 3. Feature extraction methods in BCI designs based on sensorimotor activity, VEP, P300, SCP, response to mental tasks, activity of
neural cells, and multiple neuromechanisms.

5.2.1. Time and/or frequency methods. A signal, as a provides information about the time–frequency interactions
function of time, may be considered as a representation between the components of the input signal. Thus, with the
with perfect temporal resolution. The magnitude of the CTFR the EEG data samples are not independently analyzed
Fourier transform (FT) of the signal may be considered (as in the Fourier transform case) but their relationship is also
as a representation with perfect spectral resolution but taken into account. One drawback of the CTFR resides in
with no temporal information. Frequency-based features its relative high sensitivity to noise. Consequently, the most
have been widely used in signal processing because of important values of the CTFR in terms of classification must
their ease of application, computational speed and direct be selected (Garcia et al 2003a, 2003b).
interpretation of the results. Specifically, about one-third
of BCI designs have used power-spectral features. Due 5.2.2. Parametric modeling. Parametric approaches assume
to the non-stationary nature of the EEG signals, these the time series under analysis to be the output of a given linear
features do not provide any time domain information. Thus, mathematical model. They require an a priori choice of the
mixed time–frequency representations (TFRs) that map a structure and order of the signal generation mechanism model
one-dimensional signal into a two-dimensional function of (Weitkunat 1991). The optimum model order is best estimated
time and frequency are used to analyze the time-varying not only by maximizing the fitness but also by limiting the
spectral content of the signals. It has been shown model’s complexity. For noisy signals, if the model’s order
that TFR methods may yield performance improvements is too high, spurious peaks in the spectra will result. On
comparing to the traditional FT-based methods (e.g. Qin the other hand, if the order is too low, smooth spectra are
et al (2005) and Bostanov (2004)). Most of the designs that obtained (Kelly et al 2002a, Polak and Kostov 1998, Weitkunat
employ TFR methods use wavelet-based feature extraction 1991).
algorithms. The choice of the particular wavelet used is For short EEG segments, parametric modeling results in
a crucial factor in gaining useful information from wavelet better frequency resolution and a good spectral estimate. Note
analysis. Prior knowledge of the physiological activity in the that parametric modeling may yield poor estimates if the length
brain can be useful in determining the appropriate wavelet of the EEG segments processed is too short (Birch 1988). For
function. such modeling, there is no need for a priori information about
Correlative TFR (CTFR) is another time–frequency potential frequency bands, and there is no need to window
representation method that, besides the spectral information, the data in order to decrease the spectral leakage. Also, the

R44
 Topical Review

60

50
Number of designs

40

30

20

10

0
A
LD

D
FL

M
SV

M
M
H

er
ifi

ck
s
as

ba
cl

or
ed
ed

ct
fe

P
te
as

SS
us

de
-b

g
uo
D

in
ld
M

tin

el
ho

)
N
od
on

es

-N

)
M

ANC-based BCIs
LP
C

Tr

on
ut

(M
(n
/O

n)

non-ANC-based BCIs
er

io
In

N
th

at

)
Q
O

LV
or

)
er
S)
nf

th
li

((D
ra

(O
po

N
N
em

N
(T
N
N

Figure 4. Feature classification methods in BCI designs. ‘In/out modeling’ refers to those ANC-based BCIs that directly map the input
neural recordings to the output without using a feature extraction technique.

frequency resolution does not depend on the number of data extraction algorithms for cortical control of arm prosthetics
points (Guger et al 2003a, Polak and Kostov 1998, Weitkunat can be found in Schwartz et al (2001).
1991). Estimating these parameters, however, is very sensitive One limitation of linear filter methods is that they rely on
to artifacts (Birch 1988, Guger et al 2003a). an a priori model of movement-related neuronal responses.
Special attention should be paid to the choice of the Artificial neural network (ANN) solutions can optimize each
sampling rate in parametric modeling (Weitkunat 1991), since cell’s contribution to the population prediction (Schwartz et al
severely oversampled signals tend to show only very small 2001).
amplitude differences between successive samples. Hence, Several of the algorithms used are based on the position
low-order models produce small prediction errors, giving of the moving limb. In the primary motor cortex at least, this
parameter is more poorly represented than the velocity during
the false illusion that an adequate model has been obtained.
movement. With most algorithms, the different sources of
The sampling rates dictated by the Nyquist criterion are
variability need to be specified explicitly because some sort
recommended.
of optimal function is being modeled to the cell response
(Schwartz et al 2001).
The performance of the modeling techniques is
5.2.3. Modeling the neural firing rates. Extraction constrained by their training sets and may be limited, both
algorithms for motor control operate on spike trains, recorded in terms of extrapolation beyond and interpolation within
from a population of cortical units, mostly with the purpose of the training set when new data are applied. The success
predicting arm trajectories. Several extraction methods such as of the linear filters is due to the underlying linearity of
linear filtering methods and neural networks have been used to the relationship between firing rate and movement direction.
determine arm movement trajectories from neural firing rates. These filters are limited by the conditions used to fit their
We summarize below a few important issues in modeling the coefficients and may suffer from the same training constraints
neural firing rates. A more detailed critical discussion of as ANNs (Schwartz et al 2001).

R45
Topical Review

5.3. Feature selection/dimensionality reduction non-trivial problem. Definition of the fitness function is not
straightforward in many cases and often is performed
Feature selection algorithms are used in BCI designs to
iteratively if the fittest solutions produced by a GA are not
find the most informative features for classification. This
what is desired.
is especially useful for BCI designs with high dimensional
input data, as it reduces the dimension of the feature space.
Since feature selection block reduces the complexity of 5.4. Feature classification
the classification problem, higher classification accuracies Linear classifiers are generally more robust than nonlinear
might be achieved. The experiments carried out in ones. This is because linear classifiers have fewer free
Flotzinger et al (1994) and Pregenzer and Pfurtscheller
parameters to tune, and are thus less prone to over-fitting
(1999) show that when feature selection is used, the
(Muller et al 2003a). In the presence of strong noise and
classification accuracy is better than when all the features are
outliers, even linear systems can fail. One way of overcoming
used.
this problem is to use regularization. Regularization helps limit
Principal component analysis (PCA) and genetic
(a) the influence of outliers and strong noise, (b) the complexity
algorithms (GA) are among the mostly used feature selection
of the classifier and (c) the raggedness of the decision surface
and/or dimensionality reduction methods in BCIs. PCA
(Muller et al 2003a).
has also been widely used in pre-processing stage of BCI
It is always desirable to avoid reliance on nonlinear
designs. PCA is a linear transformation that can be used
classification methods, if possible, because these methods
for dimensionality reduction in a dataset while retaining
often involve a number of parameters whose values must be
those characteristics of the dataset that contribute most to
chosen appropriately. However, when there are large amounts
its variance, by keeping lower-order principal components
of data and limited knowledge of the data, nonlinear methods
and ignoring higher-order ones. Such low-order components
are better suited in finding the potentially more complex
often contain the ‘most important’ aspects of the data. PCA
has the distinction of being the optimal linear transformation structure in the data. In particular, when the source of the
for keeping the subspace that has largest variance. PCA data to be classified is not well understood, using methods
only finds linear subspaces, works best if the individual that are good at finding nonlinear transformation of the data is
components have Gaussian distributtions, and is not optimized suggested. In these cases, kernel-based and neural-networks-
for class separability. One other possible application area based methods can be used to determine the transformations.
of PCA is in classification stage, in which, PCA is applied Kernel-based classifiers are classification methods that apply a
for weighting input features. While a standard neural linear classification in some appropriate (kernel) feature space.
network, such as the multi-layer pereceptrons (MLP), can Thus, all the beneficial properties of linear classification are
do the necessary classification itself, in some cases doing maintained, but at the same time, the overall classification
a PCA in parallel and weighting input features can give is nonlinear. Examples of such kernel-based classification
better results as it simplifies the training of the rest of the methods are support vector machines (SVMs) and kernel
system. Fisher discriminant (KFD) (Muller et al 2003a). For a more
Unlike PCA, GAs are heuristic search techniques in the detailed critical discussion regarding linear and nonlinear
problem space. GAs typically maintain a constant-sized classifiers in brain–computer interfaces, refer to Muller et al
population of individuals which represent samples of the space (2003a).
to be searched. Each individual is evaluated on the basis Some BCI designs have used classification algorithms
of its overall fitness with respect to the given application such as FIR-MLP and TBNN that utilize temporal information
domain. New individuals (samples of the search space) of the input data (Haselsteiner and Pfurtscheller 2000, Ivanova
are produced by selecting high performing individuals to et al 1995). The motivation for using such classifiers
produce ‘offspring’ which retain many of the features of is that the patterns to be recognized are not static data
their ‘parents’. This eventually leads to a population that but time series. Thus, the temporal information of the
has improved fitness with respect to the given goal. Genetic input data can be used to improve the classification results
algorithms have demonstrated substantial improvement over a (Haselsteiner and Pfurtscheller 2000). Utilizing the temporal
variety of random and local search methods (De Jong 1975). information of features is not necessarily performed directly
This is accomplished by their ability to exploit accumulating in the classification stage, and can be done with a static
information about an initially unknown search space in order classifier like MLP and a mapping of the temporal input
to bias subsequent search into promising subspaces. Since data to static data. However, using classifiers such as FIR-
GAs are basically a domain-independent search technique, MLP and TBNN that directly utilize temporal information
they are ideal for applications where domain knowledge may yield better performances as they are much better suited
and theory is difficult or impossible to provide (De Jong for exploiting temporal information contained in the time
1975). An important step in developing a GA-based search series to be classified. Regardless of the method that is
is defining a suitable fitness function. An ideal fitness used for exploiting temporal information, these approaches
function correlates closely with the algorithm’s goal, and are preferred over static classification as they may increase the
yet may be computed quickly. Speed of execution is very performance of BCI systems.
important, as a typical genetic algorithm must be iterated Using a group (committee) of classifiers rather than using
many, many times in order to produce a usable result for a a single classifier might also yield to better performances of

