DOI 10.1515/revneuro-2013-0032      Rev. Neurosci.

2013; 24(5): 537–552

Alexis Ortiz-Rosario and Hojjat Adeli*

Brain-computer interface technologies: from signal

to action
Abstract: Here, we present a state-of-the-art review of interface system (NIS; Hsu, 2011, 2012; Luis de Mello et al.,
the research performed on the brain-computer interface 2011; Lopez-Gordo et al., 2012; Manyakov et al., 2012). The
(BCI) technologies with a focus on signal processing purpose of these devices is to create a bridge between
approaches. BCI can be divided into three main compo- an individual’s intent and the outer world through brain
nents: signal acquisition, signal processing, and effector signals (Schwartz et  al., 2006; Birbaumer and Cohen,
device. The signal acquisition component is generally 2007; Donoghue et al., 2007; Hatsopoulos and Donoghue,
divided into two categories: noninvasive and invasive. 2009; Nicolelis and Lebedev, 2009; Scherberger, 2009; del
For noninvasive, this review focuses on electroencepha- Riego et  al., 2011). This technology can help a diversity
logram. For the invasive, the review includes electrocor- of patients with various challenges ranging from multi-
ticography, local field potentials, multiple-unit activity, ple neuromuscular diseases such as amyotrophic lateral
and single-unit action potentials. Signal processing tech- sclerosis (ALS) and cerebral palsy to those with trauma in
niques reviewed are divided into time-frequency methods the brain or the spinal cord (Donoghue, 2008). These com-
such as Fourier transform, autoregressive models, wave- plications impact the lives of approximately two million
lets, and Kalman filter and spatiotemporal techniques people in the US alone and millions more in the rest of the
such as Laplacian filter and common spatial patterns. world (Ficke, 1992; Murray and Lopez, 1996; Gustavsson
Additionally, various signal feature classification algo- et al., 2011; Larkindale et al., 2013).
rithms are discussed such as linear discriminant analysis, Other potential applications may come from amputa-
support vector machines, artificial neural networks, and tion or muscle deformation. These amputations can be
Bayesian classifiers. The article ends with a discussion the result of unattended fractures, war wounds, or patho-
of challenges facing BCI and concluding remarks on the physiologic conditions such as diabetes or tumors (Esque-
future of the technology. nazi, 2004). Individuals with such experiences suffer not
only from the trauma, but their everyday life is affected
Keywords: BCI; brain-computer interface; EEG; electro­ adversely and permanently (Stansbury et al., 2007). These
encephalogram; Fourier transform; Laplacian filter; challenges have created an increasing interest in the
signal processing; wavelet transform. development and improvement of BCI technologies. This
article presents a state-of-the-art review of the representa-
tive research performed on BCI technologies. This review
*Corresponding author: Hojjat Adeli, Departments of Biomedical
Engineering, Biomedical Informatics, Civil and Environmental
is divided into five sections: components of a BCI, signal
Engineering and Geodetic Science, Electrical and Computer acquisition approaches, signal processing techniques,
Engineering, Neurological Surgery, and Neuroscience, The Ohio devices currently used, and concluding remarks.
State University, 470 Hitchcock Hall, 2070 Neil Avenue, Columbus,
OH 43210, USA, e-mail: [email protected]
Alexis Ortiz-Rosario: Department of Biomedical Engineering,

Components of a BCI
The Ohio State University, 470 Hitchcock Hall, 2070 Neil Avenue,
Columbus, OH 43210, USA

A BCI is a multicomponent device that measures brain

signals to uncover the subject’s intentions. The three
Introduction basic components in a BCI are signal acquisition, signal
processing, and effector device (Figure 1). This depiction
Current challenges in rehabilitation have created the is a simplified perspective because each of these compo-
need to develop a viable and effective path of communi- nents includes other mechanisms and operations. The
cation between the brain and the exterior environment. signal acquisition component records the signal created
This technology is often called brain-computer interface by the brain. This acquisition can be either invasive or
(BCI), also known as brain-machine interface or neural noninvasive, difference to be discussed in detail in the

Brought to you by | provisional account

Unauthenticated | 82.95.122.249
Download Date | 5/30/14 10:31 AM
538      A. Ortiz-Rosario and H. Adeli: Brain-computer interface technologies: from signal to action

Signal acquisition Signal processing Effector device

• EEG • Autoregressive • Robotic arms

• ECoG • Wavelets • Wheelchairs
• LFP • Fourier transform • Cursors
• SU • Laplacian filter • Spellers
• Common spatial filters • Others
• Others

ABCDEFGHI
JKLMNOPQ
R S T U V Y X Z_

Figure 1 Basic components of a BCI: a brain signal acquisition device, a signal processing system, and an effector device.
The purpose of BCI technology is to create a path of communication to control a prosthetic device, a spelling setup, or a computer cursor.

following section. The signal processing component ana- single electrodes, or arrays (Liao et al., 2012). There are two
lyzes the signal, with the goal of discovering features and main paradigms in which signals are typically recorded:
markers to be translated into desired actions. A great deal synchronous and asynchronous. In the synchronous
of research has been conducted on the improvement of paradigm (Besio et al., 2011; Han et al., 2011), signals are
the signal processing aspect of BCI. The last component is handled through cues and specific time windows. This
the effector device, which can be a computer cursor (Wu approach is typically used with signals that have time-
et al., 2003), a prosthetic arm (Hochberg et al., 2006, 2012; dependent behavior or occur in very specific instances.
Chao et  al., 2010; Collinger et  al., 2013), a speller (Krus- On the contrary, the asynchronous paradigm acquires the
ienski et  al., 2008), or a wheelchair (Galán et  al., 2008; signal with no specific cues to the subject over a broader
Iturrate et al., 2009; Li et al., 2013). time window. This latter approach appears to be the future
direction of BCI where the user can use the device at will
(Müller-Putz and Pfurtscheller, 2008).

Signal acquisition
Noninvasive signal acquisition (EEG)
The signal acquisition component can be divided into two
categories: noninvasive and invasive. The invasive denom- Noninvasive EEG is the preferred method for a vast major-
ination is used when the device pierces through skin, skull ity of BCI researchers (Birbaumer, 2006a). This preference
or even brain matter (Millán and Carmena, 2010). For non- stems from the ease of use and no-risk noninvasive nature
invasive, this review focuses on electroencephalogram of this approach. Yet, these advantages are often accompa-
(EEG), although other technologies have been explored nied by challenges such as low spatial resolution and dif-
with limited practicality such as functional magnetic reso- ficulty of managing signal-to-noise ratios (Cabrerizo et al.,
nance imaging (fMRI; Wessberg et al., 2000; Martins et al., 2012; Cong et al., 2012). Additionally, other brain activities
2011; Rangaprakash et  al., 2013), near-infrared spectros- might supress or conceal the desired signal, thus affect-
copy (NIRS; Sitaram et al., 2006), and magnetoencepha- ing signal quality. Often, spatial filters are introduced to
lography (MEG; Mellinger et  al., 2007; Ahmadlou et  al., enhance the EEG signals that can improve feature clas-
2013). For the invasive, the review includes electrocorti- sification significantly. In addition, by not crossing the
cography (ECoG), local field potentials (LFP), multiple- skin barrier, researchers must deal with artifacts such as
unit activity (MUA), and single-unit (SU) action potentials electromyographs (EMG; electrical activity from muscles
(Figure 2). These signals are recorded through sensors above/in the eye and/or cranial muscles; Pfurtscheller
or electrodes that range from multiple scalp electrodes, et  al., 2000) and/or electrooculographs (EOG; changes

Brought to you by | provisional account

Unauthenticated | 82.95.122.249
Download Date | 5/30/14 10:31 AM
 A. Ortiz-Rosario and H. Adeli: Brain-computer interface technologies: from signal to action      539

A in electrical potential occurring between the front and

back of the eyeball as the eyes move between two fixed
points; Furdea et al., 2012). Because of these limitations,
EEG-based BCI was initially thought to be capable of only
binary (yes/no) responses, but this was disproven when
two-dimensional (Wolpaw et  al., 2004) and even three-
dimensional controls were experimentally achieved
(McFarland et  al., 2010). The EEG is often acquired
through 32 or 64 channels that measure voltage changes
B on the scalp over time (Figure 3; Martis et al., 2012).
Through the use of EEG, distinct types of signals can
be observed, which relate to specific neuronal or motor
behavior (Pasqualotto et  al., 2012). The most typical are
P300 (Finke et al., 2009), visual evoked potentials (VEP),
sensorimotor rhythms (SMR; Pfurtscheller et  al., 2000),
and slow cortical potentials (SCP). Some artificially gen-
erated signals are shown to demonstrate typical charac-
C Scalp EEG teristics of these signals (Figure 4). VEPs and P300 are
event-related potentials that are tied with outside stimu-
lus and are modulated by the subjects’ attention to their
environment. VEPs are generated from visual sensory
ECoG
stimulation, usually flashing lights, to achieve a neuronal
synchronization with a similar frequency (or harmonic)
to the flash of light presented. P300 are evoked potentials
LFP
that occur when the subject observes an expected event
unfold from a seemingly chaotic environment – similar to
SU finding a puzzle piece lost in a table or the sight of a friend
in a group of strangers (Farwell and Donchin, 1988).
Figure 2 BCI signal’s depth, neuron population, and signal waveform.
(A) Table summarizing the different levels for signal acquisition
Researchers use this low probability event, called oddball
with corresponding locations and spatial resolution. (B) Diagram paradigm, to evoke P300 expecting a consistent appear-
showing electrode locations for different signals. (C) Example ance of the wave after a 300 ms window. Event-related syn-
signals from different locations. Noninvasive scalp EEG is the most chronization (ERS) or amplitude suppression known as
distant of the sensors. Being over the scalp, there is no perfora- event-related desynchronization (ERD) are closely related
tion. This provides a disadvantage of distance, collecting noise
to SMRs. Generally found in the EEG α/β band ( < 30 Hz),
and interference from other sources making the signal less visible.
Invasive sensors cross both the scalp and the skull providing a SMRs can be generated by the imagining or realization of
more detailed and accurate signal. The spatial resolution increases movements. SCPs are synchronized polarizations of syn-
as the distance of the sensor from its focus is reduced. aptic potentials that can be voluntarily regulated by the

Figure 3 Typical 32-electrode (A) and 64-electrode (B) array for EEG signal acquisition.

