A concept for extending the applicability of constraint-induced movement therapy through motor cortex activity feedback using a neural prosthesis.
ABSTRACT This paper describes a concept for the extension of constraint-induced movement therapy (CIMT) through the use of feedback of primary motor cortex activity. CIMT requires residual movement to act as a source of feedback to the patient, thus preventing its application to those with no perceptible movement. It is proposed in this paper that it is possible to provide feedback of the motor cortex effort to the patient by measurement with near infrared spectroscopy (NIRS). Significant changes in such effort may be used to drive rehabilitative robotic actuators, for example. This may provide a possible avenue for extending CIMT to patients hitherto excluded as a result of severity of condition. In support of such a paradigm, this paper details the current status of CIMT and related attempts to extend rehabilitation therapy through the application of technology. An introduction to the relevant haemodynamics is given including a description of the basic technology behind a suitable NIRS system. An illustration of the proposed therapy is described using a simple NIRS system driving a robotic arm during simple upper-limb unilateral isometric contraction exercises with healthy subjects.
- SourceAvailable from: rsta.royalsocietypublishing.org[Show abstract] [Hide abstract]
ABSTRACT: The acute onset of a neurological deficit is the key clinical feature of stroke. In most cases, however, pathophysiological changes in the cerebral vasculature precede the event, often by many years. Persisting neurological deficits may also require long-term rehabilitation. Hence, stroke may be considered a chronic disease, and diagnostic and therapeutic efforts must include identification of specific risk factors, and the monitoring of and interventions in the acute and subacute stages, and should aim at a pathophysiologically based approach to optimize the rehabilitative effort. Non-invasive optical techniques have been experimentally used in all three stages of the disease and may complement the established diagnostic and monitoring tools. Here, we provide an overview of studies using the methodology in the context of stroke, and we sketch perspectives of how they may be integrated into the assessment of the highly dynamic pathophysiological processes during the acute and subacute stages of the disease and also during rehabilitation and (secondary) prevention of stroke.Philosophical Transactions of The Royal Society A Mathematical Physical and Engineering Sciences 11/2011; 369(1955):4470-94. · 2.89 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: Background; Brain-computer interfaces (BCIs) were recently recognized as a method to promote neuroplastic effects in motor rehabilitation. The core of a BCI is a decoding stage by which signals from the brain are classified into different brain-states. The goal of this paper was to test the feasibility of a single trial classifier to detect motor execution based on signals from cortical motor regions, measured by functional near-infrared spectroscopy (fNIRS), and the response of the autonomic nervous system. An approach that allowed for individually tuned classifier topologies was opted for. This promises to be a first step towards a novel form of active movement therapy that could be operated and controlled by paretic patients.Methods; Seven healthy subjects performed repetitions of an isometric finger pinching task, while changes in oxy- and deoxyhemoglobin concentrations were measured in the contralateral primary motor cortex and ventral premotor cortex using fNIRS. Simultaneously, heart rate, breathing rate, blood pressure and skin conductance response were measured. Hidden Markov models (HMM) were used to classify between active isometric pinching phases and rest. The classification performance (accuracy, sensitivity and specificity) was assessed for two types of input data: (i) fNIRS-signals only and (ii) fNIRS- and biosignals combined. Results; fNIRS data were classified with an average accuracy of 79.4%, which increased significantly to 88.5% when biosignals were also included (p=0.02). Comparable increases were observed for the sensitivity (from 78.3% to 87.2%, p=0.008) and specificity (from 80.5% to 89.9%, p=0.062).Conclusions; This study showed, for the first time, promising classification results with hemodynamic fNIRS data obtained from motor regions and simultaneously acquired biosignals. Combining fNIRS data with biosignals has a beneficial effect, opening new avenues for the development of brain-body-computer interfaces for rehabilitation applications. Further research is required to identify the contribution of each modality to the decoding capability of the subject's hemodynamic and physiological state.Journal of NeuroEngineering and Rehabilitation 01/2013; 10(1):4. · 2.57 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: Brain-Computer Interfaces (BCI) can potentially be used to aid in the recovery of lost motor controlin a limb following stroke. BCIs are typically used by subjects with no damage to the brain thereforerelatively little is known about the technical requirements for the design of a rehabilitative BCI forstroke. 