R46
 Topical Review

BCI systems. Only a few BCI designs have employed such an BCI system. Such error potentials have been used in a few
approach in classifying features and achieved performance BCI systems to increase the performance (Bayliss et al 2004,
improvements (Millan et al 2000b, 2002b, Peters et al Blankertz et al 2002b, 2003, Parra et al 2003b, Schalk et al
2001, Varsta et al 2000). The classification accuracy of the 2000).
committee depends on how much unique information each Another useful technique in decreasing false activations
committee member contributes to classification. A committee of BCI systems is to consider a measure of confidence in
of classifiers usually yields better classification accuracy than classification. In such a case, the output of the system can
any individual classifier could provide, and can be used to only be activated when the probability of the output being in
combine information from several channels, i.e., from different an active state is greater than a given probability threshold
spatial regions (Peters et al 2001). or some criterion. Otherwise, the response of the BCI is
As the number of epochs available for evaluating a BCI considered ‘unknown’ and rejected to avoid making risky
system is small, using a technique that reduces the bias of decisions. This is a useful way of reducing false decisions of
the estimated performance on a specific dataset is highly the system (e.g., Cincotti et al (2003b), Millan et al (1998)
recommended. This is especially important when different and Penny et al (2000)) and might be used in any BCI
architectures of a certain design are being compared. K-fold design.
cross-validation and statistical significance tests are especially Considering mechanisms like debouncing the output of
useful for these cases (e.g. refer to Anderson et al (1998), Kelly BCI designs also can reduce the number of false activations
et al (2002b), Lalor et al (2005), Obermaier et al (2001d) and (Bashashati et al 2005, Borisoff et al 2004, Fatourechi et al
Peterson et al (2005)). K-fold cross-validation can be used 2004, 2005, Muller-Putz et al 2005b, Obeid and Wolf 2004,
simply to estimate the generalization error of a given model, Pfurtscheller et al 2005, Townsend et al 2004). These methods
or it can be used for model selection by choosing one of are specifically useful for so called asynchronous (self-paced)
several models that has the smallest estimated generalization BCIs. Since false positives could happen in periods longer
error but it is not suitable for online evaluations. A value of 5 than just a few samples, using the debouncing technique in
to 10 for K is recommended for estimating the generalization a manner similar to the debouncing of physical switches is
error. For an insightful discussion of the limitations of cross- expected to improve false activation rates (but with a cost
validatory choice among several evaluation methods, see Stone in decreased re-activation time). The debounce component
(1977). continuously monitors the output of the classifier. After an
activation is detected (e.g. a change in logical state from
‘0’ to ‘1’ in a binary classifier), the output is activated for
5.5. Post-processing
one time sample, then the output is forced to an inactive
Post-processing techniques can be utilized in most of the BCI state for Td − 1 time samples, where Td is the debounce
designs to decrease the error rates. Some post-processing time period in samples. In some studies this time period is
techniques can be designed specifically for a target application. referred to as refractory period. As the debounce period is
For example, when a BCI system is used to activate a spelling increased, the false activation rate is decreased for a given
device, some letters can be omitted without losing information. true positive rate. However, with increasing this time period,
The system can also take into consideration the conditional the re-activation time of the BCI system is impacted. The
probabilities of letters provided by one or two preceding letters trade-off is clear and one needs to consider this for a given
and make corresponding suggestions to the patient (Kubler application.
et al 1999). Such techniques may also be feasible for other
applications and consequently increase the performance of the
BCI systems. 6. Conclusions
There is a possibility that just after the end of a trial,
some features of the brain signal reveal whether or not the We have completed the first comprehensive survey of signal
trial was successful (that is, whether the outcome was or was processing methods used in BCI studies and published
not what the subject desired). These features are referred to prior to January 2006. The results of this survey form a
as error potentials and can be used to detect errors in a BCI valuable and historical cross-reference for methods used in
system and void the outcome. This error detection approach the following signal processing components of a BCI design:
was encouraged by evidence that errors in conventional motor (1) pre-processing (signal enhancement), (2) feature
performances have detectable effects on the EEG recorded just selection/dimensionality reduction, (3) feature extraction,
after the error occurs (Falkenstein et al 1995, Falkenstein et al (4) feature classification and (5) post-processing methods.
2001, Gehring et al 1995). Whatever the nature of the error This survey shows which signal processing techniques have
potential, the central decision for a BCI is how useful the error received more attention and which have not. This information
potential can be in detecting errors in single trials, and thereby is also valuable for newcomers to the field, as they can now
improving accuracy. While its signal-to-noise ratio (SNR) find out which signal processing methods have been used for
is low, the error potential can improve the performance of a a certain type of a BCI system.
BCI system. In the meantime, better methods for recognizing Many signal processing methods have been proposed
and measuring the error potential could substantially improve and implemented in various brain–computer interfaces and
its SNR, and thereby increase its impact on accuracy of a comparison of these methods for different BCI applications

R47
Topical Review

would be a useful task. However, at this point we cannot

perform this task given the diversity of brain–computer Index term Description
interface systems from different aspects such as target ERS Event-related synchronization
application, neuromechanism used, amount of data tested, FLD Fisher’s linear discriminat
number of subjects and the amount of training they have FFT Fast Fourier transform
received, recording systems, and experimental paradigms. Freq-Norm Frequency normalization
GA Genetic algorithm
Also acknowledged in McFarland et al (2006), we think that
GAM Generalized additive models
comparison of methods would be possible in well-designed GLA Generalized linear models
systematic studies (Jackson et al 2006) and on established GPER Gaussian process entropy rates
datasets such as the BCI Competition datasets (Blankertz et al HMM Hidden Markov model
2004, 2006). We believe that, for a fair comparison of methods, ICA Independent component analysis
more data would be needed as comparing methods with the IFFT Inverse fast Fourier transform
data of one or two subjects does not necessarily guarantee k-NN k-nearest neighbor
the same findings on a larger subject pool. Jackson et al LDA Linear discriminant analysis
LDS Linear dynamical system
(2006) have provided a step toward this goal by proposing
LGM Linear Gaussian models implemented by Kalman
some ways to establish a systematic study both in design and filter
reporting the results and we think that this task would only be LMS Least mean square
possible with the collective help of all the researchers in this LPC Linear predictive coding
field. LPF Low-pass filter
We hope that this study will spawn further discussion of LRP Lateralized readiness potential
signal processing schemes for BCI designs. The proposed LVQ Learning vector quantization
taxonomy and classes defined in tables 3–8 represent a MD Mahalanobis distance
MLP Multi-layer perceptron neural networks
proposed set of subcategories, not a final one, and we
MN Multiple neuromechanisms
encourage others to revise or expand upon this initial MNF Maximum noise fraction
set. Our direction in the future is to establish an online MRA Movement-related activity
public-accessible database where research groups will be NID3 –
able to submit their signal processing designs as well as NMF Non-negative matrix factorization
propose revisions/expansions of the proposed definitions and NN Neural networks
categories presented in this paper. OLS1 Orthogonal least square
OPM Outlier processing method
PCA Principal component analysis (a.k.a. Karhounen
Appendix A. Index of terms Loeve transform)
PLV Phase locking values
PPM Piecewise Prony method
Index term Description PSD Power-spectral density
AAR Adaptive auto-regressive RBF Radial basis function
AEP Auditory evoked potential RFE Recursive feature/channel elimination
AGR Adaptive Gaussian representation RNN Recurrent neural network
ALN Adaptive logic network SA-UK Successive averaging and/or considering choice of
ANC Activity of neural cells unknown
ANN Artificial neural networks SCP Slow cortical potentials
AR Auto-regressive SE Spectral-entropy
ARTMAP Adaptive resonance theory MAP SFFS Sequential forward feature selection
ARX Auto-regressive with exogenous input SL Surface Laplacian
BPF Band-pass filter SOFNN Self-organizing feature neural network
C4.5 – SOM Self-organizing map
CAR Common average referencing SSEP Somatosensory evoked potential
CBR Changes in brain rhythms SSP Signal space projection
CCTM Cross-correlation-based template matching SSVEP Steady state visual evoked potential
CER Coarse-grained entropy rate STD Standard deviation
CHMM Coupled hidden Markov model SVD Singular value decomposition
CN2 – SVM Support vector machine
CSP Common spatial patterns SVR Support vector machine regression
CSSD Common spatial subspace decomposition SWDA Stepwise discriminant analysis
CSSP Common spatio-spectral patterns TBNN Tree-based neural network
CTFR Correlative time–frequency representation TFR Time–frequency representation
CTFSR Correlative time–frequency–space representation VEFD Variable epoch frequency decomposition
DFT Discrete Fourier transform VEP Visual evoked potential
DSLVQ Distinctive sensitive learning vector quantization WE Wavelet entropy
ERD Event-related desynchronization WK Wiener–Khinchine
ERN Event-related negativity ZDA Z-scale-based discriminant analysis