Brought to you by | provisional account

Unauthenticated | 82.95.122.249
Download Date | 5/30/14 10:31 AM
540      A. Ortiz-Rosario and H. Adeli: Brain-computer interface technologies: from signal to action

A B SMR

Boundary

SCP 2

0 5 10 15 20 25 30 Time

SCP 1
0 50 Time

C D P300

0 75 0 75

Visual
cortex
Light Light Brain
(pulses in Hz) (eyes) 300 ms
source Time

Figure 4 Sample EEG signal types.

(A) SCPs are synchronized synaptic potentials that can be voluntarily regulated by the subject through positive reinforcement. (B) SMRs are
sensorimotor rhythms generated by the imagining of movements and like SCPs can be voluntarily controlled. (C) VEPs are signals acquired
on the visual cortex, which possess similar frequencies to visual stimulus. (D) P300 are evoked potentials that occur when the user is faced
with an unexpected event such as finding a puzzle piece lost in a table or the sight of a friend in a group of strangers and appear approxi-
mately 300 ms after the stimulus.

subject through positive reinforcement. The SCP, like the as the time between targets. Flash patterns are generated
SMR, can be voluntarily controlled through training. based on the total number of flashes per trial per letter.
In BCI, the typical setup for P300 spellers works Steady-state VEPs (SSVEPs) are extensively used in
randomly lighting one letter at a time on a screen. The BCI research due to their high signal-to-noise ratio and
appearance of the P300 wave is recorded when the straightforwardness. These evoked potentials are gener-
expected letter is activated and observed by the subject. ated by flashing lights at specific frequencies that can be
Typically, the letters flash in a random order a number of observed in the neuronal activity of the visual cortex. The
times per trial. Krusienski et  al. (2008) used a stepwise typical setup for SSVEP-BCI is to have subjects observe
linear discriminant analysis (SWLDA) to optimize P300 flashing lights in the surrounding area (monitor or pros-
speller performance. The SWLDA approach, similar to a thetic device) allowing the BCI to correlate the observed
linear regression, seeks to model an observation with light with a desired action (cursor control or robotic arm
linear combinations of features or measurements. They movement). The disadvantage of SSVEPs comes with the
experimented on channel selection, channel reference, need to have gaze control for which patients with differ-
removal of redundant features, and optimal values of ent neuromuscular diseases do not have. Müller-Putz and
features and concluded that only the spatial selection Pfurtscheller (2008) demonstrated the use of an SSVEP-
of channels statistically affected the P300 detection and BCI to control asynchronously an electrical prosthesis in
thus the speller performance. Jin et al. (2012a) evaluated four subjects.
the effect of target-to-target intervals and flash patterns in Recent experiments have tried to uncover a better
a P300 speller. The target-to-target intervals are defined understanding of SMR generation and its relation to motor

Brought to you by | provisional account

Unauthenticated | 82.95.122.249
Download Date | 5/30/14 10:31 AM
 A. Ortiz-Rosario and H. Adeli: Brain-computer interface technologies: from signal to action      541

behavior. Researchers have questioned whether SMRs are researchers must acquire data during clinical observation
just an epiphenomenon or perhaps comprise a relevant in the presence of a physician only, which limits greatly
component in the physiologic transmission of information experimental procedures. The ECoG, being closer to the
in motor control (Kilgard et al., 2007). Boulay et al. (2011) brain, provides a higher spatial resolution and signal
studied six able-bodied individuals using EEG-BCI and clarity than noninvasive scalp EEG.
SMR and concluded that SMR signals posses a causal role By providing a clearer signal, ECoG has been used in
on regulation of behavior, thus confirming the usability of the context of BCI for some time. Using ECoG, Chao et al.
SMRs in BCI. Researchers have also looked at the advan- (2010) achieved asynchronous decoding of arm motion in
tages of continuous adaptation in translation accuracy monkeys. Through this decoding, they were able to acquire
in both SMRs and P300. More recently, SMRs have been kinematics with high degrees of freedom. Their decoder
used to provide communication to minimally conscious was used for months without experiencing any effect
patients using single switch BCI (Müller-Putz et al., 2013). in accuracy or signal fidelity. Davis et  al. (2011) recorded
SCPs stand as the most challenging signals to acquire ECoG signals wirelessly to continuously monitor epilep-
from EEG. The training of SCPs can be long (months to tic episodes in canines, with the goal of creating a better
years) and may not be successful (Pasqualotto et  al., platform for communication between the implant and BCI.
2012). Using fMRI, researchers have identified the areas This experiment provided over a year of data in six dogs
where SCPs are generated, which can be used to expe- and a potential platform for a wireless and more stable BCI.
dite BCI training (Hinterberger et al., 2004, 2005). Iversen
et  al. (2008) demonstrated the possibility of training an
ALS patient to control SCPs and that patients with ALS LFP, MUA, and SU
still possess SCPs even in the late stages of the disease.
In contrast to ECoG, LFP, MUA, and SU cross the cortical
surface to record inside the brain matter. LFP reaches to
Invasive signal acquisition (ECoG, LFP, MUA, a small group of neurons recording their field potentials
and SU) (Bansal et  al., 2012), whereas MUA and SU record the
action potentials of single neurons (Maynard et al., 1997;
For a signal to be considered invasive, the recording Carmena et  al., 2003; Collinger et  al., 2013). Research
approach must cross the scalp-skull level. These signals involving LFPs, MUA, and SUs are mostly done in animals
have been explored mostly in animals and only to a limited with few human exceptions (Hochberg et al., 2006, 2012;
extent in human cases (Georgopoulos et al., 1986; Chapin Collinger et al., 2013). These methods provide the highest
et al., 1999; Leuthardt et al., 2004; Krusienski and Shih, spatial resolution yet are the most invasive, typically
2010). For human experimentation, researchers must deal causing damage to tissue surrounding the recording site
with a range of challenges from recording instability to and cells in the area (Navarro et  al., 2005). In addition,
minimizing the risk/benefit ratio. these sensors must be designed to be biocompatible to
withstand the biologically active environment of the brain
to attain reliable and continuous recordings through time
ECoG (Donoghue, 2008).
Using SU recordings in rats, Laubach et  al. (2000)
ECoGs are acquired from the cortical surface of the brain studied the learning of movement and activation of motor
and bear great similarity to scalp EEGs in their character- cortex neurons. They observed neuronal firing rate and
istic shape and are often called intracranial EEG (iEEG). pattern over time. The authors correlated this activity with
ECoGs are often obtained from severely epileptic and Par- motor learning and the activation of new cortical areas,
kinsonian patients (Wu et  al., 2010; Andres et  al., 2011; thus observing a predictable behavior suitable for BCI
Liu et al., 2013) and sometimes from patients with major applications.
depressive disorder (MDD; Ahmadlou et al., 2012a). In the While making SU recordings in monkeys, Carmena
case of epileptic patients, subdural electrode arrays are et al. (2005) observed that one neuron might have different
surgically implanted over multiple cortical areas (fronto, contributions to a single action. A neuron participating in
parietal, and temporal) in foci related to the source of motion generation in one time window may remain silent
the neurologic disorder. This provides researchers with a in another. This observation proved that single-neuron par-
window of 4 to 5 days from the implantation of the elec- ticipation was nonstationary, pointing out the need for con-
trode to surgical resection of the foci. In most occasions, sidering a larger pool of neurons for future BCI applications.

Brought to you by | provisional account

Unauthenticated | 82.95.122.249
Download Date | 5/30/14 10:31 AM
542      A. Ortiz-Rosario and H. Adeli: Brain-computer interface technologies: from signal to action

Chase et al. (2012) used SU recordings in monkeys to Standard BCI setup

A
observe primary motor neuron adaptation capabilities to
Signal acquisition Signal processing
errors in motor tasks. Assuming a linear encoding model, (EEG, ECoG,...)
Device

the authors extracted preferred directions from neurons Hybrid BCI setups
and studied their changes when presented with a pertur-
bation. They argue that error reduction was attributed to a BCI
(Ex. P300)
global re-aiming adaptation from the population instead Signal processing Device
of individual re-tuning of neurons strengthening the idea
of using larger neuronal populations. BCI
(Ex. SSVEP)
Using a 96-electrode array on a monkey, Bansal et al. or
(2012) compared the information encoded within certain B
frequency bands of LFPs and SUs resulting from primary BCI
motor cortex (M1) and ventral premotor area (PMv). They (P300)
Signal processing Device
examined the impact of movement encoded in both LFPs
and SUs within the M1 and PMv and concluded that a large Physio. device
(EMG or EOG)
enough pool of SUs ( > 16 neurons) provides better or equal
information than LFPs, but including LFPs improves Figure 5 Standard BCI setup vs. hybrid BCI setup.
robustness for BCI applications. (A) Example of standard BCI setup including signal acquisition,
More recently, MUAs have been demonstrated in the signal processing, and device. (B) Examples of a hybrid BCI. These
BCI field as a recording method. This approach acquires systems acquire and use more than one type of signals to improve
the operation of the BCI.
multiple neuronal activities that can be isolated in mul-
tiple channels. It was used recently to demonstrate
robotic control in tetraplegic patients (Hochberg et  al.,
2006, 2012). Collinger et al. (2013) demonstrated a seven
Signal processing: feature
degrees-of-freedom robotic arm used by a tetraplegic extraction
patient. Their application allowed the subject to move
the robotic arm freely in a three-dimensional workspace After the signal is acquired by any of the aforementioned
achieving a 91.6% mean success rate. mediums, it has to be processed. This task is performed
by the signal processing component. In a BCI system, this
component consists of two major steps: feature extraction
Hybrid signals and feature translation/classification. The feature extrac-
tion is the process of evaluating the acquired signal and
A recent trend in BCI signal acquisition, called hybridiza- identifying potential signal features or markers (Acharya
tion or hybrid BCI, is the use of more than one signal type et al., 2011, 2012; Hou et al., 2011; Cong et al., 2013; Hsu,
(Figure 5; Nam et al., 2011). These devices can range from 2013). A signal extraction algorithm is used to discover
two different BCIs working together to physiologic devices the features that correlate most strongly with the sub-
such as EMG or EOG being added to the BCI. These addi- ject’s intent. The result from the extraction is sent to a
tional signals can work within the BCI system to turn it on feature classification algorithm that relates the extracted
or change control states for a more accurate and robust feature with the output device used. Effective and accu-
operation. rate feature extraction and classification (Ahmadlou and
Another type of signal used in a hybrid BCI is motion- Adeli, 2010a) are of paramount importance for the success
onset VEPs (M-VEPs), a type of VEP evoked by motion of the BCI. Many signal processing techniques have been
instead of flashing frequency (Torriente et  al., 1999). used in BCI research (McFarland et  al., 2006; Waldert
Jin et al. (2012b) proposed a hybrid BCI using P300 and et al., 2009). The most common ones are reviewed in this
M-VEP. Their experimentation was designed to assess section.
M-VEP performance for BCI, test hybrid feasibility, and Feature extraction is the most important and chal-
find possible alternatives for BCI operation. Another lenging aspect of BCI signal processing because of the
notable aspect of the research is the use of color stimula- complex processes involved in the brain. Feature extrac-
tion. The hybrid approach yielded a mean classification tion algorithms must deal with the source of the signal,
accuracy of 96%, higher than using stand-alone P300 which is often noisy and complex, and detect features of
and M-VEP BCIs. interest. For the discovery of these features, researchers