32-channel electroencephalogram (EEG) was recorded during a finger-tapping task from 10 healthysubjects for one session and 5 stroke patients for two sessions approximately 6 months apart. Anoff-line BCI design based on Filter Bank Common Spatial Patterns (FBCSP) was implemented to testand compare the efficacy and accuracy of training a rehabilitative BCI with both stroke-affected andhealthy data. Stroke-affected EEG datasets have lower 10-fold cross validation results than healthy EEG datasets.When training a BCI with healthy EEG, average classification accuracy of stroke-affected EEG islower than the average for healthy EEG. Classification accuracy of the late session stroke EEG isimproved by training the BCI on the corresponding early stroke EEG dataset. This exploratory study illustrates that stroke and the accompanying neuroplastic changes associatedwith the recovery process can cause significant inter-subject changes in the EEG features suitable formapping as part of a neurofeedback therapy, even when individuals have scored largely similar withconventional behavioural measures. It appears such measures can mask this individual variability incortical reorganization. Consequently we believe motor retraining BCI should initially be tailored toindividual patients.Journal of NeuroEngineering and Rehabilitation 01/2014; 11(1):9. · 2.57 Impact Factor
Hindawi Publishing Corporation
Computational Intelligence and Neuroscience
Volume 2007, Article ID 51363, 9 pages
A Concept for Extending the Applicability of
Constraint-Induced Movement Therapy through Motor
Cortex Activity Feedback Using a Neural Prosthesis
Tomas E. Ward,1Christopher J. Soraghan,2Fiachra Matthews,3and Charles Markham4
1Department of Electronic Engineering, National University of Ireland, Maynooth, County Kildare, Ireland
2Department of Computer Science and Department of Experimental Physics, National University of Ireland,
Maynooth, County Kildare, Ireland
3Hamilton Institute, National University of Ireland, Maynooth, County Kildare, Ireland
4Department of Computer Science, National University of Ireland, Maynooth, County Kildare, Ireland
Correspondence should be addressed to Tomas E. Ward, email@example.com
Received 17 February 2007; Revised 11 May 2007; Accepted 14 July 2007
Recommended by Fabio Babiloni
This paper describes a concept for the extension of constraint-induced movement therapy (CIMT) through the use of feedback of
primary motor cortex activity. CIMT requires residual movement to act as a source of feedback to the patient, thus preventing its
application to those with no perceptible movement. It is proposed in this paper that it is possible to provide feedback of the motor
to drive rehabilitative robotic actuators, for example. This may provide a possible avenue for extending CIMT to patients hitherto
attempts to extend rehabilitation therapy through the application of technology. An introduction to the relevant haemodynamics
is given including a description of the basic technology behind a suitable NIRS system. An illustration of the proposed therapy is
described using a simple NIRS system driving a robotic arm during simple upper-limb unilateral isometric contraction exercises
with healthy subjects.
Copyright © 2007 Tomas E. Ward et al.ThisisanopenaccessarticledistributedundertheCreativeCommonsAttributionLicense,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Strokes are characterised by an acute, nonconvulsive loss of
neurological function as a result of an ischemic or hemor-
rhagic intracranial vascular event . Worldwide, there are
over 20 million cases of stroke each year  and of these ap-
proximately 75% are nonfatal with survivors left with a spec-
trum of poststroke disabilities ranging from mild numbness
to severe motor and cognitive impairments. The dysfunc-
tion introduced depends on the site and extent of the infarc-
tion. Immediately following stroke, there is generally some
degree of spontaneous recovery where some lost function is
restored as a result of collateral circulation, reduction in in-
flammation, and haematoma compression among other fac-
tors. However, there are in nearly all cases significant resid-
ual neurological impairment, which if left untreated will re-
sult in severe degradation in life quality for the survivor. It is
not surprising then that stroke is the leading cause of phys-
ical disability in Europe and the United States , and with
75% of stroke survivors suffering syndromes severe enough
to affect their employability, the economic cost of the cer-
brovascular disease stretches far beyond the immediate med-
ical one. Poststroke rehabilitation is therefore critical to re-
store as much function as possible for the patient. This takes
the form of neuro-rehabilitation, an interdisciplinary branch
of clinical and medical science; the purpose of which is to re-
store neurological function and quality of life to people fol-
lowing disease or injury of the nervous system. Neuroreha-
bilitation science draws on many techniques and therapies
but the cornerstone of most treatments lies in physical ther-
The basic tenet of physical therapy or motor rehabilitation
as it is often termed is that repetitive practise of proscribed
movement can have a highly significant effect on rehabilita-
tion outcome .