R48
 Topical Review

References Bayliss J D and Ballard D H 2000a Recognizing evoked potentials

in a virtual environment Advances in Neural Information
Processing Systems vol 12, ed S A Solla, T K Leen and K R
Allison B Z and Pineda J A 2003 ERPs evoked by different matrix Müller (Cambridge, MA: MIT Press) pp 3–9
sizes: implications for a brain–computer interface (BCI) Bayliss J D and Ballard D H 2000b A virtual reality testbed for
system IEEE Trans. Neural Syst. Rehabil. Eng. 11 110–3 brain–computer interface research IEEE Trans. Rehabil. Eng.
Allison B Z and Pineda J A 2005 Effects of SOA and flash pattern 8 188–90
manipulations on ERPs, performance, and preference: Bayliss J D, Inverso S A and Tentler A 2004 Changing the
implications for a BCI system Int. J. Psychophysiol. 59 127–40 P300 brain–computer interface Cyberpsychol.
Anderson C W, Devulapalli S V and Stolz E A 1995a Signal Behav. 7 694–704
classification with different signal representations Proc. IEEE Birbaumer N, Ghanayim N, Hinterberger T, Iversen I,
Workshop on Neural Networks for Signal Processing Kotchoubey B, Kubler A, Perelmouter J, Taub E and Flor H
pp 475–83 1999 A spelling device for the paralysed Nature 398 297–8
Anderson C W, Stolz E A and Shamsunder S 1995b Discriminating Birbaumer N, Kubler A, Ghanayim N, Hinterberger T,
mental tasks using EEG represented by AR models Proc. 17th Perelmouter J, Kaiser J, Iversen I, Kotchoubey B, Neumann N
Annual Int. Conf. of the IEEE Engineering in Medicine and and Flor H 2000 The thought translation device (TTD) for
Biology Society (Montreal) pp 875–6 completely paralyzed patients IEEE Trans. Rehabil. Eng.
Anderson C W, Stolz E A and Shamsunder S 1998 Multivariate 8 190–3
autoregressive models for classification of spontaneous Birch G E 1988 Single trial EEG signal analysis using outlier
electroencephalographic signals during mental tasks IEEE information PhD Thesis The University of British Columbia,
Trans. Biomed. Eng. 45 277–86 Vancouver, Canada
Babiloni C, Babiloni F, Carducci F, Cappa S F, Cincotti F, Birch G E, Bozorgzadeh Z and Mason S G 2002 Initial on-line
Del Percio C, Miniussi C, Moretti D V, Rossi S, Sosta K and evaluations of the LF-ASD brain–computer interface with
Rossini P M 2004 Human cortical responses during one-bit able-bodied and spinal-cord subjects using imagined voluntary
short-term memory. A high-resolution EEG study on delayed motor potentials IEEE Trans. Neural Syst. Rehabil. Eng.
choice reaction time tasks Clin. Neurophysiol. 115 161–70 10 219–24
Babiloni F, Bianchi L, Semeraro F, Millan J R, Mourino J, Birch G E, Mason S G and Borisoff J F 2003 Current trends in
Cattini A, Salinari S, Marciani M G and Cincotti F 2001a brain–computer interface research at the neil squire foundation
Mahalanobis distance-based classifiers are able to recognize IEEE Trans. Neural Syst. Rehabil. Eng. 11 123–6
EEG patterns by using few EEG electrodes Proc. 23rd Annual Black M J, Bienenstock E, Donoghue J P, Wu W and Gao Y 2003
Int. Conf. of the IEEE Engineering in Medicine and Biology Connecting brains with machines: the neural control of 2D
Society (Istanbul) pp 651–4 cursor movement Proc. 2nd IEEE-EMBS Conf. on Neural
Babiloni F, Cincotti F, Bianchi L, Pirri G, Millan J R, Mourino J, Engineering (Arlington, VA) pp 580–3
Salinari S and Marciani M G 2001b Recognition of imagined Blanchard G and Blankertz B 2004 BCI competition 2003–data set
hand movements with low resolution surface laplacian and IIa: Spatial patterns of self-controlled brain rhythm
linear classifiers Med. Eng. Phys. 23 323–8 modulations IEEE Trans. Biomed. Eng. 51 1062–6
Babiloni F, Cincotti F, Lazzarini L, Millan J, Mourino J, Varsta M, Blankertz B, Curio G and Muller K R 2002a Classifying single trial
Heikkonen J, Bianchi L and Marciani M G 2000 Linear EEG: Towards brain–computer interfacing Advances in Neural
classification of low-resolution EEG patterns produced by Information Processing Systems vol 14, ed T G Dietterich,
imagined hand movements IEEE Trans. Rehabil. Eng. S Becker and Z Ghahramani (Cambridge, MA: MIT Press)
8 186–8 pp 157–64
Balbale U H, Higgins J E, Bement S L and Levine S P 1999 Blankertz B, Dornhege G, Schñfer C, Krepki R, Kolmorgen J,
Multi-channel analysis of human event-related cortical Muller K R, Kunzmann V, Losch F and Curio G 2003 Boosting
potentials for the development of a direct brain interface Proc. bit rates and error detection for the classification of fast-paced
21st Annual Int. Conf. of the IEEE Engineering in Medicine motor commands based on single-trial EEG analysis IEEE
and Biology Society & Annual Fall meeting of the Biomedical Trans. Neural Syst. Rehabil. Eng. 11 127–31
Engineering Society (Atlanta, GA) p 447 Blankertz B, Muller K R, Curio G, Vaughan T M, Schalk G,
Barreto A B, Taberner A M and Vicente L M 1996a Classification of Wolpaw J R, Schlogl A, Neuper C, Pfurtscheller G,
spatio-temporal EEG readiness potentials towards the Hinterberger T, Schroder M and Birbaumer N 2004 The BCI
development of a brain–computer interface Proc. IEEE competition 2003: progress and perspectives in detection and
Southeast Conf. (Raleigh, NC) pp 99–102 discrimination of EEG single trials IEEE Trans. Biomed. Eng.
Barreto A B, Taberner A M and Vicente L M 1996b Neural network 51 1044–51
classification of spatio-temporal EEG readiness potentials Blankertz B, Muller K R, Krusienski D J, Schalk G, Wolpaw J R,
Proc. IEEE 15th Southern Biomedical Engineering Conf. Schlogl A, Pfurtscheller G, Millan J R, Schroder M and
(Dayton, OH) pp 73–6 Birbaumer N 2006 The BCI competition: III. Validating
Bashashati A, Ward R K and Birch G E 2005 A new design of the alternative approaches to actual BCI problems IEEE Trans.
asynchronous brain–computer interface using the knowledge of Neural Syst. Rehabil. Eng. 14 153–9
the path of features Proc. 2nd IEEE-EMBS Conf. on Neural Blankertz B, Schñfer C, Dornhege G and Curio G 2002b Single trial
Engineering (Arlington, VA) pp 101–4 detection of EEG error potentials: a tool for increasing BCI
Bashashati A, Ward R K, Birch G E, Hashemi M R and transmission rates Proc. Int. Conf. on Artificial Neural
Khalilzadeh M A 2003 Fractal dimension-based EEG Networks (Madrid, Spain) pp 1137–43
biofeedback system Proc. 25th Annual Int. Conf. of the IEEE Boostani R and Moradi M H 2004 A new approach in the BCI
Engineering in Medicine and Biology Society (Cancun, research based on fractal dimension as feature and adaboost as
Mexico) pp 2220–3 classifier J. Neural Eng. 1 212–7
Bayliss J D 2003 Use of the evoked potential P3 component for Borisoff J F, Mason S G, Bashashati A and Birch G E 2004
control in a virtual apartment IEEE Trans. Neural Syst. Brain–computer interface design for asynchronous control
Rehabil. Eng. 11 113–6 applications: Improvements to the LF-ASD asynchronous brain
Bayliss J D and Ballard D H 1999 Single trial P300 recognition in a switch IEEE Trans. Biomed. Eng. 51 985–92
virtual environment Proc. Int. ICSC Symp. on Soft Computing Bostanov V 2004 BCI competition 2003–data sets ib and IIb:
in Biomedicine (Genova, Italy) feature extraction from event-related brain potentials with the

R49
Topical Review

continuous wavelet transform and the t-value scalogram IEEE Costa E J and Cabral E F 2000 EEG-based discrimination between
Trans. Biomed. Eng. 51 1057–61 imagination of left and right hand movements using adaptive
Bozorgzadeh Z, Birch G E and Mason S G 2000 The LF-ASD BCI: gaussian representation Med. Eng. Phys. 22 345–8
on-line identification of imagined finger movements in Coyle D, Prasad G and McGinnity T M 2004 Extracting features for
spontaneous EEG with able-bodied subjects Proc. IEEE Int. a brain–computer interface by self-organising fuzzy neural
Conf. on Acoustics, Speech and Signal Processing (Istanbul, network-based time series prediction Proc. 26th Annual Int.
Turkey) pp 109–13 Conf. of the IEEE Engineering in Medicine and Biology
Burke D, Kelly S, de Chazal P and Reilly R 2002 A simultaneous Society (San Francisco, CA) pp 4371–4
filtering and feature extraction strategy for direct brain Coyle D, Prasad G and McGinnity T M 2005 A time-frequency
interfacing Proc. 24th Annual Int. Conf. of the IEEE approach to feature extraction for a brain–computer interface
Engineering in Medicine and Biology Society (Houston, TX) with a comparative analysis of performance measures
pp 279–80 EURASIP J. Appl. Signal Process. 2005 3141–51
Burke D P, Kelly S P, de Chazal P, Reilly R B and Finucane C 2005 Curran E, Sykacek P, Stokes M, Roberts S J, Penny W, Johnsrude I
A parametric feature extraction and classification strategy for and Owen A M 2004 Cognitive tasks for driving a
brain–computer interfacing IEEE Trans. Neural Syst. Rehabil. brain–computer interfacing system: a pilot study IEEE Trans.
Eng. 13 12–7 Neural Syst. Rehabil. Eng. 12 48–54
Byron O, He J P, Hu J and Si J 2005 A conceptual brain machine Darmanjian S, Kim S P, Nechbya M C, Morrison S, Principe J,
interface system Proc. 1st Int. Conf. on Neural Interface and Wessberg J and Nicolelis M A 2003 Bimodal brain-machine
Control (Wuhan,China) pp 40–44 interface for motor control of robotic prosthetic Proc.
Calhoun G L and McMillan G R 1996 EEG-based control for IEEE/RSJ Int. Conf. on Intelligent Robots and Systems
human-computer interaction Proc. 3rd Symp. On Human (Las Vegas, NV) pp 3612–7
Interaction with Complex Systems (Dayton, OH) pp 4–9 De Jong K A 1975 An analysis of the behaviour of a class of genetic
Carmena J M, Lebedev M A, Crist R E, O’Doherty J E, adaptive systems PhD Thesis University of Michigan, Ann
Santucci D M, Dimitrov D F, Patil P G, Henriquez C S and Arbor, MI (Diss. Abstr. Int. 36(10), 5140B, University
Nicolelis M A 2003 Learning to control a brain-machine Microfilms No. 76–9381. 44)
interface for reaching and grasping by primates Plos Biol. Deecke L and Kornhuber H 1976 Voluntary finger movement in
1 193–208 man: Cerebral potentials and theory Biol. Cybern. 23 99–119
Carmena J M, Lebedev M A, Henriquez C S and Nicolelis M A Donchin E, Spencer K M and Wijesinghe R 2000 The mental
2005 Stable ensemble performance with single-neuron prosthesis: assessing the speed of a P300-based
variability during reaching movements in primates J. Neurosci. brain–computer interface IEEE Trans. Rehabil. Eng. 8 174–9
25 10712–6 Donoghue J P 2002 Connecting cortex to machines: recent advances
Chapin J K, Moxon K A, Markowitz R S and Nicolelis M A 1999 in brain interfaces Nat. Neurosci. 5 1085–8
Real-time control of a robot arm using simultaneously recorded Dornhege G, Blankertz B and Curio G 2003 Speeding up
neurons in the motor cortex Nat. Neurosci. 2 664–70 classification of multi-channel brain–computer interfaces:
Cheng M and Gao S 1999 An EEG-based cursor control system common spatial patterns for slow cortical potentials Proc. 2nd
Proc. 21st Annual Int. Conf. of the IEEE Engineering in IEEE-EMBS Conf. on Neural Engineering (Arlington, VA)
Medicine and Biology Society & Annual Fall meeting of the pp 595–8
Biomedical Engineering Society (Atlanta, GA) p 669 Dornhege G, Blankertz B, Curio G and Muller K R 2004 Boosting
Cheng M, Gao X, Gao S and Wang B 2005 Stimulation frequency bit rates in noninvasive EEG single-trial classifications by
extraction in SSVEP-based brain–computer interface Proc. 1st feature combination and multiclass paradigms IEEE Trans.
Int. Conf. on Neural Interface and Control (Wuhan, China) Biomed. Eng. 51 993–1002
pp 64–7 Erfanian A and Erfani A 2004 ICA-based classification scheme for
Cheng M, Gao X, Gao S and Xu D 2002 Design and EEG-based brain–computer interface: the role of mental
implementation of a brain–computer interface with high practice and concentration skills Proc. 26th Annual Int. Conf.
transfer rates IEEE Trans. Biomed. Eng. 49 1181–6 of the IEEE Engineering in Medicine and Biology Society
Cheng M, Jia W, Gao X, Gao S and Yang F 2004 Mu rhythm-based (San Francisco, CA) pp 235–8
cursor control: an offline analysis Clin. Neurophysiol. Fabiani G E, McFarland D J, Wolpaw J R and Pfurtscheller G 2004
115 745–51 Conversion of EEG activity into cursor movement by a
Cheng M, Xu D, Gao X and Gao S 2001 Brain–computer interface brain–computer interface (BCI) IEEE Trans. Neural Syst.
with high transfer rates Proc. 8th Int. Conf. on Neural Rehabil. Eng. 12 331–8
Information Processing (Shanghai, China) Falkenstein M, Hielscher H, Dziobek I, Schwarzenau P, Hoormann
Cho B H, Kim S, Shin D I, Lee J H, Lee S M, Kim I Y and Kim S I J, Sunderman B and Hohnsbein J 2001 Action monitoring,
2004 Neurofeedback training with virtual reality for inattention error detection, and the basal ganglia: an ERP study
and impulsiveness Cyberpsychol. Behav. 7 519–26 Neuroreport 12 157–61
Cincotti F, Bianchi L, Millan J, Mourino J, Salinari S, Falkenstein M, Hohnsbein J and Hoormann J 1995 Event-related
Marciani M G and Babiloni F 2001 Brain–computer interface: potential correlates of errors in reaction tasks Perspectives of
the use of low resolution surface laplacian and linear classifiers Event-Related Brain Potentials Research ed Karmos et al
for the recognition of imagined hand movements Proc. 23rd (Amsterdam: Elsevier) pp 287–96
Annual Int. Conf. of the IEEE Engineering in Medicine and Farwell L A and Donchin E 1988 Talking off the top of your head:
Biology Society (Istanbul) pp 655–8 Towards a mental prosthesis utilizing event-related brain
Cincotti F, Mattia D, Babiloni C, Carducci F, Salinari S, Bianchi L, potentials Electroencephalogr. Clin. Neurophysiol. 80 510–23
Marciani M G and Babiloni F 2003a The use of EEG Fatourechi M, Bashashati A, Borisoff J F, Birch G E and Ward R K
modifications due to motor imagery for brain–computer 2004 Improving the performance of the LF-ASD
interfaces IEEE Trans. Neural Syst. Rehabil. Eng. 11 131–3 brain–computer interface by means of genetic algorithm Proc.
Cincotti F, Scipione A, Timperi A, Mattia D, Marciani M G, IEEE Symp. on Signal Processing and Information Technology
Millan J, Salinari S, Bianchi L and Babiloni F 2003b (Rome, Italy) pp 38–41
Comparison of different feature classifiers for brain–computer Fatourechi M, Bashashati A, Ward R K and Birch G E 2005
interfaces Proc. 25th Annual Int. Conf. of the IEEE Engineering A hybrid genetic algorithm approach for improving the
in Medicine and Biology Society (Cancun, Mexico) pp 645–7 performance of the LF-ASD brain–computer interface Proc.