Brought to you by | provisional account

Unauthenticated | 82.95.122.249
Download Date | 5/30/14 10:31 AM
 A. Ortiz-Rosario and H. Adeli: Brain-computer interface technologies: from signal to action      543

rely on time-frequency and/or temporal-spatial analysis. equations to make the predictions. Most of the information
The most common techniques used in time-frequency extracted from the signal is through the conversion of the
analysis include Fourier transforms (FT), autoregressive time domain into the frequency domain and finding spec-
(AR) models, wavelet transform, and Kalman filters (KF). tral amplitudes in the frequency domain. An optional use
The temporal-spatial techniques typically used are Lapla- for AR model is to directly use the acquired values (or coef-
cian filter and common spatial patterns (CSP). ficients) as features instead of the frequency spectrum.
Wolpaw et  al. (2004) presented an application of
AR model in EEG-BCI using frequency amplitudes from
Time-frequency methods μ (8–12 Hz) and β (18–26 Hz) bands for two-dimensional
control of a computer cursor on human subjects. The
Fourier transforms amplitudes in μ and β bands were modulated through
motor imagery, making the cursor move in the horizon-
FT is one of the oldest and most widely used methods for tal or vertical direction. With this control, the subject was
signal processing and BCI. Discrete FT (DFT) is often used able to move the cursor to one of eight predefined loca-
to convert time series functions into a frequency domain tions on a screen. Other applications of AR model can be
representation using discrete samples of a continuous found (Guger et  al., 2001; Kübler, 2005; Vidaurre et  al.,
time signal. Diez et  al. (2011) used DFT in an EEG-BCI 2006; Mehta et al., 2011).
experiment with six human subjects to decode high-fre-
quency SSVEP to achieve asynchronous control of a com-
puter cursor. Advantages of high frequencies ( > 30 Hz) Wavelet
attainable in SSVEP are lower visual fatigue (due to less
perceived flickering) and less interference with the α band A more recent time-frequency analysis technique is the
(an information-rich frequency band). The authors set a wavelet transform that has been used in many fields from
monitor with four light-emitting diodes (LEDs) above, image and pattern recognition (Ghodrati Amiri et al., 2012;
below, and to each side of the monitor. Each LED flickered Jiang et al., 2012; Lin et al., 2012; Tao et al., 2012; Xiang and
at a slightly different frequency (37, 38, 39, and 40 Hz) rep- Liang, 2012) to automated EEG-based diagnosis of neuro-
resenting the four possible cursor movements (up, down, logic and psychiatric disorders (Adeli and Ghosh-Dastidar,
right, and left). Subjects were asked to control the cursor 2010). This methodology has been applied to epilepsy
through different mazes by gazing at the one of the four (Adeli et al., 2003, 2007; Ghosh-Dastidar and Adeli, 2007,
flickering lights and determine direction. Using a linear 2009; Ghosh-Dastidar et  al., 2008), Alzheimer’s disease
combination of Fourier series, some researchers have tried (Adeli et  al., 2005, 2008; Ahmadlou et  al., 2010a, 2011;
to quantify the relationship between two or more brain Sankari and Adeli, 2011; Sankari et al., 2012), autism spec-
signals usually using measures based on FT of the signal. trum disorder (ASD; Ahmadlou et al., 2010b, 2012b), atten-
Krusienski et  al. (2012) compared three such measures: tion-deficit and hyperactivity disorder (ADHD; Ahmadlou
coherence, spectral power, and the phase-locking value and Adeli, 2010b, 2011; Ahmadlou et al., 2012c), and MDD
(PLV) of two EEG signals obtained from seven able-bodied (Ahmadlou et  al., 2012a). The approaches presented in
users trained on a simple one-dimensional/two-target the aforementioned papers have demonstrated the power
cursor control task. They conclude that ‘spectral power of wavelets as denoisers and feature extractors and their
produced classification at least as good as PLV, coherence, effectiveness in brain signal processing. Wavelet uses the
or any possible combination of these measures.’ This was concept of resonance to acquire features within a specific
expected because all signals were obtained from the same frequency band. The main concept in wavelet analysis is to
area of the brain – sensorimotor cortex. Other applica- decompose a given signal into scale components in both
tions of fast FT (FFT) in BCI are reported in Chin-Teng et al. frequency and time domains. This decomposition allows
(2008). specific frequency bands to be extracted, processed, and/
or analyzed. This signal processing technique is espe-
cially helpful for nonperiodic/nonstationary signals with
AR model discontinuities.
The discrete wavelet packet transform (DWPT) pro-
The AR model attempts to estimate a given data set, or vides more coefficients than the conventional discrete
signal, by linear weighted sum of prior values. It is an wavelet transform (DWT), representing additional subtle
efficient method because it uses only linear regression details of a signal and can be considered a generalization

Brought to you by | provisional account

Unauthenticated | 82.95.122.249
Download Date | 5/30/14 10:31 AM
544      A. Ortiz-Rosario and H. Adeli: Brain-computer interface technologies: from signal to action

of DWT (Jiang and Adeli, 2004). One of the challenges of compared linear multiple regression analysis with KF for
EEG signal processing is the location of the optimal sub- LFP signals obtained from recorded motor cortex neurons
ject-based band, the slight difference found in subjects’ in nonhuman primates. The authors were able to decode
frequency bands due to physiologic differences. Yang endpoint forces from the hand accurately, thus potentially
et  al. (2007) proposed an adaptive subject-based feature improving the BCI operation. Other applications of KF can
extraction using DWPT. The authors determined the best be found (Sykacek et al., 2004; Gage et al., 2005; Koyama
wavelet basis for the most suitable frequency subbands et al., 2010; White et al., 2010).
for signal representation. They tested the idea on three
different motor imagery tasks: playing basketball using
the left hand, playing basketball using the right hand, and Spatiotemporal techniques
braking using the right foot. Hsu et  al. (2012) classified
single-trial left finger lifting and resting EEGs obtained Laplacian filter
from male and female human subjects using Daubechies
wavelets and amplitude modulation. Another application The methods in previous sections were applied on time
of wavelet can be found in Donchin et al. (2000). series independent of their spatial distribution on or in
the brain. The methods covered in this section deal with
multiple signals distributed over the brain (Figure 6).
Kalman filter Laplacian filters are a family of signal processing tech-
niques that observe the change of behavior of spatially
KF, or linear quadratic estimation, is an algorithm used distributed signals. Mostly used for image processing,
to estimate unknown variables from measurements that Laplacian filters can delineate signal changes between
contain noise or inaccuracies. The algorithm works with spatial locations obtained from electrodes over the scalp.
a series of inputs and recursive measurements to achieve As such, they provide a useful tool for spatiotemporal pro-
its statistically optimal estimates. The filter will estimate cessing of BCI signals.
a variable including noise, and once estimated, it will use Qin et  al. (2005) reported a pilot study using source
weights to adjust the estimation in a recursive manner. analysis, a technique for approximation of the source of
The advantage of the KF is its ease of use for real-time an EEG signal, for the classification of motor imagery.
data analysis. Malik et  al. (2011) applied a steady-state The authors used various techniques to pre-process the
KF (SSKF) on ECoG data obtained from two patients to signal including Laplacian filters, time-frequency analy-
assess its computational efficiency against standard KF. sis, band-pass temporal filtering, and independent com-
The authors recorded LFPs using a silicon microelectrode ponents analysis (ICA; Al-Naser and Soderstrom, 2012)
over the precentral gyrus opposite the dominant hand. before performing the source analysis. Faller et al. (2012)
The cursor velocity and position were decoded using applied Laplacian transform to EEG signals obtained from
SSKF and KF. SSKF displayed improved computational three channels for autocalibration of motor imagery BCI,
efficiency over stand-alone KF proving a potential method performed classification using LDA (Garcia-Cuesta et al.,
for online neuroprosthetic control. Gupta and Ashe (2009) 2011), and reported a calibration accuracy of 70%. Other

A Time signal B Spatio-temporal signal

5 Single electrode Multiple electrodes
3
1
-1 0
CZ C4 T4
50 100 150 200 250
-3
Time Time
-5

Space

Signals

Figure 6 Time domain signal vs. spatial domain signal.

(A) Typical time series signal. (B) Typical spatiotemporal signal. A time-series signal is usually taken from a single electrode or recording
site and is evaluated over time. On the contrary, spatiotemporal signals are a group of signals recorded over a grid. Operations applied to
spatiotemporal signals may range from simple averaging of neighboring signals to elaborate transformations.

Brought to you by | provisional account

Unauthenticated | 82.95.122.249
Download Date | 5/30/14 10:31 AM
 A. Ortiz-Rosario and H. Adeli: Brain-computer interface technologies: from signal to action      545

applications of Laplacian filters have been reported in by LDA have been reported by Lemm et al. (2005), Galán et al.
Millán et al. (2004) and Kamousi et al. (2005). (2008), and Finke et al. (2009).
SVM, similar to LDA, uses hyperplanes to distinguish
different classes. The major difference between the two
Common spatial pattern is in the application of an optimal hyperplane that maxi-
mizes the distance between the hyperplane and each
A commonly used statistical approach used as a spatial class using support vectors (boundary points between
filter for feature extraction in BCI is the CSP algorithm, classes to be differentiated). Through the optimal hyper-
a data-driven supervised statistical learning algorithm. plane, SVM achieves a higher generalization for the clas-
It works by maximizing the variance of the spatially fil- sification, thus becoming more robust. SVM classifiers
tered signal of one class and minimizing the variance for can be converted into nonlinear classifiers by converting
the other class in a two-class BCI problem. This is for- the feature space into a higher-dimensional space using
mulated as a generalized eigenvalue problem where the a kernel function, usually Gaussian or radial basis func-
largest and smallest eigenvalues are used to measure the tion (RBF; Junfei and Honggui, 2010; Patrinos et al., 2010;
difference between the two classes (Blankertz et al., 2011). Wu et  al., 2010). Spüler et  al. (2012) used a RBF SVM to
Another application of CSP has been reported by Lemm improve performance for P300 classification. Furdea et al.
et al. (2005). (2012) use RBF SVM for classification in a pseudo lie detec-
tion test along with several other classification methods
including LDA. They report the best classification results
Signal processing: feature classification/ by RBF SVM. Other examples of SVM are presented by
translation Rakotomamonjy and Guigue (2008) and Hsu et al. (2012).
NNs are used extensively in many fields including the
The second component of signal processing is respon- BCI field. They are capable of approximating any continu-
sible for translating the features with desired actions or ous function and solve multiple-class classification prob-
controls. The translation (or classification) algorithms use lems effectively (Freitag et al., 2011; Puscasu and Codres,
statistical or discriminative methods to place the feature 2011; Setiono et  al., 2011; Graf et  al., 2012; Hsiao et  al.,
into classes. Proper algorithms must be selected for each 2012; Osornio-Rios et al., 2012). White et al. (2010) used a
problem, otherwise faulty representation of the observed combination of KF and NN to predict the user’s intent for
features or operational errors in the BCI may result. The prosthetic limb control. Another application of NN in BCI
following approaches are reviewed: LDA, support vector is presented by Yang et al. (2007).
machines (SVM; Jumutc et al., 2011; Wandekokem et al., Bayesian classifiers set the decision boundaries based
2011; Dai et al., 2012), neural networks (NN; Ahmed et al., on probabilities in contrast to LDA, SVM, and NNs, which
2011; Colici et  al., 2011; Rossello et  al., 2012; Tomasevic discriminate deterministically. Jin et  al. (2012a) use a
et  al., 2012; Yamanishi et  al., 2012), and Bayesian clas- Bayesian LDA for the optimization of target-to-target inter-
sifiers (Saliminejad and Gharaibeh, 2012). A review of vals explained in the signal acquisition section above.
EEG-BCI classification algorithms up to 2007 is presented Additional examples are presented by Shin et  al. (2010)
in by Lotte et  al. (2007). This review will discuss more and Malik et al. (2011).
recent applications of the aforementioned techniques.
LDA is a linear method that seeks to minimize the
interclass variance while maximizing the distance
between means of two classes. LDA assumes the data Devices
possess a normal distribution and separates them using a
linear hyperplane. For a multiple-class problem, a gener- The final component of the BCI system is the output device.
alization of LDA is used called multiple discriminant anal- The initial purpose of BCIs was to provide a rehabilitation
ysis (MDA), which applies several hyperplanes to separate tool, but this purpose has now expanded to other applica-
the features. LDA and other linear discriminant methods, tions (Lance et al., 2012). Because earlier BCI applications
however, cannot solve nonlinear classification problems were intended for patients with severe neuromuscu-
effectively, which commonly occur in the context of BCI lar disease, most of the output devices used in research
(Linderman et al., 2008). Iturrate et al. (2009) use SWLDA are in the form of spellers or computer cursors (Kayagil
for classification of a noninvasive P300 BCI for the auto- et  al., 2009; Shih et  al., 2012). Further research intends
mated navigation of a wheelchair. Other applications of to broaden this list by using augmented BCIs, which

Brought to you by | provisional account

Unauthenticated | 82.95.122.249
Download Date | 5/30/14 10:31 AM
546      A. Ortiz-Rosario and H. Adeli: Brain-computer interface technologies: from signal to action

use improved sensors to acquire signals. BCI has been combination of movements to achieve the objective. The
extended to nonrehabilitation purposes such as leisure goal-oriented BCI attempts to mimic the complex pathway
and entertainment. Liao et al. (2012) demonstrated the use of motor control, which the control-oriented BCI bypasses
of wireless EEG-BCI for gaming control. The approach was (Figure 7).
to use attention levels to control an archery video game Another challenge is the ‘BCI illiteracy’ (Hammer et al.,
using the BCI. With three electrodes encased in a conduc- 2012), that is, a percentage of people that are unable to use
tive fabric and foam setup, they were able to assess atten- BCI due to advanced stages of neuromuscular diseases or
tion through FFT of the signal and demonstrated a reliable other unknown reasons. In some cases, the patient is able
mechanism for control. to regain proficiency by switching from BCI approach. Yet,
Galán et  al. (2008) demonstrated asynchronous in some cases, such as in advanced stages of locked-in
control for a wheelchair using EEG-BCI and a multiple LDA syndrome (LIS) or complete LIS (CLIS), the signals need for
for feature extraction and a Gaussian classifier for classi- the BCI become challenging to acquire as the disease pro-
fication. More recently, Huang et al. (2012) used ERS/ERD gresses (Birbaumer, 2006b; Kübler and Birbaumer, 2008).
on both sides of the brain in an EEG-BCI to set different For signal acquisition, invasive sensors currently face
commands to control activation of a wheelchair. a reliability issue. These sensors lose signal quality with
BCI is also being used increasingly to control pros- the passage of time due to the harsh environments inside
thetic or robotic devices. Pfurtscheller et  al. (2003) used the brain. They have not achieved the long-term reliabil-
functional electrical stimulation (FES) in an EEG-BCI to ity required for commercial applications (Schwartz et al.,
control and restore hand grasp in a tetraplegic patient. 2006). For noninvasive sensors, most require applica-
Using foot motor imagery, the patient was able to control tion of gels that make their implementation complicated
grasping of a cylinder with his disabled hand. Moritz et al. and not ready-to-use. However, new sensors are cur-
(2008) also used FES for wrist flexion in monkeys using rently being developed to overcome this problem (Liao
cortical firing rates. Earlier approaches of movement et al., 2012). Another challenge is the long training times
control through BCI are given in Guger et  al. (1999) and (Kreilinger, 2012) and to make single-trial classifications
Lauer et  al. (1999). Gancet (2012) presented the applica- (Blankertz et  al., 2011). Single-trial classification seeks
tion of BCI to a lower limb exoskeleton for patients with to calibrate the device in one trial (or as close). Müller-
spinal cord injury. Putz et al. (2005) reported training of an individual for an
Cincotti et  al. (2008) presented a BCI suite that pro- EEG-BCI using motor imagery in the relatively short time
vided control to multiple devices using a computer screen of 3 days, demonstrating the possibility of quicker train-
with icons each representing a device. It was tested for ing in the near future. These are only a sample of the chal-
the directional control of a toy dog in a maze. This pro- lenges BCI currently faces.
posed suite has applications for control of in-house robots
responding to the user’s wish, such as opening the front
door or calling for a caretaker. Cortex

Basal ganglia Cerebellum

Cortex

Challenges
Thalamus

For BCI to move from the laboratory to real-life applica-

Brain stem
tions, a number of challenges must be overcome. Most BCI BCI
research is done under heavily controlled environment.
Spinal cord
The technology must reach a stand-alone status to be in the
reach of patients and health providers (Fetz, 2007). Two Muscles
possible setup approaches to BCI have emerged referred to
as goal-oriented and control-oriented. The goal-oriented
Action Action
or cognitive BCI seeks to acquire the subject’s objectives or
desires instead of specific control. These systems require Normal motor pathway BCI pathway

a complex output device to translate an objective such as Figure 7 Normal central nervous system operation involves interac-
‘reach a glass of water’ to the action (Andersen et al., 2005, tion of multiple components to produce motor action.
2010). The output device needs to decipher obstacles and Current BCI only records actions of one of the components.