R50
 Topical Review

IEEE Int. Conf. on Acoustics, Speech and Signal Processing channels Proc. 1st IEEE-EMBS Conf. on Neural Engineering
(Philadelphia, PA) pp 345–8 (Capri Island, Italy) pp 614–7
Flotzinger D, Pregenzer M and Pfurtscheller G 1994 Feature Graimann B, Huggins J E, Levine S P and Pfurtscheller G 2004
selection with distinction sensitive learning vector quantisation Toward a direct brain interface based on human subdural
and genetic algorithms Proc. IEEE Int. Conf. on Neural recordings and wavelet-packet analysis IEEE Trans. Biomed.
Networks (Orlando, FL) pp 3448–51 Eng. 51 954–62
Fukada S, Tatsumi D, Tsujimoto H and Inokuchi S 1998 Studies of Graimann B, Huggins J E, Schlogl A, Levine S P and
input speed of word inputting system using event-related Pfurtscheller G 2003b Detection of movement-related
potential Proc. 20th Annual Int. Conf. of the IEEE desynchronization patterns in ongoing single-channel
Engineering in Medicine and Biology Society (Hong Kong) electrocorticogram IEEE Trans. Neural Syst. Rehabil. Eng.
pp 1458–60 11 276–81
Gao Y, Black M J, Bienenstock E, Shoham S and Donoghue J P Guan J, Chen Y, Lin J, Yun Y and Huang M 2005 N2 components
2002 Probabilistic inference of hand motion from neural as features for brain–computer interface Proc. 1st Int. Conf. on
activity in motor cortex Advances in Neural Information Neural Interface and Control (Wuhan, China) pp 45–9
Processing Systems vol 14, ed T G Dietterich, S Becker and Guan C, Thulasidas M and Wu J 2004 High performance P300
Z Ghahramani (Cambridge, MA: MIT Press) pp 213–20 speller for brain–computer interface Proc. IEEE Int.
Gao Y, Black M J, Bienenstock E, Wu W and Donoghue J P 2003a Workshop on Biomedical Circuits and Systems (Singapore)
A quantitative comparison of linear and non-linear models of pp S3/5/INV-S3/13–16
motor cortical activity for the encoding and decoding of arm Guger C, Edlinger G, Harkam W, Niedermayer I and
motions Proc. 25th Annual Int. Conf. of the IEEE Engineering Pfurtscheller G 2003a How many people are able to operate an
in Medicine and Biology Society (Hong Kong) pp 189–92 EEG-based brain–computer interface (BCI)? IEEE Trans.
Gao X, Xu D, Cheng M and Gao S 2003b A BCI-based Neural Syst. Rehabil. Eng. 11 145–7
environmental controller for the motion-disabled IEEE Trans. Guger C, Edlinger G and Pfurtscheller G 2003b How many people
Neural Syst. Rehabil. Eng. 11 137–40 are able to operate an EEG-based brain–computer interface
Gao X, Xu N, Hong B, Gao S and Yang F 2004 Optimal selection of (BCI)? Presented at the 2nd Int. Meeting on Brain–Computer
independent components for event-related brain electrical Interfaces for Communication and Control (Albany, NY)
potential enhancement Proc. IEEE Int. Workshop on Guger C, Muller G, Neuper C, Krausz G, Niedermayer I and
Biomedical Circuits and Systems (Singapore) Pfurtscheller G 2000a Brain–computer communication
pp S3/5/INV–S3/5/1-4 system: the EEG-based control of a hand orthesis in a
Garcia G and Ebrahimi T 2002 Time–frequency–space kernel for quadriplegic patient Presented at the Int. Conf. on
single EEG-trial classification Presented at the 5th Nordic Computers Helping People with Special Needs (Karlsruhe,
Signal Processing Symp. (Norway) Germany)
Garcia G, Ebrahimi T and Vesin J M 2002 Classification of EEG Guger C, Ramoser H and Pfurtscheller G 2000b Real-time EEG
signals in the ambiguity domain for brain–computer interface analysis with subject-specific spatial patterns for a
applications Proc. IEEE 14th Int. Conf. on Digital Signal brain–computer interface (BCI) IEEE Trans. Rehabil. Eng.
Processing (Santorini, Greece) pp 301–305 8 447–56
Garcia G, Ebrahimi T and Vesin J M 2003a Correlative exploration Guger C, Schloegl A, Walterspacher D and Pfurtscheller G 1999
of EEG signals for direct brain–computer communications Design of an EEG-based brain–computer interface (BCI) from
Proc. IEEE Int. Conf. on Acoustics, Speech, and Signal standard components running in real-time under windows
Processing (Hong Kong) pp 816–9 Biomed. Tech. (Berl) 44 12–6
Garcia G, Ebrahimi T and Vesin J M 2003b Support vector EEG Gysels E and Celka P 2004 Phase synchronization for the
classification in the fourier and time-frequency correlation recognition of mental tasks in a brain–computer interface
domains Proc. 1st IEEE-EMBS Conf. on Neural Engineering IEEE Trans. Neural Syst. Rehabil. Eng. 12 406–15
(Capri Island, Italy) pp 591–4 Gysels E, Renevey P and Celka P 2005 SVM-based recursive
Garcia G N, Ebrahimi T, Vesin J M and Villca A 2003c Direct feature elimination to compare phase synchronization
brain–computer communication with user rewarding computed from broadband and narrowband EEG signals in
mechanism Proc. IEEE Int. Symp. in Information Theory brain–computer interfaces Signal Process. 85 2178–89
(Yokohama, Japan) p 221 Hallett M 1994 Movement-related cortical potentials Electromyogr.
Garrett D, Peterson D A, Anderson C W and Thaut M H 2003 Clin. Neurophysiol. 34 5–13
Comparison of linear, nonlinear, and feature selection methods Haselsteiner E and Pfurtscheller G 2000 Using time-dependent
for EEG signal IEEE Trans. Neural Syst. Rehabil. Eng. neural networks for EEG classification IEEE Trans. Rehabil.
11 141–4 Eng. 8 457–63
Gehring W J, Coles M G H, Meyer D E and Donchin E 1995 Hatsopoulos N, Joshi J and O’Leary J G 2004 Decoding continuous
A brain potential manifestation of error-related processing and discrete motor behaviors using motor and premotor
Perspectives of Event-Related Brain Potentials Research (EEG cortical ensembles J. Neurophysiol. 92 1165–74
Suppl vol 44) ed Karmos et al (Amsterdam:Elsevier) pp 261–72 Hill N J, Lal T N, Bierig K, Birbaumer N and Scholkopf B 2004
Georgopoulos A P, Langheim F J, Leuthold A C and Merkle A N Attention modulation of auditory event-related potentials in a
2005 Magnetoencephalographic signals predict movement brain–computer interface Proc. IEEE Int. Workshop on
trajectory in space Exp. Brain Res. 167 132–5 Biomedical Circuits and Systems (Singapore)
Gevins A, Bressler S L, Morgan N H, Cutillo B A, White R M, pp S3/5/INV-S3/17–20
Greer D S and Illes J 1989 Event-related covariances during a Hinterberger T and Baier G 2005 Parametric orchestral sonification
bimanual visuomotor taks: I. Methods and analysis of of EEG in real time IEEE Multimedia 12 70–9
stimulus- and response-locked data Electroencephalogr. Clin. Hinterberger T, Kubler A, Kaiser J, Neumann N and Birbaumer N
Neurophysiol. 74 58–75 2003 A brain–computer interface (BCI) for the locked-in:
Glassman E L 2005 A wavelet-like filter based on neuron action comparison of different EEG classifications for the thought
potentials for analysis of human scalp electroencephalographs translation device Electroencephalogr. Clin. Neurophysiol. 114
IEEE Trans. Biomed. Eng. 52 1851–62 416–25
Graimann B, Huggins J E, Levine S P and Pfurtscheller G 2003a Hinterberger T, Neumann N, Pham M, Kubler A, Grether A,
Detection of ERP and ERD/ERS patterns in single ECoG Hofmayer N, Wilhelm B, Flor H and Birbaumer N 2004a