Brought to you by | provisional account

Unauthenticated | 82.95.122.249
Download Date | 5/30/14 10:31 AM
 A. Ortiz-Rosario and H. Adeli: Brain-computer interface technologies: from signal to action      547

Conclusion (Malik et al., 2011). This number is expected to increase in

the coming years.
BCI holds a promising future for a broad range of appli- Brilliant minds from around the world are doing
cations. Current technological advances and applications research in BCI using multidisciplinary concepts and
in the field of BCI are growing fast. BCI, as a technology, approaches. Growing exponentially in the past 15 years,
has achieved a great deal of breakthroughs from control of BCI is a field with a bright future and great potentials for
robotic arms to three-dimensional cursor manipulation. amazing technological breakthroughs.
A limited number of systems are currently in the market
(Debener et  al., 2012) and others under clinical trials Received May 18, 2013; accepted August 21, 2013

References
Acharya, U.R., Sree, S.V., and Suri, J.S. (2011). Automatic detection Ahmadlou, M., Adeli, H., and Adeli, A. (2012b). Improved visibility
of epileptic EEG signals using higher order cumulant features. graph fractality with application for diagnosis of autism
Int. J. Neural Syst. 5, 403–414. spectrum disorder. Physica A 391, 4720–4726.
Acharya, U.R., Sree, S.V., Alvin, A.P., Yanti, R., and Suri, J.S. (2012). Ahmadlou, M., Adeli, H., and Adeli, A. (2012c). Graph theoretical
Application of non-linear and wavelet based features for the analysis of organization of functional brain networks in ADHD.
automated identification of epileptic EEG signals. Int. J. Neural Clin. EEG Neurosci. 43, 5–13.
Syst. 22, 1250002. Ahmadlou, M., Adeli, A., Bajo, R., and Adeli, H. (2013). Complexity of
Adeli, H. and Ghosh-Dastidar, S. (2010). Automated EEG-Based functional connectivity networks in mild cognitive impairment
Diagnosis of Neurological Disorders – Inventing the Future of patients during a working memory task. Clin. Neurophysiol. In
Neurology. (Boca Raton, FL: CRC Press/Taylor & Francis). press.
Adeli, H., Zhou, Z., and Dadmehr, N. (2003). Analysis of EEG records Ahmed, S., Shahjahan, M., and Murase, K. (2011). A Lempel Ziv
in an epileptic patient using wavelet transform. J. Neurosci. complexity-based neural network pruning algorithm. Int. J.
Methods 123, 69–87. Neural Syst. 21, 427–441.
Adeli, H., Ghosh-Dastidar, S., and Dadmehr, N. (2005). Alzheimer’s Al-Naser, M. and Soderstrom, U. (2012). Reconstruction of occluded
disease: models of computation and analysis of EEGs. Clin. EEG facial images using asymmetrical principal component
Neurosci. 36, 131–140. analysis. Integr. Comput. Aided Eng. 19, 273–283.
Adeli, H., Ghosh-Dastidar, S., and Dadmehr, N. (2007). Andersen, R.A., Musallam, S., Burdick, J., and Cham, J.G. (2005).
A wavelet-chaos methodology for analysis of EEGs and EEG Cognitive based neural prosthetics. Proc. IEEE 2, 1908–1913.
sub-bands to detect seizure and epilepsy. IEEE Trans. Biomed. Andersen, R.A., Eun, J.H., and Mulliken, G.H. (2010). Cognitive
Eng. 54, 205–211. neural prosthetics. Annu. Rev. Psychol. 61, 169–190.
Adeli, H., Ghosh-Dastidar, S., and Dadmehr, N. (2008). A spatio- Andres, D.S., Cerquetti, D., and Merello, M. (2011). Finite
temporal wavelet-chaos methodology for EEG-based diagnosis dimensional structure of the GPI discharge in patients
of Alzheimer’s disease. Neurosci. Lett. 444, 190–194. Parkinson’s disease. Int. J. Neural Syst. 21, 175–186.
Ahmadlou, M. and Adeli, H. (2010a). Enhanced probabilistic neural Bansal, A.K., Truccolo, W., Vargas-Irwin, C.E., and Donoghue, J.P.
network with local decision circles: a robust classifier. Integr. (2012). Decoding 3D reach and grasp from hybrid signals in
Comput. Aided Eng. 17, 197–210. motor and premotor cortices: spikes, multiunit activity, and
Ahmadlou, M. and Adeli, H. (2010b). Wavelet-synchronization local field potentials. J. Neurophysiol. 107, 1337–1355.
methodology: a new approach for EEG-based diagnosis of Besio, W.G., Liu, X., Wang, L., Medvedev, A.V., and Koka, K. (2011).
ADHD. Clin. EEG Neurosci. 41, 1–10. Transcutaneous focal electrical stimulation via concentric ring
Ahmadlou, M. and Adeli, H. (2011). Fuzzy synchronization likelihood electrodes reduces synchrony induced by pentylenetetrazole in
with application to attention-deficit/hyperactivity disorder. beta and gamma bands in rats. Int. J. Neural Syst. 21, 139–149.
Clin. EEG Neurosci. 42, 6–13. Birbaumer, N. (2006a). Brain-computer-interface research: coming
Ahmadlou, M., Adeli, H., and Adeli, A. (2010a). New diagnostic EEG of age. Clin. Neurophysiol. 117, 479–482.
markers of the Alzheimer’s disease using visibility graph. Birbaumer, N. (2006b). Breaking the silence: brain-computer
J. Neural Transm. 117, 1099–1109. interfaces (BCI) for communication and motor control.
Ahmadlou, M., Adeli, H., and Adeli, A. (2010b). Fractality and a Psychophysiology 43, 517–532.
wavelet-chaos-neural network methodology for EEG-based Birbaumer, N. and Cohen, L.G. (2007). Brain-computer interfaces:
diagnosis of autistic spectrum disorder. J. Clin. Neurophysiol. communication and restoration of movement in paralysis.
27, 328–333. J. Physiol. 579, 621–636.
Ahmadlou, A., Adeli, H., and Adeli, A. (2011). Fractality and a Blankertz, B., Lemm, S., Treder, M., Haufe, S., and Müller,
wavelet-chaos methodology for EEG-based diagnosis of K.-R. (2011). Single-trial analysis and classification of ERP
Alzheimer’s disease. Alzheimer’s Dis. Assoc. Dis. 25, 85–92. components – a tutorial. NeuroImage 56, 814–825.
Ahmadlou, M., Adeli, H., and Adeli, A. (2012a). Fractality analysis of Boulay, C.B., Sarnacki, W.A., Wolpaw, J.R., and McFarland, D.J.
frontal brain in major depressive disorder. Int. J. Psychol. 85, (2011). Trained modulation of sensorimotor rhythms can affect
206–211. reaction time. Clin. Neurophysiol. 122, 1820–1826.

Brought to you by | provisional account

Unauthenticated | 82.95.122.249
Download Date | 5/30/14 10:31 AM
548      A. Ortiz-Rosario and H. Adeli: Brain-computer interface technologies: from signal to action

Cabrerizo, M., Ayala, M., Goryawala, M., Jayakar, P., and del Riego, R., Otero, J., and Ranilla, J. (2011). A low cost 3D human
Adjouadi, M. (2012). A new parametric feature descriptor interface device using GPU-based optical flow algorithms.
for the classification of epileptic and control EEG records in Integr. Comput. Aided Eng. 18, 391–400.
pediatric population. Int. J. Neural Syst. 22, 1250001. Diez, P.F., Mut, V.A., Avila, P.E.M., and Laciar, L.E. (2011).
Carmena, J.M., Lebedev, M.A., Crist, R.E., O’Doherty, J.E., Asynchronous BCI control using high-frequency SSVEP.
Santucci, D.M., Dimitrov, D.F., Patil, P.G., and Nicolelis, M.A.L. J. Neuroeng. Rehabil. 8, 39.
(2003). Learning to control a brain-machine interface for Donchin, E., Spencer, K.M., and Wijesinghe, R. (2000).
reaching and grasping by primates. PLoS Biol. 1, 193–208. The mental prosthesis: assessing the speed of a
Carmena, J.M., Lebedev, M.A., Henriquez, C.S., and Nicolelis, M.A. P300-based brain-computer interface. IEEE Trans. Rehabil.
(2005). Stable ensemble performance with single-neuron Eng. 8, 174–179.
variability during reaching movements in primates. J. Neurosci. Donoghue, J.P. (2008). Bridging the brain to the world:
25, 10712–10716. a perspective on neural interface systems. Neuron
Chao, Z.C., Nagasaka, Y., and Fujii, N. (2010). Long-term asynchronous 60, 511–521.
decoding of arm motion using electrocorticographic signals in Donoghue, J.P., Nurmikko, A., Black, M., and Hochberg, L.R. (2007).
monkeys. Front. Neuroeng. 3, 3. Assistive technology and robotic control using motor cortex
Chapin, J.K., Moxon, K.A., Markowitz, R.S., and Nicolelis, M.A. ensemble-based neural interface systems in humans with
(1999). Real-time control of a robot arm using simultaneously tetraplegia. J. Physiol. 579, 603–611.
recorded neurons in the motor cortex. Nat. Neurosci. 2, Esquenazi, A. (2004). Amputation rehabilitation and prosthetic
664–670. restoration. From surgery to community reintegration. Disabil.
Chase, S.M., Kass, R.E., and Schwartz, A.B. (2012). Behavioral Rehabil. 26, 831–836.
and neural correlates of visuomotor adaptation observed Faller, J., Solis-Escalante, T., Scherer, R., Neuper, C., and
through a brain-computer interface in primary motor cortex. Vidaurre, C. (2012). Autocalibration and recurrent adaptation:
J. Neurophysiol. 108, 624–644. towards a plug and play online ERD-BCI. IEEE Trans. Neural
Chin-Teng, L., Li-Wei, K., Jin-Chern, C., Jeng-Ren, D., Ruey-Song, H., Syst. Rehabil. Eng. 20, 313–319.
Sheng-Fu, L., and Tzai-Wen, C. (2008). Noninvasive neural Farwell, L.A. and Donchin, E. (1988). Talking off the top of your
prostheses using mobile and wireless EEG. Proc. IEEE 96, head: toward a mental prosthesis utilizing event-related brain
1167–1183. potentials. Electroencephalogr. Clin. Neurol. 70, 510–523.
Cincotti, F., Mattia, D., Aloise, F., Bufalari, S., Marciani, M.G., Fetz, E.E. (2007). Volitional control of neural activity: implications
Babiloni, F., Schalk, G., and Cherubini, A. (2008). Non-invasive for brain-computer interfaces. J. Physiol. 579, 571–579.
brain-computer interface system: towards its application as Ficke, R.C. (1992). Digest of Data on Persons with Disabilities.
assistive technology. Brain Res. Bull. 75, 796–803. (Washington, DC: National Institute on Disability and
Colici, S., Zalay, O.C., and Bardakjian, B.L. (2011). Response Rehabilitation Research).
neuromodulators based on artificial neural networks used Finke, A., Lenhardt, A., and Ritter, H. (2009). The MindGame:
to control seizure-like events in a computational model of a P300-based brain-computer interface game. Neural Netw.
epilepsy. Int. J. Neural Syst. 21, 367–383. 22, 1329–1333.
Collinger, J.L., Wodlinger, B., Downey, J.E., Wang, W., Tyler-Kabara, E.C., Freitag, S., Graf, W., and Kaliske, M.I. (2011). Recurrent neural
Weber, D.J., McMorland, A.J., Velliste, M., Boninger, M.L., and networks for fuzzy data. Integr. Comput. Aided Eng. 18,
Schwartz, A.B. (2013). High-performance neuroprosthetic control 265–280.
by an individual with tetraplegia. Lancet 381, 557–564. Furdea, A., Ruf, C.A., Halder, S., De, M.D., Bogdan, M., Rosenstiel,
Cong, F., Phan, A.H., Zhao, Q., Huttunen-Scott, T., Kaartinen, J., W., and Matuz, T. (2012). A new (semantic) reflexive
Ristaniemi, T., Lyytinen, H., and Cichocki, A. (2012). Benefits brain-computer interface: in search for a suitable classifier.
of multi-domain feature of mismatch negativity extracted by J. Neurosci. Methods 203, 233–240.
nonnegative tensor factorization from EEG collected by low Gage, G.J., Ludwig, K.A., Otto, K.J., Ionides, E.L., and Kipke, D.R.
density array. Int. J. Neural Syst. 22, 1250025. (2005). Naïve coadaptive cortical control. J. Neural Eng. 2,
Cong, F., Phan, A.H., Astikainen, P., Zhao, Q., Wu, Q., Hietanen, J.K., 52–63.
Ristaniemi, T., and Cichocki, A. (2013). Multi-domain feature Galán, F., Nuttin, M., Lew, E., Ferrez, P.W., Vanacker, G., Philips, J.,
extraction for event-related potential through nonnegative and Millán, J.R. (2008). A brain-actuated wheelchair:
multi-way array decomposition from low dense array EEG. Int. J. asynchronous and non-invasive brain-computer interface
Neural Syst. 23, 1350006. for continuous control of robots. Clin. Neurophysiol. 119,
Dai, H., Zhang, H., and Wang, W. (2012). Structural reliability 2159–2169.
assessment by local approximation of limit state functions Gancet, J. (2012). MINDWALKER: going one step further with
using adaptive Markov chain simulation and support vector assistive lower limbs exoskeleton for SCI condition
regression. Comput. Aided Civ. Inf. Eng. 27, 676–686. subjects. J. Proc. IEEE RAS EMBS Int. Conf. Biomed. Robot.
Davis, K.A., Sturges, B.K., Vite, C.H., Ruedebusch, V., Worrell, G., Biomechatronics 1794–1800.
Gardner, A.B., and Leyde, K. (2011). A novel implanted device Garcia-Cuesta, E., Galvan, I.M., and de Castro, A.J. (2011). Recursive
to wirelessly record and analyze continuous intracranial canine discriminant regression analysis to find homogeneous
EEG. Epilepsy Res. 96, 116–122. structures. Int. J. Neural Syst. 21, 95–101.
Debener, S., Minow, F., Emkes, R., Gandras, K., and de Vos, M. Georgopoulos, A.P., Schwartz, A.B., and Kettner, R.E. (1986).
(2012). How about taking a low-cost, small, and wireless EEG Neuronal population coding of movement direction. Science
for a walk. Psychophysiology 49, 1617–1621. 233, 1416–1419.

Brought to you by | provisional account

Unauthenticated | 82.95.122.249
Download Date | 5/30/14 10:31 AM
 A. Ortiz-Rosario and H. Adeli: Brain-computer interface technologies: from signal to action      549

Ghodrati Amiri, G., Abdolahi Rad, A., and Khorasani, M. (2012). Hsu, W.Y. (2011). Continuous EEG signal analysis for asynchronous
Generation of near-field artificial ground motions compatible BCI application. Int. J. Neural Syst. 21, 335–350.
with median predicted spectra using PSO-based neural network Hsu, W.Y. (2012). Application of competitive Hopfield neural network
and wavelet analysis. Comput. Aided Civ. Inf. Eng. 27, 711–730. clustering to brain-computer interface systems. Int. J. Neural
Ghosh-Dastidar, S. and Adeli, H. (2007). Improved spiking neural Syst. 22, 51–62.
networks for EEG classification and epilepsy and seizure Hsu, W.Y. (2013). Single-trial motor imagery classification using
detection. Integr. Comput. Aided Eng. 14, 187–212. asymmetry ratio, phase relation and wavelet-based fractal
Ghosh-Dastidar, S. and Adeli, H. (2009). A new supervised learning features, and their selected combination. Int. J. Neural Syst. 23,
algorithm for multiple spiking neural networks with application 1350007.
in epilepsy and seizure detection. Neural Netw. 22, 1419–1431. Hsu, W.Y., Lin, C.H., Hsu, H.J., Chen, P.H., and Chen, I.R. (2012).
Ghosh-Dastidar, S., Adeli, H., and Dadmehr, N. (2008). Principal Wavelet-based envelope features with automatic EOG artifact
component analysis – enhanced cosine radial basis function removal: application to single-trial EEG data. Expert Syst. Appl.
neural network for robust epilepsy and seizure detection. IEEE 39, 2743–2749.
Trans. Biomed. Eng. 55, 512–518. Huang, D., Qian, K., Fei, D.-Y., Bai, O., Jia, W., and Chen, X. (2012).
Graf, W., Freitag, S., Sickert, J.U., and Kaliske, M. (2012). Structural Electroencephalography (EEG)-based brain-computer interface
analysis with fuzzy data and neural network-based material (BCI): a 2-D virtual wheelchair control based on event-related
description. Comput. Aided Civ. Inf. Eng. 27, 640–654. desynchronization/synchronization and state control. IEEE
Guger, C., Harkam, W., Hertnaes, C., and Pfurtscheller, G. (1999). Trans. Neural Syst. Rehabil. Eng. 20, 379–388.
Prosthetic control by an EEG-based brain-computer Interface Iturrate, I., Antelis, J.M., Kubler, A., and Minguez, J. (2009). A
(BCI). Proc. AAATE 5th Eur. Conf. Adv. Assist. Technol. 590–595. noninvasive brain-actuated wheelchair based on a P300
Guger, C., Schlögl, A., Neuper, C., Walterspacher, D., Strein, T., and neurophysiological protocol and automated navigation. IEEE
Pfurtscheller, G. (2001). Rapid prototyping of an EEG-based Trans. Robot. 25, 614–627.
brain-computer interface (BCI). IEEE Trans. Neural Syst. Iversen, I.H., Ghanayim, N., Kubler, A., Neumann, N., Birbaumer, N.,
Rehabil. Eng. 9, 49–58. and Kaiser, J. (2008). Conditional associative learning
Gupta, R. and Ashe, J. (2009). Offline decoding of end-point forces examined in a paralyzed patient with amyotropic lateral
using neural ensembles: application to a brain-machine sclerosis using brain-computer interface technology. Behav.
interface. IEEE Trans. Neural Syst. Rehabil. Eng. 17, 254–262. Brain Funct. 4, 1–53.
Gustavsson, A., Svensson, M., Jacobi, F., Allgulander, C., Alonso, J., Jiang, X. and Adeli, H. (2004). Wavelet packet-autocorrelation
Beghi, E., and Dodel, R. (2011). Cost of disorders of the brain in function method for traffic flow pattern analysis. Comput.
Europe 2010. J. Eur. Coll. Neuropsychopharmacol. 21, 718–779. Aided Civ. Inf. Eng. 19, 324–337.
Hammer, E.M., Halder, S., Blankertz, B., Sannelli, C., Dickhaus, T., Jiang, X., Ma, Z.J., and Ren, W.X. (2012). Crack detection
Kleih, S., Müller, K.-R., and Kubler, A. (2012). Psychological from the slope of the mode shape using complex
predictors of SMR-BCI performance. Biol. Psychol. 89, 80–86. continuous wavelet transform. Comput. Aided Civ.
Han, F., Wiercigroch, M., Fang, J.A., and Wang, Z. (2011). Excitement Inf. Eng. 27, 187–201.
and synchronization of small-world neuronal networks with Jin, J., Sellers, E.W., and Wang, X. (2012a). Targeting an efficient
short-term synaptic plasticity. Int. J. Neural Syst. 21, 415–425. target-to-target interval for P300 speller brain-computer
Hatsopoulos, N.G. and Donoghue, J.P. (2009). The science of neural interface. Med. Biol. Eng. Comput. 50, 289–296.
interface systems. Annu. Rev. Neurosci. 32, 249–266. Jin, J., Allison, B.Z., Wang, X., and Neuper, C. (2012b). A combined
Hinterberger, T., Weiskopf, N., Veit, R., Wilhelm, B., Betta, E., and brain-computer interface based on P300 potentials and
Birbaumer, N. (2004). An EEG-driven brain-computer interface motion-onset visual evoked potentials. J. Neurosci. Methods
combined with functional magnetic resonance imaging (fMRI). 205, 265–275.
IEEE Trans. Biomed. Eng. 51, 971–974. Jumutc, V., Zayakin, P., and Borisov, A. (2011). Ranking-based
Hinterberger, T., Veit, R., Wilhelm, B., Weiskopf, N., Vatine, J.J., and kernels in applied biomedical diagnostics using support vector
Birbaumer, N. (2005). Neuronal mechanisms underlying control machine. Int. J. Neural Syst. 21, 459–473.
of a brain-computer interface. Eur. J. Neurosci. 21, 3169–3181. Junfei, Q. and Honggui, H. (2010). A repair algorithm for
Hochberg, L.R., Serruya, M.D., Friehs, G.M., Mukand, J.A., radial basis function neural network with application to
Saleh, M., Caplan, A.H., Branner, A., Chen, D., Penn, R.D., and chemical oxygen demand modeling. Int. J. Neural Syst.
Donoghue, J.P. (2006). Neuronal ensemble control of prosthetic 20, 63–74.
devices by a human with tetraplegia. Nature 442, 164–171. Kamousi, B., Liu, Z., and He, B. (2005). Classification of motor
Hochberg, L.R., Bacher, D., Jarosiewicz, B., Masse, N.Y., imagery tasks for brain-computer interface applications by
Simeral, J.D., Vogel, J., Haddadin, S., Liu, J., Cash, S.S., and means of two equivalent dipoles analysis. IEEE Trans. Neural
Van der, P. (2012). Reach and grasp by people with tetraplegia Syst. Rehabil. Eng. 13, 166–171.
using a neurally controlled robotic arm. Nature 485, 372–375. Kayagil, T.A., Bai, O., Henriquez, C.S., Lin, P., Furlani, S.J.,
Hou, J., Chen, Z., Qin, X., and Zhang, D. (2011). Automatic image Vorbach, S., and Hallett, M. (2009). A binary method for
search based on improved feature descriptors and decision simple and accurate two-dimensional cursor control from
tree. Integr. Comput. Aided Eng. 18, 167–180. EEG with minimal subject training. J. Neuroeng. Rehabil. 6,
Hsiao, F.Y., Wang, S.S., Wang, W.C., Wen, C.P., and Yu, W.D. (2012). 1–14.
Neuro-fuzzy cost estimation model enhanced by fast messy Kilgard, M.P., Vazquez, J.L., Engineer, N.D., and Pandya, P.K.
genetic algorithms for semiconductor hookup construction. (2007). Experience dependent plasticity alters cortical
Comput. Aided Civ. Inf. Eng. 27, 764–781. synchronization. Hear. Res. 229, 171–179.