R51
Topical Review

A multimodal brain-based feedback and communication Kalcher J, Flotzinger D and Pfurtscheller G 1993 Graz
system Exp. Brain Res. 154 521–6 brain–computer interface: an EEG-based cursor control system
Hinterberger T, Veit R, Wilhelm B, Weiskopf N, Vatine J J and Proc. 15th Annual Int. Conf. of the IEEE Engineering in
Birbaumer N 2005a Neuronal mechanisms underlying control Medicine and Biology Society (San Diego, CA) pp 1264–5
of a brain–computer interface Eur. J. Neurosci. 21 3169–81 Kaper M, Meinicke P, Grossekathoefer U, Lingner T and Ritter H
Hinterberger T, Weiskopf N, Veit R, Wilhelm B, Betta E and 2004 BCI competition 2003–data set IIb: support vector
Birbaumer N 2004b An EEG-driven brain–computer interface machines for the P300 speller paradigm IEEE Trans. Biomed.
combined with functional magnetic resonance imaging (fMRI) Eng. 51 1073–6
IEEE Trans. Biomed. Eng. 51 971–4 Kaper M and Ritter H 2004a Generalizing to new subjects in
Hinterberger T, Wilhelm B, Mellinger J, Kotchoubey B and brain–computer interfacing Proc. 26th Annual Int. Conf. of
Birbaumer N 2005b A device for the detection of cognitive the IEEE Engineering in Medicine and Biology Society
brain functions in completely paralyzed or unresponsive (San Francisco, CA) pp 4363–6
patients IEEE Trans. Biomed. Eng. 52 211–20 Kaper M and Ritter H 2004b Progress in P300-based
Hu J, Si J, Olson B P and He J 2004 Principle component feature brain–computer interfacing Proc. IEEE Int. Workshop on
detector for motor cortical control Proc. 26th Annual Int. Conf. Biomedical Circuits and Systems (Singapore)
of the IEEE Engineering in Medicine and Biology Society pp S3/5/INV-S3/59–12
(San Francisco, CA) pp 4021–4 Karniel A, Fleming K M, Sanguineti V, Alford S and
Huan N J and Palaniappan R 2004 Neural network classification of Mussa-Ivaldi F A 2002 Dynamic properties of the lamprey’s
autoregressive features from electroencephalogram signals for neuronal circuits as it drives a two-wheeled robot Proc.
brain–computer interface design J. Neural Eng. 1 142–50 SAB’2002 Workshop on Motor Control in Humans and Robots:
Huan N J and Palaniappan R 2005 Classification of mental tasks on the Interplay of Real Brains and Artificial Devices
using fixed and adaptive autoregressive models of EEG signals (Edinburgh, Scotland) pp 29–36
Proc. 27th Annual Int. Conf. of the IEEE Engineering in Kawakami T, Inoue M, Kobayashi Y and Nakashima K 1996
Medicine and Biology Society (Shanghai, China) pp 633–6 Application of event related potentials to communication aids
Huggins J E et al 2003 Electrocorticogram as the basis for a direct Proc. 18th Annual Int. Conf. of the IEEE Engineering in
brain interface: Opportunities for improved detection accuracy Medicine and Biology Society (Amsterdam, Holland)
Proc. 1st IEEE-EMBS Conf. on Neural Engineering (Capri pp 2229–31
Island, Italy) pp 587–90 Keirn Z A and Aunon J I 1990 A new mode of communication
Huggins J E, Levine S P, Bement S L, Kushwaha R K, Schuh L A, between man and his surroundings IEEE Trans. Biomed. Eng.
Passaro E A, Rohde M M, Ross D A, Elisevich K V and 37 1209–14
Smith B J 1999 Detection of event-related potentials for Kelly S, Burke D, Chazal P d and Reilly R 2002a Parametric models
development of a direct brain interface J. Clin. Neurophysiol. and classification for direct brain interfaces Proc. Irish Signal
16 448–55 and Systems Conf. (Cork, Ireland) pp 25–6
Hung C I, Lee P L, Wu Y T, Chen L F, Yeh T C and Hsieh J C 2005 Kelly S, Burke D, Chazal P d and Reilly R 2002b Parametric models
Recognition of motor imagery electroencephalography using and spectral analysis for classification in brain–computer
independent component analysis and machine classifiers Ann. interfaces Proc. 14th IEEE Int. Conf. on Digital Signal
Biomed. Eng. 33 1053–70 Proceesing (Santorini, Greece) pp 307–10
Isaacs R E, Weber D J and Schwartz A B 2000 Work toward Kelly S P, Lalor E, Finucane C and Reilly R B 2004 A comparison
real-time control of a cortical neural prothesis IEEE Trans of covert and overt attention as a control option in a steady-state
Rehabil. Eng. 8 196–8 visual evoked potential-based brain–computer interface Proc.
Ivanova I, Pfurtscheller G and Andrew C 1995 AI-based 26th Annual Int. Conf. of the IEEE Engineering in Medicine
classification of single-trial EEG data Proc. 17th Annual Int. and Biology Society (San Francisco, CA) pp 4725–8
Conf. of the IEEE Engineering in Medicine and Biology Kelly S P, Lalor E, Reilly R B and Foxe J J 2005a Independent
Society (Montreal, Canada) pp 703–4 brain–computer interface control using visual spatial attention-
Jackson M M, Mason S G and Birch G E 2006 Analyzing trends in dependent modulations of parieto-occipital alpha Proc.
brain interface technology: a method to compare studies Ann. IEEE-EMBS Conf. on Neural Engineering (Arlington, VA)
Biomed. Eng. 34 859–78 pp 667–70
Jansen B H, Allam A, Kota P, Lachance K, Osho A and Kelly S P, Lalor E C, Finucane C, McDarby G and Reilly R B
Sundaresan K 2004 An exploratory study of factors affecting 2005b Visual spatial attention control in an independent
single trial P300 detection IEEE Trans. Biomed. Eng. 51 975–8 brain–computer interface IEEE Trans. Biomed. Eng.
Jasper H and Penfield W 1949 Electrocortiograms in man: effect of 52 1588–96
voluntary movement upon the electrical activity of the Kelly S P, Lalor E C, Reilly R B and Foxe J J 2005c Visual spatial
precentral gyrus Arch. Psychiat. Nervenkr. 183 163–74 attention tracking using high-density SSVEP data for
Jia W, Zhao X, Liu H, Gao X, Gao S and Yang F 2004 Classification independent brain–computer communication IEEE Trans.
of single trial EEG during motor imagery based on ERD Proc. Neural Syst. Rehabil. Eng. 13 172–8
26th Annual Int. Conf. of the IEEE Engineering in Medicine Kemere C, Shenoy K V and Meng T H 2004 Model-based neural
and Biology Society (San Francisco, CA) pp 5–8 decoding of reaching movements: a maximum likelihood
Kaiser J, Kubler A, Hinterberger T, Neumann N and Birbaumer N approach IEEE Trans. Biomed. Eng. 51 925–32
2002 A non-invasive communication device for the paralyzed Kennedy P R, Bakay R A, Moore M M, Adams K and
Minim. Invasive Neurosurg. 45 19–23 Goldwaithe J 2000 Direct control of a computer from the
Kaiser J, Perelmouter J, Iversen I H, Neumann N, Ghanayim N, human central nervous system IEEE Trans. Rehabil. Eng.
Hinterberger T, Kubler A, Kotchoubey B and Birbaumer N 8 198–202
2001 Self-initiation of EEG-based communication in paralyzed Kim K H, Kim S S and Kim S J 2005a Superiority of nonlinear
patients Electroencephalogr. Clin. Neurophysiol. 112 551–4 mapping in decoding multiple single-unit neuronal spike trains:
Kalcher J, Flotzinger D and Pfurtscheller G 1992 A new approach to a simulation study J. Neurosci. Methods 150 202–11
a brain–computer-interface (BCI) based on learning vector Kim S P, Rao Y N, Erdogmus D, Sanchez J C, Nicolelis M A L and
quantization (LVQ3) Proc. 14th Annual Int. Conf. of the IEEE Principe J C 2005b Determining patterns in neural activity for
Engineering in Medicine and Biology Society (Paris, France) reaching movements using nonnegative matrix factorization
pp 1658–9 EURASIP J. Appl. Signal Process. 19 3113–21

R52
 Topical Review

Kira K and Rendell L A 1992 The feature selection problem: Lee P L, Wu C H, Hsieh J C and Wu Y T 2005 Visual evoked
traditional methods and a new algorithm Proc. of the 10th potential actuated brain–computer interface: a brain-actuated
National Conf. on Artificial Intelligence (San Jose, CA) cursor system Electron. Lett. 41 832–4
pp 129–34 Leeb R and Pfurtscheller G 2004 Walking through a virtual city by
Kositsky M, Karniel A, Alford S, Fleming K M and thought Proc. 26th Annual Int. Conf. of the IEEE Engineering
Mussa-Ivaldi F A 2003 Dynamical dimension of a hybrid in Medicine and Biology Society (San Francisco, CA)
neurorobotic system IEEE Trans. Neural Syst. Rehabil. Eng. pp 4503–6
11 155–9 Lemm S, Blankertz B, Curio G and Muller K R 2005 Spatio-spectral
Kostov A and Polak M 1997 Prospects of computer access using filters for improving the classification of single trial EEG IEEE
voluntary modulated EEG signal Proc. ECPD Symposium on Trans. Biomed. Eng. 52 1541–8
Brain & Consciousness (Belgrade, Yugoslavia) pp 233–6 Lemm S, Schafer C and Curio G 2004 BCI competition 2003–data
Kostov A and Polak M 2000 Parallel man-machine training in set III: probabilistic modeling of sensorimotor mu rhythms for
development of EEG-based cursor control IEEE Trans. classification of imaginary hand movements IEEE Trans.
Rehabil. Eng. 8 203–5 Biomed. Eng. 51 1077–80
Kozelka J W and Pedley T A 1990 Beta and mu rhythms J. Clin. Levine S P, Huggins J E, Bement S L, Kushwaha R K, Schuh L A,
Neurophysiol. 7 191–207 Passaro E A, Rohde M M and Ross D A 1999 Identification of
Krauledat M, Dornhege G, Blankertz B, Losch F, Curio G and electrocorticogram patterns as the basis for a direct brain
Muller K 2004 Improving speed and accuracy of interface J. Clin. Neurophysiol. 16 439–47
brain–computer interfaces using readiness potential features Levine S P, Huggins J E, Bement S L, Kushwaha R K, Schuh L A,
Proc. 26th Annual Int. Conf. of the IEEE Engineering in Rohde M M, Passaro E A, Ross D A, Elisevich K V and
Medicine and Biology Society (San Francisco, CA) pp 4511–5 Smith B J 2000 A direct brain interface based on event-related
Krausz G, Scherer R, Korisek G and Pfurtscheller G 2003 Critical potentials IEEE Trans. Rehabil. Eng. 8 180–5
decision-speed and information transfer in the ‘graz Li Y, Cichocki A, Guan C and Qin J 2004a Sparse factorization
brain–computer interface’ Appl. Psychophysiol. Biofeedback preprocessing-based offline analysis for a cursor control
28 233–40 experiment Proc. IEEE Int. Workshop on Biomedical Circuits
Kubler A, Kotchoubey B, Hinterberger T, Ghanayim N, and Systems (Singapore) pp S3/5/INV-S3/5/5–8
Perelmouter J, Schauer M, Fritsch C, Taub E and Birbaumer N Li Y, Gao X, Liu H and Gao S 2004b Classification of single-trial
1999 The thought translation device: a neurophysiological electroencephalogram during finger movement IEEE Trans.
approach to communication in total motor paralysis Exp. Brain Biomed. Eng. 51 1019–25
Res. 124 223–2 Lisogurski D and Birch G E 1998 Identification of finger flexions
Kubler A, Kotchoubey B, Kaiser J, Wolpaw J R and Birbaumer N from continuous EEG as a brain–computer interface Proc. 20th
2001a Brain–computer communication: unlocking the locked Annual Int. Conf. of the IEEE Engineering in Medicine and
Psychol. Bull. 127 358–75 Biology Society (Hong Kong) pp 2004–7
Kubler A, Kotchoubey B, Salzmann H P, Ghanayim N, Liu H S, Gao X, Yang F and Gao S 2003 Imagined hand movement
Perelmouter J, Homberg V and Birbaumer N 1998 identification based on spatio-temporal pattern recognition of
Self-regulation of slow cortical potentials in completely EEG Proc. IEEE-EMBS Conf. on Neural Engineering
paralyzed human patients Neurosci. Lett. 252 171–4 (Capri Island, Italy) pp 599–602
Kubler A, Neumann N, Kaiser J, Kotchoubey B, Hinterberger T and Liu H, Wang J and Zheng C 2005 Mental tasks classification and
Birbaumer N P 2001b Brain–computer communication: their EEG structures analysis by using the growing hierarchical
Self-regulation of slow cortical potentials for verbal self-organizing map Proc. 1st Int. Conf. on Neural Interface
communication Arch. Phys. Med. Rehabil. 82 1533–9 and Control (Wuhan, China) pp 115–8
Kubler A, Nijboer F, Mellinger J, Vaughan T M, Pawelzik H, Mahmoudi B and Erfanian A 2002 Single-channel EEG-based
Schalk G, McFarland D J, Birbaumer N and Wolpaw J R 2005 prosthetic hand grasp control for amputee subjects Proc. 24th
Patients with ALS can use sensorimotor rhythms to operate a Annual Int. Conf. of the IEEE Engineering in Medicine and
brain–computer interface Neurology 64 1775–7 Biology Society (Houston, TX) pp 2406–7
LaCourse J R and Wilson E 2003 Brainiac: a brain computer link Mason S G and Birch G E 2000 A brain-controlled switch for
Proc. IEEE Instrumentation and Measurement Technology asynchronous control applications IEEE Trans. Biomed. Eng.
Conf (Vail, CO) pp 587–92 47 1297–307
Lal T N, Schroder M, Hinterberger T, Weston J, Bogdan M, Mason S G and Birch G E 2003 A general framework for
Birbaumer N and Scholkopf B 2004 Support vector channel brain–computer interface design IEEE Trans. Neural Syst.
selection in BCI IEEE Trans. Biomed. Eng. 51 1003–10 Rehabil. Eng. 11 70–85
Lalor E, Kelly S, Finucane C, Burke R, Smith R, Reilly R and Mason S G, Bohringer R, Borisoff J F and Birch G E 2004
McDarby G 2005 Steady-state VEP-based brain–computer Real-time control of a video game with a direct
interface control in an immersive 3-D gaming environment brain–computer interface J. Clin. Neurophysiol.
EURASIP J. Appl. Signal Process. 19 3156–64 21 404–8
Laubach M, Wessberg J and Nicolelis M A 2000 Cortical ensemble McFarland D J, Anderson C W, Muller K R, Schlogl A and
activity increasingly predicts behaviour outcomes during Krusienski D J 2006 BCI meeting 2005—workshop on BCI
learning of a motor task Nature 405 567–71 signal processing: feature extraction and translation IEEE
Lebedev M A, Carmena J M, O’Doherty J E, Zacksenhouse M, Trans. Neural Syst. Rehabil. Eng. 14 135–8
Henriquez C S, Principe J C and Nicolelis M A 2005 Cortical McFarland D J, McCane L M, David S V and Wolpaw J R 1997
ensemble adaptation to represent velocity of an artificial Spatial filter selection for EEG-based communication
actuator controlled by a brain-machine interface J. Neurosci. Electroencephalogr. Clin. Neurophysiol. 103 386–94
25 4681–93 McFarland D J, Sarnacki W A, Vaughan T M and Wolpaw J R
Lee H and Choi S 2002 PCA-based linear dynamical systems for 2005 Brain–computer interface (BCI) operation: signal and
multichannel EEG classification Proc. 9th Int. Conf. on Neural noise during early training sessions Clin. Neurophysiol.
Information Processing (Singapore) pp 745–9 116 56–62
Lee H and Choi S 2003 PCA+HMM+SVM for EEG pattern McFarland D J, Sarnacki W A and Wolpaw J R 2003
classification Proc. IEEE 7th Int. Symp. on Signal Processing Brain–computer interface (BCI) operation: optimizing
and Its Applications (Paris, France) pp 541–4 information transfer rates Biol. Psychol. 63 237–51