Brought to you by | provisional account

Unauthenticated | 82.95.122.249
Download Date | 5/30/14 10:31 AM
550      A. Ortiz-Rosario and H. Adeli: Brain-computer interface technologies: from signal to action

Koyama, S., Chase, S.M., Whitford, A.S., Velliste, M., Schwartz, A.B., Lopez-Gordo, M.A., Pelayo, F., Prieto, A., and Fernandez, E. (2012).
and Kass, R.E. (2010). Comparison of brain-computer interface An auditory brain-computer interface with accuracy prediction.
decoding algorithms in open-loop and closed-loop control. Int. J. Neural Syst. 22, 1250009.
J. Comput. Neurosci. 29, 73–87. Lotte, F., Congeado, M., Lecuyer, A., Lamarche, F., and Arnaldi, B.
Kreilinger, A. (2012). Error potential detection during continuous (2007). A review of classification algorithms for EEG-based
movement of an artificial arm controlled by brain-computer brain-computer interfaces. J. Neural Eng. 4, R1–R13.
interface. Med. Biol. Eng. Comput. 50, 223–230. Luis de Mello, F., Strauss, E., and Fernandes de Oliveira, A.A. (2011).
Krusienski, D. and Shih, J. (2010). Control of a visual keyboard Computer theory and digital image processing applied to
using an electrocorticographic brain-computer interface. emotional brain activation recognition. Integr. Comput. Aided
Neurorehabil. Neural Repair 25, 323–331. Eng. 18, 157–166.
Krusienski, D.J., Sellers, E.W., McFarland, D.J., Vaughan, T.M., Malik, W.Q., Brown, E.N., Hochberg, L.R., and Truccolo, W. (2011).
and Wolpaw, J.R. (2008). Towards enhanced P300 speller Efficient decoding with steady-state Kalman filter in neural
performance. J. Neurosci. Methods 167, 15–21. interface systems. IEEE Trans. Neural Syst. Rehabil. Eng. 19,
Krusienski, D.J., McFarland, D.J., and Wolpaw, J.R. (2012). Value of 25–34.
amplitude, phase, and coherence features for a sensorimotor Manyakov, N.V., Chumerin, N., and Van Hulle, M.M. (2012).
rhythm-based brain-computer interface. Brain Res. Bull. 87, Multichannel decoding for phase-coded SSVEP brain-computer
130–134. interface. Int. J. Neural Syst. 22, 1250022.
Kübler, A. (2005). Patients with ALS can use sensorimotor rhythms to Martins, A.L.D., Mascarenhas, N.D.A., and Suazo, C.A.T. (2011).
operate a brain-computer interface. Neurology 64, 1775–1777. Spatio-temporal resolution enhancement of vocal tract MRI
Kübler, A. and Birbaumer, N. (2008). Brain-computer interfaces sequences based on image registration. Integr. Comput. Aided
and communication in paralysis: extinction of goal directed Eng. 18, 143–156.
thinking in completely paralysed patients. Clin. Neurophysiol. Martis, R.J., Acharya, U.R., Tan, J.H., Petznick, A., Yanti, R.,
119, 2658–2666. Chua, C.K., Ng, E.Y., and Tong, L. (2012). Application of
Lance, B.J., Kerick, S.E., Ries, A.J., Oie, K.S., and McDowell, K. empirical mode decomposition (EMD) for automated
(2012). Brain-computer interface technologies in the coming detection of epilepsy using EEG signals. Int. J. Neural Syst.
decades. Proc. IEEE 100, 1585–1599. 22, 1250027.
Larkindale, J., Yang, W., Hogan, P.F., Simon, C.J., Zhang, Y., Jain, A., Maynard, E.M., Nordhausen, C.T., and Normann, R.A. (1997).
Habeeb-Louks, E.M., and Cwik, V.A. (2013). Cost of illness for The Utah intracortical electrode array: a recording structure
neuromuscular diseases in the U.S. Muscle Nerve for potential brain-computer interfaces. Electroencephalogr.
DOI: 10.1002/mus.23942. Clin. Neurophysiol. 102, 228–239.
Laubach, M., Wessberg, J., and Nicolelis, M.A. (2000). Cortical McFarland, D.J., Anderson, C.W., Müller, K.R., Schlögl, A., and
ensemble activity increasingly predicts behaviour outcomes Krusienski, D.J. (2006). BCI Meeting 2005 workshop on BCI
during learning of a motor task. Nature 405, 567–571. signal processing: feature extraction and translation. IEEE
Lauer, R.T., Peckham, P.H., and Kilgore, K.L. (1999). EEG-based Trans. Neural Syst. Rehabil. Eng. 14, 135–139.
control of a hand grasp neuroprosthesis. NeuroReport 10, McFarland, D.J., Sarnacki, W.A., and Wolpaw, J.R. (2010).
1767–1771. Electroencephalographic (EEG) control of three-dimensional
Lemm, S., Blankertz, B., Curio, G., and Müller, K.R. (2005). Spatio- movement. J. Neural Eng. 7, 036007.
spectral filters for improving the classification of single trial Mehta, N.A., Hameed, S.H.S., and Jackson, M.M. (2011). Optimal
EEG. IEEE Trans. Biomed. Eng. 52, 1541–1548. control strategies for an SSVEP-based brain-computer
Leuthardt, E.C., Schalk, G., Wolpaw, J.R., Ojemann, J.G., and interface. Int. J. Hum. Comput. Int. 27, 85–101.
Moran, D.W. (2004). A brain-computer interface using Mellinger, J., Schalk, G., Braun, C., Preissl, H., Rosenstiel, W.,
electrocorticographic signals in humans. J. Neural Eng. 1, 63–71. Birbaumer, N., and Kubler, A. (2007). An MEG-based
Li, J., Liang, J., Zhao, Q., Hong, K., and Zhang, L. (2013). Design brain-computer interface (BCI). Neuroimage 36,
of wheelchair assistive system directly steered by human 581–593.
thoughts. Int. J. Neural Syst. 23, 1350013. Millán, J.R. and Carmena, J. (2010). Invasive or noninvasive:
Liao, L.-D., Chang, J.-Y., Lin, C.-T., Ko, L.-W., McDowell, K., understanding brain-machine interface technology. IEEE Eng.
Wickenden, A.E., and Gramann, K. (2012). Biosensor Med. Biol. Mag. 29, 16–22.
technologies for augmented brain-computer interfaces in Millán, J.R., Renkens, F., Mourinũo, J., and Gerstner, W. (2004).
the next decades. Proc. IEEE 100, 1553–1566. Brain-actuated interaction. Artif. Intell. 159, 241–259.
Lin, C.M., Ting, A.B., Hsu, C.F., and Chung, C.M. (2012). Adaptive Moritz, C.T., Perlmutter, S.I., and Fetz, E.E. (2008). Direct control
control for MIMO uncertain nonlinear systems using recurrent of paralysed muscles by cortical neurons. Nature 456,
wavelet neural network. Int. J. Neural Syst. 22, 37–50. 639–643.
Linderman, M.D., Santhanam, G., Gilja, V., O’Driscoll, S., Yu, B.M., Müller-Putz, G.R., and Pfurtscheller, G. (2008). Control of an
Afshar, A., Kemere, C.T., and Meng, T.H. (2008). Signal electrical prosthesis with an SSVEP-Based BCI. IEEE Trans.
processing challenges for neural prostheses. IEEE Signal Proc. Biomed. Eng. 55, 361–364.
Mag. 25, 18–28. Müller-Putz, G.R., Scherer, R., Pfurtscheller, G., and Rupp, R. (2005).
Liu, C., Wang, J., Chen, Y.Y., Deng, B., Wei, X.L., and Li, H.Y. (2013). EEG-based neuroprosthesis control: a step towards clinical
Closed-loop control of the thalamocortical relay neuron’s practice. Neurosci. Lett. 382, 169–174.
parkinsonian state based on slow variable. Int. J. Neural Syst. Müller-Putz, G.R., Pokorny, C., Klobassa, D.S., and Horki,
23, 1350017. P. (2013). A single switch BCI based on passive and