R53
Topical Review

McFarland D and Wolpaw J R 1998 EEG-based communication and Muller-Putz G R, Scherer R, Brauneis C and Pfurtscheller G 2005a
control: short-term role of feedback IEEE Trans. Rehabil. Eng. Steady-state visual evoked potential (SSVEP)-based
6 7–11 communication: impact of harmonic frequency components
McMillan G R and Calhoun G L 1995 Direct brain interface J. Neural Eng. 2 123–30
utilizing self-regulation of steady-state visual evoked response Muller-Putz G R, Scherer R, Pfurtscheller G and Rupp R 2005b
(SSVER) Proc. RESNA ‘95 Annual Conf. (Vancouver, BC) EEG-based neuroprosthesis control: a step towards clinical
pp 693–5 practice Neurosci. Lett. 382 169–74
Mensh B D, Werfel J and Seung H S 2004 BCI competition Musallam S, Cornell B D, Greger B, Scherberger H and
2003–data set ia: combining gamma-band power with slow Andersen R A 2004 Cognitive control signals for neural
cortical potentials to improve single-trial classification of prosthetics Science 305 258–62
electroencephalographic signals IEEE Trans. Biomed. Eng. Neumann N, Hinterberger T, Kaiser J, Leins U, Birbaumer N and
51 1052–6 Kubler A 2004 Automatic processing of self-regulation of slow
Middendorf M, McMillan G, Calhoun G and Jones K S 2000 cortical potentials: evidence from brain–computer
Brain–computer interfaces based on the steady-state communication in paralysed patients Electroencephalogr. Clin.
visual-evoked response IEEE Trans. Rehabil. Eng. 8 211–4 Neurophysiol. 115 628–35
Millan J R 2004 On the need for on-line learning in brain–computer Neumann N, Kubler A, Kaiser J, Hinterberger T and Birbaumer N
interfaces Proc. Annual Int. Joint Conf. on Neural Networks 2003 Conscious perception of brain states: Mental strategies
(Budapest, Hungary) for brain–computer communication Neuropsychologia
Millan J R, Renkens F, Mourino J and Gerstner W 2004a 41 1028–36
Noninvasive brain-actuated control of a mobile robot by human Neuper C, Muller G R, Kubler A, Birbaumer N and
EEG IEEE Trans. Biomed. Eng. 51 1026–33 Pfurtscheller G 2003 Clinical application of an EEG-based
Millan J R, Renkens F, Mourino J and Gerstner W 2003a brain–computer interface: a case study in a patient with
Non-invasive brain-actuated control of a mobile robot Proc. severe motor impairment Clin. Neurophysiol.
18th Int. Joint Conf. on Artificial Intelligence (Acapulco, 114 399–409
Mexico) pp 1121–6 Neuper C, Scherer R, Reiner M and Pfurtscheller G 2005 Imagery
Millan J R, Renkens F, Mouriño J and Gerstner W 2004b of motor actions: differential effects of kinesthetic and
Brain-actuated interaction Artif. Intell. 159 241–59 visual-motor mode of imagery in single-trial EEG Brain Res.
Millan J, Franze M, Mourino J, Cincotti F and Babiloni F 2002a Cogn. Brain Res. 25 668–77
Relevant EEG features for the classification of spontaneous Neuper C, Schloegl A and Pfurtscheller G 1999 Enhancement of
motor-related tasks Biol. Cybern. 86 89–95 left-right sensorimotor EEG differences during feedback-
Millan J, Mourino J, Franze M, Cincotti F, Varsta M, Heikkonen J regulated motor imagery J. Clin. Neurophysiol. 16 373–82
and Babiloni F 2002b A local neural classifier for the Niedermeyer E and Lopes da Silva F 1998 Electroencephalography:
recognition of EEG patterns associated to mental tasks IEEE Basic Principles, Clinical Applications, and Related Fields
Trans. Neural Netw. 13 678–86 4th edn, ed C Mitchell (Baltimore, MD: Williams & Wilkins)
Millan J R, Mourino J, Babiloni F, Cincotti F, Varsta M and Obeid I and Wolf P D 2004 Evaluation of spike-detection algorithms
Heikkonen J 2000a Local neural classifier for EEG-based for a brain-machine interface application IEEE Trans. Biomed.
recognition of mental tasks Proc. IEEE-INNS-ENNS Int. Joint Eng. 51 905–11
Conf. on Neural Networks (Como, Italy) pp 632–6 Obermaier B, Guger C, Neuper C and Pfurtscheller G 2001a Hidden
Millan J R, Mourino J, Cincotti F, Varsta M, Heikkonen J, Topani F, markov models for online classification of single trial EEG
Marciani M G, Kaski K and Babiloni F 2000b Neural networks data Pattern Recognit. Lett. 22 1299–309
for robust classification of mental tasks Proc. 22nd Annual Int. Obermaier B, Muller G R and Pfurtscheller G 2001b ‘Virtual
Conf. of the IEEE Engineering in Medicine and Biology keyboard’ controlled by spontaneous EEG activity IEEE Trans.
Society (Chicago, IL) pp 1380–2 Neural Syst. Rehabil. Eng. 11 636–41
Millan J R, Mourino J, Marciani M G, Babiloni F, Topani F, Obermaier B, Munteanu C, Rosa A and Pfurtscheller G 2001c
Canale I, Heikkonen J and Kaski K 1998 Adaptive brain Asymmetric hemisphere modeling in an offline brain–computer
interfaces for physically-disabled people Proc. 20th Annual Int. interface IEEE Trans. Syst. Man Cybern. 31 536–40
Conf. of the IEEE Engineering in Medicine and Biology Obermaier B, Neuper C, Guger C and Pfurtscheller G 2001d
Society (Hong Kong) pp 2008–11 Information transfer rate in a five-classes brain–computer
Millan J R and Mourino J 2003b Asynchronous BCI and local interface IEEE Trans. Neural Syst. Rehabil. Eng. 9 283–8
neural classifiers: an overview of the adaptive brain interface Olson B P, Si J, Hu J and He J 2005 Closed-loop cortical control of
project IEEE Trans. Neural Syst. Rehabil. Eng. 11 159–61 direction using support vector machines IEEE Trans. Neural
Molina G G, Ebrahimi T and Vesin J M 2003 Joint time-frequency- Syst. Rehabil. Eng. 13 72–80
space classification of EEG in a brain–computer interface Otto K J, Vetter R J, Marzullo T C and Kipke D R 2003
application EURASIP J. Appl. Signal Process. 7 713–29 Brain-machine interfaces in rat motor cortex: implications of
Mourino J, Chiappa S, Jane R and Millan J R 2002 Evolution of the adaptive decoding algorithms Proc. IEEE-EMBS Conf. on
mental states operating a brain–computer interface Proc. of the Neural Engineering (Capri Island, Italy) pp 100–3
Int. Federation for Medical and Biological Engineering Palaniappan R 2005 Brain–computer interface design using band
(Vienna, Austria) pp 600–1 powers extracted during mental tasks Proc. IEEE-EMBS Conf.
Muller K R, Anderson C W and Birch G E 2003a Linear and on Neural Engineering (Arlington, VA) pp 321–4
nonlinear methods for brain–computer interfaces IEEE Trans. Palaniappan R, Paramesran R, Nishida S and Saiwaki N 2002 A new
Neural Syst. Rehabil. Eng. 11 165–9 brain–computer interface design using fuzzy ARTMAP IEEE
Muller K R, Curio G, Blankertz B and Dornhege G 2003b Trans. Neural Syst. Rehabil. Eng. 10 140–8
Combining features for BCI Advances in Neural Information Panuccio A, Bicego M and Murino V 2002 A hidden markov
Processing Systems (NIPS02) vol 15, ed S Becker, S Thrun and model-based approach to sequential data clustering Proc. Joint
K Obermayer pp 1115–22 IAPR Int. Workshop on Structural, Syntactic, and Statistical
Muller G R, Neuper C and Pfurtscheller G 2003c Implementation of Pattern Recognition (Windsor, Canada) pp 734–42
a telemonitoring system for the control of an EEG-based Parra L, Alvino C, Tang A, Pearlmutter B, Yeung N, Osman A and
brain–computer interface IEEE Trans. Neural Syst. Rehabil. Sajda P 2002 Linear spatial integration for single-trial
Eng. 11 54–9 detection in encephalography Neuroimage 17 223–30

R54
 Topical Review

Parra L, Alvino C, Tang A, Pearlmutter B, Yeung N, Osman A and Pineda J A, Silverman D S, Vankov A and Hestenes J 2003 Learning
Sajda P 2003a Single-trial detection in EEG and MEG: keeping to control brain rhythms: making a brain–computer interface
it linear Neurocomputing 52–54 177–83 possible IEEE Trans. Neural Syst. Rehabil. Eng. 11 181–4
Parra L C, Spence C D, Gerson A D and Sajda P 2003b Response Platt J 1991 A resource-allocating network for function
error correction—a demonstration of improved interpolation Neural Comput. 3 213–25
human-machine performance using real-time EEG monitoring Polak M and Kostov A 1997 Development of brain–computer
IEEE Trans. Neural Syst. Rehabil. Eng. 11 173–7 interface: preliminary results Proc. 19th Annual Int. Conf. of
Patil P G, Carmena J M, Nicolelis M A and Turner D A 2004 the IEEE Engineering in Medicine and Biology Society
Ensemble recordings of human subcortical neurons as a source (Chicago, IL) pp 1543–6
of motor control signals for a brain-machine interface Polak M and Kostov A 1998 Feature extraction in development of
Neurosurgery 55 27–35 brain–computer interface: a case study Proc. 20th Annual Int.
Penny W D, Roberts S J, Curran E A and Stokes M J 2000 Conf. of the IEEE Engineering in Medicine and Biology
EEG-based communication: a pattern recognition approach Society (Hong Kong) pp 2058–61
IEEE Trans. Rehabil. Eng. 8 214–5 Polak M and Kostov A 1999 Training setup for control of neural
Peters B O, Pfurtscheller G and Flyvbjerg H 2001 Automatic prosthesis using brain–computer interface Proc. 21st Annual
differentiation of multichannel EEG signals IEEE Trans. Int. Conf. of the IEEE Engineering in Medicine and Biology
Biomed. Eng. 48 111–6 Society & Annual Fall Meeting of the Biomedical Engineering
Peterson D A, Knight J N, Kirby M J, Anderson C W and Society (Atlanta, GA) p 446
Thaut M H 2005 Feature selection and blind source separation Polikoff J B, Bunnell H T and Borkowski W 1995 Toward a
in an EEG-based brain–computer interface EURASIP J. Appl. P300-based computer interface Proc. RESNA’95 Annual Conf.
Signal Process. 19 3128–40 (Vancouver, Canada) pp 178–80
Pfurtscheller G and Aranibar A 1977 Event-related cortical Pregenzer M and Pfurtscheller G 1995 Distinction sensitive learning
desynchronization detected by power measurements of scalp vector quantization (DSLVQ)—application as a classifier based
EEG Electroencephalogr. Clin. Neurophysiol. 42 817–26 feature selection method for a brain–computer interface Proc.
Pfurtscheller G et al 2003b Graz-BCI: state of the art and clinical 4th Int. Conf. on Artificial Neural Networks (Cambridge, UK)
applications IEEE Trans. Neural Syst. Rehabil. Eng. 11 177–80 pp 433–6
Pfurtscheller G, Flotzinger D and Kalcher J 1993 Brain–computer Pregenzer M and Pfurtscheller G 1999 Frequency component
interface—a new communication device for handicapped selection for an EEG-based brain to computer interface IEEE
persons J. Microcomput. Appl. 16 293–9 Trans. Rehabil. Eng. 7 413–9
Pfurtscheller G, Flotzinger D and Neuper C 1994 Differentiation Qin L, Deng J, Ding L and He B 2004a Motor imagery classification
between finger, toe and tongue movement in man based on by means of source analysis methods Proc. 26th Annual Int.
40-Hz EEG Electroencephalogr. Clin. Neurophysiol. Conf. of the IEEE Engineering in Medicine and Biology
90 456–60 Society (San Francisco, CA) pp 4356–8
Pfurtscheller G and Guger C 1999 Brain–computer communication Qin L, Ding L and He B 2004b Motor imagery classification by
system: EEG-based control of hand orthesis in a tetraplegic means of source analysis for brain–computer interface
patient Acta Chir. Austriaca 31 23–5 applications J. Neural Eng. 1 135–41
Pfurtscheller G, Kalcher J, Neuper C, Flotzinger D and Pregenzer M Qin L and He B 2005 A wavelet-based time-frequency analysis
1996 On-line EEG classification during externally-paced hand approach for classification of motor imagery for
movements using a neural network-based classifier brain–computer interface applications J. Neural Eng.
Electroencephalogr. Clin. Neurophysiol. 99 416–25 2 65–72
Pfurtscheller G and Lopes da Silva F H 1999 Event-related Qin L, Kamousi B, Liu Z M, Ding L and He B 2005 Classification
EEG/MEG synchronization and desynchronization: basic of motor imagery tasks by means of time-frequency-spatial
principles Clin. Neurophysiol. 110 1842–57 analysis for brain–computer interface applications Proc.
Pfurtscheller G, Muller G R, Pfurtscheller J, Gerner H J and Rupp R IEEE-EMBS Conf. on Neural Engineering (Arlington, VA)
2003a ‘Thought’—control of functional electrical stimulation pp 374–6
to restore hand grasp in a patient with tetraplegia Neurosci. Lett. Ramachandran V S and Histein W 1998 The perception of phantom
351 33–6 limbs Brain 121 1603–30
Pfurtscheller G, Muller-Putz G R, Pfurtscheller J and Rupp R 2005 Ramoser H, Muller-Gerking J and Pfurtscheller G 2000 Optimal
EEG-based asynchronous BCI controls functional electrical spatial filtering of single trial EEG during imagined hand
stimulation in a tetraplegic patient EURASIP J. Appl. Signal movement IEEE Trans. Rehabil. Eng. 8 441–6
Process. 19 3152–5 Rao Y N, Kim S, Sanchez J C, Erdogmus D, Principe J C,
Pfurtscheller G and Neuper C 2001 Motor imagery and direct Carmena J M, Lebedev M A and Nicolelis M A 2005
brain–computer communication Proc. IEEE 89 1123–34 Learning mappings in brain machine interfaces with echo
Pfurtscheller G, Neuper C, Flotzinger D and Pregenzer M 1997 state networks Proc. IEEE Int. Conf. on Acoustics, Speech
EEG-based discrimination between imagination of right and and Signal Processing (Philadelphia, PA)
left hand movement Electroencephalogr. Clin. Neurophysiol. pp v/233–v/236
103 642–51 Reger B D, Fleming K M, Sanguineti V, Alford S and
Pfurtscheller G, Neuper C, Guger C, Harkam W, Ramoser H, Mussa-Ivaldi F A 2000a Connecting brains to robots: the
Schlogl A, Obermaier B and Pregenzer M 2000 Current trends development of a hybrid system for the study of learning in
in graz brain–computer interface (BCI) research IEEE Trans. neural tissues Proc. 7th Int. Conf. on Artificial Life
Rehabil. Eng. 8 216–9 (Portland, OR) pp 263–72
Pfurtscheller G, Neuper C, Schlogl A and Lugger K 1998 Reger B D, Fleming K M, Sanguineti V, Alford S and
Separability of EEG signals recorded during right and left Mussa-Ivaldi F A 2000b Connecting brains to robots: an
motor imagery using adaptive autoregressive parameters IEEE artificial body for studying the computational properties of
Trans. Rehabil. Eng. 6 316–25 neural tissues Artif. Life 6 307–24
Pineda J A, Allison B Z and Vankov A 2000 The effects of Rezek I, Roberts S and Sykacek P 2003 Ensemble coupled hidden
self-movement, observation, and imagination on mu rhythms markov models for joint characterization of dynamic signals
and readiness potentials (RP’s): toward a brain–computer Proc. 9th Int. Workshop on Artificial Intelligence and Statistics
interface (BCI) IEEE Trans. Rehabil. Eng. 8 219–22 (Key West, FL) ed C M Bishop and B J Frey