Brought to you by | provisional account

Unauthenticated | 82.95.122.249
Download Date | 5/30/14 10:31 AM
 A. Ortiz-Rosario and H. Adeli: Brain-computer interface technologies: from signal to action      551

imagined movements: towards restoring communication Sankari, Z., Adeli, H., and Adeli, A. (2012). Wavelet coherence model
in minimally conscious patients. Int. J. Neural Syst. 23, for diagnosis of Alzheimer’s disease. Clin. EEG Neurosci. 43,
1250037. 268–278.
Murray, C.J.L. and Lopez, A.D. (1996).The Global Burden of Scherberger, H. (2009). Neural control of motor prostheses. Curr.
Disease: A Comprehensive Assessment of Mortality and Opin. Neurobiol. 19, 629–633.
Disability from Diseases, Injuries, and Risk Factors in 1990 and Schwartz, A.B., Cui, X.T., Weber, D.J., and Moran, D.W. (2006).
Projected to 2020. (Cambridge, MA, USA: Harvard School of Brain-controlled interfaces: movement restoration with neural
Public Health). prosthetics. Neuron 52, 205–220.
Nam, C.S., Schalk, G., and Jackson, M.M. (2011). Editorial: current Setiono, R., Baesens, B., and Mues, C. (2011). Rule extraction from
trends in brain-computer interface (BCI) research and minimal neural network for credit card screening. Int. J. Neural
development. Int. J. Hum. Comput. Int. 27, 40912. Syst. 21, 265–276.
Navarro, X., Krueger, T.B., Lago, N., Micera, S., Stieglitz, T., and Shih, J.J., Krusienski, D.J., and Wolpaw, J.R. (2012). Brain-computer
Dario, P. (2005). A critical review of interfaces with the interface in medicine. Mayo Clin. Proc. 87, 268–279.
peripheral nervous system for the control of neuroprostheses Shin, H-C., Aggarwal, V., Acharya, S., Thakor, N.V., and
and hybrid bionic systems. J. Peripher. Nerv. Syst. 10, Schieber, M.H. (2010). Neural decoding of finger movements
229–258. using Skellam-based maximum-likelihood decoding. IEEE
Nicolelis, M.A.L. and Lebedev, M.A. (2009). Principles of Trans. Biomed. Eng. 57, 754–760.
neural ensemble physiology underlying the operation of Sitaram, R., Zhang, H.H., Guan, C.T., Thulasidas, M., Hoshi, Y.,
brain-machine interfaces. Nat. Rev. Neurosci. 10, 530–540. Ishikawa, A., Shimizu, K., and Birbaumer, N. (2006). Temporal
Osornio-Rios, R.A., Amezquita-Sanchez, J.P., Romero-Troncoso, R.J., classification of multichannel near-infrared spectroscopy
and Garcia-Perez, A. (2012). Music-neural network analysis for signals of motor imagery for developing a brain-computer
locating structural damage in Truss-type structures by means interface. Neuroimage 34, 1416–1427.
of vibrations. Comput. Aided Civ. Inf. Eng. 27, 687–698. Spüler, M., Bensch, M., Kleih, S., Rosenstiel, W., Bogdan, M.,
Pasqualotto, E., Federici, S., and Belardinelli, M.O. (2012). Towards and Kübler, A. (2012). Online use of error-related potentials
functioning and usable brain-computer interfaces (BCIs): a in healthy users and people with sever motor impairment
literature review. Disabil. Rehabil. Assist. Technol. 7, 89–103. increases performance of P300-BCI. Clin. Neurophysiol. 123,
Patrinos, P., Alexandridis, A., Ninos, K., and Sarimveis, H. (2010). 1328–1337.
Variable selection in nonlinear modeling based on RBF Stansbury, L.G., Branstetter, J.G., and Lalliss, S.J. (2007).
networks and evolutionary computation. Int. J. Neural Syst. 20, Amputation in military trauma surgery. J. Trauma 63,
365–379. 940–944.
Pfurtscheller, G., Guger, C., Müller, G., Krausz, G., and Neuper, C. Sykacek, P., Roberts, S.J., and Stokes, M. (2004). Adaptive BCI
(2000). Brain oscillations control hand orthosis in a based on variational Bayesian Kalman filtering; an empirical
tetraplegic. Neurosci. Lett. 292, 211–214. evaluation. IEEE Trans. Biomed. Eng. 51, 719–727.
Pfurtscheller, G., Müller, G.R., Pfurtscheller, J., Gerner, H.J., and Tao, H., Zain, J.M., Ahmed, M.M., Abdalla, A.N., and Jing, W. (2012).
Rupp, R. (2003). ‘Thought’ – control of functional electrical A wavelet-based particle swarm optimization algorithm for
stimulation to restore hand grasp in a patient with tetraplegia. digital image watermarking. Integr. Comput. Aided Eng. 19,
Neurosci. Lett. 351, 33–36. 81–91.
Puscasu, G. and Codres, B. (2011). Nonlinear system identification Tomasevic, N.M., Neskovic, A.M., and Neskovic, N.J. (2012). Artificial
and control based on modular neural networks. Int. J. Neural neural network-based approach to EEG signal simulation. Int. J.
Syst. 21, 319–334. Neural Syst. 22, 1250008.
Qin, L., Ding, L., and He, B. (2005). Motor imagery classification Torriente, I., Valdes-Sosa, M., Ramirez, D., and Bobes, M.A.
by means of source analysis for brain computer interface (1999). Visual evoked potentials related to motion-onset are
application. J. Neural Eng. 1, 135–141. modulated by attention. Vision Res. 39, 4122–4139.
Rakotomamonjy, A. and Guigue, V. (2008). BCI competition III: Vidaurre, C., Schlögl, A., Cabeza, R., Scherer, R., and
dataset II-ensemble of SVMs for BCI P300 Speller. IEEE Trans. Pfurtscheller, G. (2006). A fully on-line adaptive BCI. IEEE
Biomed. Eng. 55, 1147–1154. Trans. Biomed. Eng. 53, 1214–1219.
Rangaprakash, D., Hu, X., and Deshpande, G. (2013). Phase Waldert, S., Pistohl, T., Braun, C., Ball, T., Aertsen, A., and Mehring, C.
synchronization in brain networks derived from correlation (2009). A review on directional information in neural signals for
between probabilities of recurrences in functional MRI data. brain-machine interfaces. J. Physiol. Paris 103, 244–254.
Int. J. Neural Syst. 23, 1350003. Wandekokem, E.D., Mendel, E., Fabris, F., Valentim, M., Batista, R.J.,
Rossello, J.L., Canals, V., Morro, A., and Oliver, A. (2012). Hardware Varejao, F.M., and Rauber, T.W. (2011). Diagnosing multiple
implementation of stochastic spiking neural networks. Int. J. faults in oil rig motor pumps using support vector machine
Neural Syst. 22, 1250014. classifier ensembles. Integr. Comput. Aided Eng. 18, 61–74.
Saliminejad, S. and Gharaibeh, N.G. (2012). A spatial-Bayesian Wessberg, J., Stambaugh, C.R., Kralik, J.D., Beck, P.D., Laubach, M.,
technique for imputing pavement network repair data. Comput. Chapin, J.K., and Kim, J. (2000). Real-time prediction of hand
Aided Civ. Inf. Eng. 27, 594–607. trajectory by ensembles of cortical neurons in primates. Nature
Sankari, Z. and Adeli, H. (2011). Probabilistic neural networks 408, 361–365.
for EEG-based diagnosis of Alzheimer’s disease using White, J.R., Levy, T., Bishop, W., and Beaty, J.D. (2010). Real-time
conventional and wavelet coherence. J. Neurosci. Methods 197, decision fusion for multimodal neural prosthetic device. PLoS
165–170. ONE 5, 1.

Brought to you by | provisional account

Unauthenticated | 82.95.122.249
Download Date | 5/30/14 10:31 AM
552      A. Ortiz-Rosario and H. Adeli: Brain-computer interface technologies: from signal to action

Wolpaw, J.R., McFarland, D.J., and Bizzi, E. (2004). Control of based on particle swarm optimization. Int. J. Neural Syst. 20,
a two-dimensional movement signal by a noninvasive 109–116.
brain-computer interface in humans. Proc. Natl. Acad. Sci. USA Xiang, J. and Liang, M. (2012). Wavelet-based detection of beam
101, 17849–17854. cracks using modal shape and frequency measurements.
Wu, W., Black, M.J., Gao, Y., Bienenstock, E., Serruya, M., Comput. Aided Civ. Inf. Eng. 27, 439–454.
Shaikhouni, A., and Donoghue, J.P. (2003). Neural decoding Yamanishi, T., Liu, J.Q., and Nishimura, H. (2012). Modeling
of cursor motion using a Kalman filter. Adv. Neural Info. Proc. fluctuations in default-mode brain network using a spiking
Syst. 15, 130–140. neural network. Int. J. Neural Syst. 22, 1250016.
Wu, D., Warwick, K., Ma, Z., Gasson, M.N., Burgess, J.G., Pan, Yang, B., Yan, G., Yan, R., and Wu, T. (2007). Adaptive subject-based
S., and Aziz, T.Z. (2010). Prediction of Parkinson’s disease feature extraction in brain-computer interfaces using wavelet
tremor onset using radial basis function neural network packet best basis decomposition. Med. Eng. Phys. 29, 48–53.

Alexis Ortiz-Rosario was born in San Juan, Puerto Rico. He received Hojjat Adeli is Professor of Civil, Environmental, and Geodetic Engi-
his B.S. in Industrial Engineering from the University of Puerto neering, Biomedical Engineering, Biomedical Informatics, Electrical
Rico (UPR) at Mayagüez, Puerto Rico, in 2011. He is currently a and Computer Engineering, Neurological Surgery, and Neurosci-
PhD student in Biomedical Engineering and a Graduate Research ence at The Ohio State University. He has authored/co-authored 15
Associate at the Motor Systems and Neurophysiology Lab at The books. He is the author of Automated EEG-based Diagnosis of Neu-
Ohio State University. His current research interests include neural rological Disorders – Inventing the Future of Neurology (CRC Press,
networks, neural signal processing, brain-computer interfaces, 2010). In 1998 he received The Ohio State University’s highest
motor systems, and electrophysiology. research honor, the Distinguished Scholar Award ‘in recognition of
extraordinary accomplishment in research and scholarship’. He is
a Fellow of AAAS and IEEE. He is the Editor-in-Chief of the interna-
tional research journals Computer-Aided Civil and Infrastructure
Engineering which he founded in 1986 and Integrated Computer-
Aided Engineering which he founded in 1993. He is also the Editor-
in-Chief of the International Journal of Neural Systems.

Brought to you by | provisional account

Unauthenticated | 82.95.122.249
Download Date | 5/30/14 10:31 AM