R55
Topical Review

Roberts S J and Penny W 2003 Real-time brain–computer Serruya M D, Hatsopoulos N G, Paninski L, Fellows M R and
interfacing: a preliminary study using bayesian learning Med. Donoghue J P 2002 Instant neural control of a movement
Biol. Eng. Comput. 38 56–61 signal Nature 416 141–2
Roberts S J, Penny W and Rezek I 1999 Temporal and spatial Sheikh H, McFarland D J, Sarnacki W A and Wolpaw J R 2003
complexity measures for electroencephalogram based Electroencephalographic (EEG)-based communication: EEG
brain–computer interfacing Med. Biol. Eng. Comput. control versus system performance in humans Neurosci. Lett.
37 93–8 345 89–92
Roushe P, Clement R, Visser C, Cruz J and Si J 2003 ‘Neurabotics’: Stone M 1977 Asymptotics for and against cross-validation
an integration of neurons, rats and robots for advanced studies Biometrika 64 29–35
of brain–computer interface systems Proc. IEEE-EMBS Conf. Su Ryu C, Song Y, Yoo D S, Choi S, Moon S S and Sohn J H 1999
on Neural Engineering (Capri Island, Italy) pp 462–5 EEG-based discrimination between yes and no Proc. 21st
Sanchez J C, Carmena J M, Lebedev M A, Nicolelis M A, Annual Int. Conf. of the IEEE Engineering in Medicine and
Harris J G and Principe J C 2004 Ascertaining the importance Biology Society & Annual Fall Meeting of the Biomedical
of neurons to develop better brain-machine interfaces IEEE Engineering Society (Atlanta, GA) p 444
Trans. Biomed. Eng. 51 943–53 Sutter E E 1992 The brain response interface: communication
Sanchez J C, Erdogmus D, Rao Y N, Principe J and Nicolelis M A through visually-induced electrical brain responses
2002a A comparison between nonlinear mappings and linear J. Microcomput. Appl. 15 31–45
state estimation to model the relation from motor cortical Sykacek P, Roberts S J and Stokes M 2004 Adaptive BCI based on
neuronal firing to hand movements Proc. SAB Workshop on variational bayesian kalman filtering: an empirical evaluation
Motor Control in Humans and Robots (Edinburgh, Scotland) IEEE Trans. Biomed. Eng. 51 719–27
pp 59–65 Sykacek P, Roberts S, Stokes M, Curran E, Gibbs M and Pickup L
Sanchez J C, Erdogmus D, Rao Y N, Principe J, Nicolelis M A and 2003 Probabilistic methods in BCI research IEEE Trans.
Wessberg J 2003 Learning the contributions of the motor, Neural Syst. Rehabil. Eng. 11 192–5
premotor, and posterior parietal cortices for hand trajectory Tavakolian K, Nasrabadi A M and Rezaei S 2004 Selecting better
reconstruction in a brain machine interface Proc. IEEE-EMBS EEG channels for classification of mental tasks Proc. IEEE Int.
Conf. on Neural Engineering (Capri Island, Italy) pp 59–62 Symp. on Circuits and Systems 537–40
Sanchez J C, Kim S P, Erdogmus D, Rao Y N, Principe J, Wessing J Taylor D M, Tillery S I and Schwartz A B 2002 Direct cortical
and Nicolelis M A 2002b Input–output mapping performance control of 3D neuroprosthetic devices Science
of linear and nonlinear models for estimating hand trajectories 296 1829–32
from cortical neuronal firing patterns Proc. IEEE Workshop on Taylor D M, Tillery S I H and Schwartz A B 2003 Information
Neural Networks and Signal Processing (Martigny, conveyed through brain-control: cursor versus robot IEEE
Switzerland) pp 139–48 Trans. Neural Syst. Rehabil. Eng. 11 195–9
Schalk G, Wolpaw J R, McFarland D J and Pfurtscheller G 2000 Thulasidas M, Guan C, Ranganatha S, Wu J K, Zhu X and Xu W
EEG-based communication: presence of an error potential 2004 Effect of ocular artifact removal in brain–computer
Clin. Neurophysiol. 111 2138–44 interface accuracy Proc. 26th Annual Int. Conf. of the IEEE
Scherer R, Muller G R, Neuper C, Graimann B and Pfurtscheller G Engineering in Medicine and Biology Society (San Francisco,
2004 An asynchronously controlled EEG-based virtual CA) pp 4385–8
keyboard: improvement of the spelling rate IEEE Trans. Townsend G, Graimann B and Pfurtscheller G 2004 Continuous
Biomed. Eng. 51 979–84 EEG classification during motor imagery—simulation of an
Schloegl A, Lugger K and Pfurtscheller G 1997a Using adaptive asynchronous BCI IEEE Trans. Neural Syst. Rehabil. Eng.
autoregressive parameters for a brain–computer-interface 12 258–65
experiment Proc. 19th Annual Int. Conf. of the IEEE Trejo L J, Wheeler K R, Jorgensen C C, Rosipal R, Clanton S T,
Engineering in Medicine and Biology Society (Chicago, IL) Matthews B, Hibbs A D, Matthews R and Krupka M 2003
pp 1533–5 Multimodal neuroelectric interface development IEEE Trans.
Schloegl A, Neuper C and Pfurtscheller G 1997b Subject specific Neural Syst. Rehabil. Eng. 11 199–204
EEG patterns during motor imaginary Proc. 19th Annual Int. Varsta M, Heikkonen J, Millan J D R and Mourino J 2000 Evaluating
Conf. of the IEEE Engineering in Medicine and Biology the performance of three feature sets for brain–computer
Society (Chicago, IL) pp 1530–2 interfaces with an early stopping MLP commitee Int. Conf. on
Schlogl A, Keinrath C, Scherer R and Pfurtscheller G 2003 Pattern Recognition (Barcelona, Spain) pp 907–10
Information transfer of an EEG-based brain–computer Vetter R J, Otto K J, Marzullo T C and Kipke D R 2003 Brain
interface Proc. IEEE-EMBS Conf. on Neural Engineering machine interfaces in rat motor cortex: neuronal operant
(Capri Island, Italy) pp 641–4 conditioning to perform a sensory detection task Proc.
Schroder M, Bogdan M, Rosenstiel W, Hinterberger T and IEEE-EMBS Conf. on Neural Engineering (Capri Island, Italy)
Birbaumer N 2003 Automated EEG feature selection for pp 637–40
brain–computer interfaces Proc. IEEE-EMBS Conf. on Neural Vidal J J 1977 Real-time detection of brain events in EEG Proc.
Engineering (Capri Island, Italy) pp 626–9 IEEE 65 633–41
Schröder M, Lal T N, Hinterberger T, Bogdan M, Hill N J, Wang T, Deng J and He B 2004a Classifying EEG-based motor
Birbaumer N, Rosenstiel W and Schölkopf B 2005 Robust imagery tasks by means of time-frequency synthesized spatial
EEG channel selection across subjects for brain–computer patterns Clin. Neurophysiol. 115 2744–53
interfaces EURASIP J. Appl. Signal Process. 19 3103–12 Wang T, Deng J and He B 2004b Classification of motor imagery
Schwartz A B, Taylor D M and Tillery S I 2001 Extraction EEG patterns and their topographic representation Proc. 26th
algorithms for cortical control of arm prosthetics Curr. Opin. Annual Int. Conf. of the IEEE Engineering in Medicine and
Neurobiol. 11 701–7 Biology Society (San Francisco, CA) pp 4359–62
Serby H, Yom-Tov E and Inbar G F 2005 An improved P300-based Wang D, Kochiyama T, Lu S and Wu J 2005a Measurement and
brain–computer interface IEEE Trans. Neural Syst. Rehabil. analysis of electroencephalogram (EEG) using directional
Eng. 13 89–98 visual stimuli for brain–computer interface Proc. Int. Conf. on
Serruya M, Hatsopoulos N, Fellows M, Paninski L and Donoghue J Active Media Technology (Takamatsu, Japan) pp 34–9
2003 Robustness of neuroprosthetic decoding algorithms Wang Y, Wang R, Gao X and Gao S 2005b Brain–computer
Biol. Cybern. 88 219–28 interface based on the high-frequency steady-state visual

R56
 Topical Review

evoked potential Proc. 1st Int. Conf. on Neural Interface and development program IEEE Trans. Neural Syst. Rehabil. Eng.
Control (Wuhan, China) pp 37–9 11 204–7
Wang Y, Zhang Z, Gao X and Gao S 2004c Lead selection for Wu W, Black M J, Gao Y, Bienenstock E, Serruya M D and
SSVEP-based brain–computer interface Proc. 26th Annual Int. Donoghue J P 2002a Inferring hand motion from multi-cell
Conf. of the IEEE Engineering in Medicine and Biology recordings in motor cortex using a kalman filter Proc. SAB
Society (San Francisco, CA) pp 4507–10 Workshop on Motor Control in Humans and Robots
Wang Y, Zhang Z, Li Y, Gao X, Gao S and Yang F 2004d BCI (Edinburgh, Scotland) pp 66–73
competition 2003–data set IV: an algorithm based on CSSD Wu W, Black M J, Gao Y, Bienenstock E, Shaikhounix A and
and FDA for classifying single-trial EEG IEEE Trans. Biomed. Donoghue J P 2002b Neural decoding of cursor motion using a
Eng. 51 1081–6 kalman filter Adv. Neural Inform. Process. Syst. 15 133–40
Weiskopf N, Mathiak K, Bock S W, Scharnowski F, Veit R, Wu R C, Liang S F, Lin C T and Hsu C F 2004 Applications of
Grodd W, Goebel R and Birbaumer N 2004 Principles of a event-related-potential-based brain–computer interface to
brain–computer interface (BCI) based on real-time functional intelligent transportation systems Proc. IEEE Int. Conf. on
magnetic resonance imaging (fMRI) IEEE Trans. Biomed. Eng. Networking, Sensing and Control (Taipei, Taiwan) pp 813–8
51 966–70 Xu N, Gao X, Hong B, Miao X, Gao S and Yang F 2004a BCI
Weiskopf N, Veit R, Erb M, Mathiak K, Grodd W, Goebel R and competition 2003–data set IIb: enhancing P300 wave detection
Birbaumer N 2003 Physiological self-regulation of regional using ICA-based subspace projections for BCI applications
brain activity using real-time functional magnetic resonance IEEE Trans. Biomed. Eng. 51 1067–72
imaging (fMRI): methodology and exemplary data Neuroimage Xu W, Guan C, Siong C E, Ranganatha S, Thulasidas M and Wu J
19 577–86 2004b High accuracy classification of EEG signal Proc. 17th
Weitkunat R 1991 Digital Biosignal Processing (New York: Int. Conf. on Pattern Recognition (Cambridge, UK) pp 391–4
Elsevier) Yan W, Sutherland M T, Sanfratello L L and Tang A C 2004
Wessberg J, Stambaugh C R, Kralik J D, Beck P D, Laubach M, Single-trial classification of ERPS using second-order blind
Chapin J K, Kim J, Biggs S J, Srinivasan M A and identification (SOBI) Proc. IEEE Int. Conf. on Machine
Nicolelis M A 2000 Real-time prediction of hand trajectory by Learning and Cybernetics (Shanghai, China) pp 4246–51
ensembles of cortical neurons in primates Nature 408 361–5 Yom-Tov E and Inbar G F 2001 Selection of relevant features for
Wolpaw J R, Birbaumer N, McFarland D J, Pfurtscheller G and classification of movements from single movement-related
Vaughan T M 2002 Brain–computer interfaces for potentials using a genetic algorithm Proc. 23rd Annual Int.
communication and control Clin. Neurophysiol. 113 767–91 Conf. of the IEEE Engineering in Medicine and Biology
Wolpaw J R, Flotzinger D, Pfurtscheller G and McFarland D J 1997 Society (Istanbul, Turkey) pp 1364–6
Timing of EEG-based cursor control J. Clin. Neurophysiol. Yom-Tov E and Inbar G F 2002 Feature selection for the
14 529–38 classification of movements from single movement-related
Wolpaw J R and McFarland D J 2004 Control of a two-dimensional potentials IEEE Trans. Neural Syst. Rehabil. Eng. 10 170–7
movement signal by a noninvasive brain–computer interface in Yom-Tov E and Inbar G F 2003 Detection of movement-related
humans Proc. Natl Acad. Sci. USA 101 17849–54 potentials from the electro-encephalogram for possible use in a
Wolpaw J R and McFarland D J 1994 Multichannel EEG-based brain–computer interface Med. Biol. Eng. Comput. 41 85–93
brain–computer communication Electroencephalogr. Clin. Yoo S S, Fairneny T, Chen N K, Choo S E, Panych L P, Park H,
Neurophysiol. 90 444–9 Lee S Y and Jolesz F A 2004 Brain–computer interface using
Wolpaw J R, McFarland D, Neat G W and Forneris C A 1991 An fMRI: spatial navigation by thoughts Neuroreport 15 1591–5
EEG-based brain–computer interface for cursor control Yoon H, Yang K and Shahabi C 2005 Feature subset selection and
Electroencephalogr. Clin. Neurophysiol. 78 252–9 feature ranking for multivariate time series IEEE Trans. Knowl.
Wolpaw J R, McFarland D J and Vaughan T M 2000 Data Eng. 17 1186–98
Brain–computer interface research at the wadsworth center Yu Z, Mason S G and Birch G E 2002 Enhancing the performance
IEEE Trans. Rehabil. Eng. 8 222–6 of the LF-ASD brain–computer interface Proc. 24th Annual
Wolpaw J R, McFarland D J, Vaughan T M and Schalk G 2003 The Int. Conf. of the IEEE Engineering in Medicine and Biology
wadsworth center brain–computer interface (BCI) research and Society (Houston, TX) pp 2443–4

R57

View publication stats