2 Computational Intelligence and Neuroscience
Currently, a particular type of motor rehabilitation
has been shown to be highly effective for use in hemiplegic
stroke rehabilitation . CIMT requires the subject have
the unaffected limb constrained while they are encouraged
actively to use the affected limb over long periods. With
large periods of practise, the weakened side is strengthened
significantly—possibly as a result of cortical reorganisation
and changes in motor cortex excitability . It appears that
this method has statistically significant outcome improve-
ments over equal intensity approaches [9, 10] and is cur-
rently the focus of concerted refinement and development to
expand its theoretical basis and extend the application scope
The flurry of research around this technique has had
an impact on rehabilitation engineering where over the past
few years the therapy has been augmented by robot-assisted
training . Such techniques extend CIMT to patients who
have such severe disability that they are unable to engage
in unassisted movement. This concept is underpinned by
studies which have shown that the benefits of CIMT could
be extended to such patients through the application of ex-
ternal forces applied to the limb  or functional elec-
trical stimulation [14, 15]. The philosophy behind the ap-
cess can be enhanced further for this group if some mea-
sure of attempted activity in the motor regions could be
presented either as a direct form of biofeedback or har-
nessed as a trigger to induce robotic-assisted movement or
FES. Functional magnetic resonance imaging (fMRI) studies
have shown that increases in bilateral cortical activation 
are exhibited during CIMT; therefore appropriate feedback
could be established using a brain-machine interface driven
by signals derived directly from the motor cortical areas—a
neural or more specifically, neurocortical prosthesis. It has
been widely reported that NIRS systems are capable of de-
neural prosthesis for this application is a NIRS-based brain
computer interface (NIRS-BCI). The successful evoking of
cortical NIRS responses could serve as the triggering event
for biofeedback. NIRS has the additional benefit of showing
oxyhaemoglobin (HbO) level changes as well as blood vol-
ume and deoxyhaemoglobin (Hb) changes which are the re-
ported etiology of the MRI signal. The idea of extending the
biofeedback loop directly to the motor areas responsible for
movement is novel in this case of CIMT and the technique
sits well with current opinions in neurorehabilitation which
advocate enhanced motor learning techniques [18, 20]. The
remainder of this paper is as follows. Section 2 begins with a
description of the rehabilitation context for stroke survivors
including a short presentation of the CIMT model. In ad-
dition, some background to BCI and robotics in a neurore-
habilitation context is given including the relevant physio-
logical measurement modality of NIRS. Section 3 comprises
a technical illustration of the proposed concept to facilitate
an appreciation for obstacles and issues facing practical em-
bodiments of the idea. Section 4 discusses implications and
prospects before a short summary is given.
2. BACKGROUND AND RELATED WORK
Stroke rehabilitation therapies have until recently been char-
acterised by empirically derived approaches rather than on
the basis of scientifically derived theories. Traditional prac-
tises have a compensatory philosophy where targeted mus-
cular and action accommodation techniques serve to cir-
cumnavigate impaired function. Lack of standards, poor val-
idation, poor evaluation, and above all the lack of a neu-
roscientific basis has meant that neurorehabilitation clini-
cal practises have languished outside the realm of evidence-
based medicine. The advent of CIMT changed this percep-
tion and revolutionised rehabilitation medicine. CIMT is de-
rived from a rigorously constructed conceptual framework
which has its origins in the theory of learned non-use—an
explanation for certain neurocortical and behavioural phe-
nomena evident in monkey models of neurological dysfunc-
tion. . In such models, the consequence of the paretic
limb, for example, is the onset of a neuroplastic process in
which the motor circuits undergo alteration which has de-
generative impacts for the affected limb. CIMT aims to undo
and forced repetitive training—a practise somewhat analo-
gous to the use of the eye patch in amblyopia or lazy eye.
With its efficacy confirmed during the largest ever controlled
trial in neurological rehabilitation , the continued clin-
ical practise of this therapy and its further refinement and
extension are assured.
The rigorous psychological and neurological basis un-
derlying CIMT makes it very amenable to integration with
assistive technologies which yield quantitative measures and
assessment criteria. The development of robotic actuators in
procedures is a natural development for future CIMT vari-
ants. The recent literature exhibits a growing and versatile
range of potential systems that may be effective in applica-
tion with CIMT.
The use of robotic systems as aids in neurorehabilitation is
not new with systems such as MIT-MANUS  demon-
strating the efficacy of the technique almost ten years ago.
Their application to neurorehabilitation is quite natural as it
is well known that intensive goal-directed movement repeti-
tion facilitates improved recovery outcome following stroke
[23, 24] and as robots can engage in repetitive tasks consis-
tently and unobtrusively, they are excellent deliverers of re-
habilitation therapy. Clinical effectiveness has been reported
in several studies [25–27] and it seems that these rehabilita-
tive devices will be incorporated into standard clinical prac-
tise in the near future. Comprehensive reviews of such de-
Such systems also have the benefit that they may be altered to
incorporate automatic kinematic and kinetic data collection
allowing the possibility of quantitative measures of subject
performance and recovery of function. While there are many
devices reported at present, the common feature is their fa-
cilitation of repetitious exercise. The most notable recent
Tomas E. Ward et al.3
developments which provide context for this work are elec-
tromyogram feature-triggered systems reported by Dipietro
activate robot assistance; such patients might be able to gener-
cient movement to trigger the robot.” Previous systems rely on
exceeding kinematic/kinetic thresholds based on limb veloc-
ity, for example, to trigger movement. Therefore, Dipietro’s
system can be regarded as harnessing peripheral nervous sys-
tem activity as recorded through electrical muscular activity
to trigger feedback. It is proposed in the present paper that
central nervous system activity measures may serve as an al-
ternative trigger—a concept that suggests a new application
area for brain-computer interfaces.
Brain-computer interfaces (BCIs) are devices that act as neu-
transfer between the brain and the outside world indepen-
dent of the peripheral nervous system. While the primary
focus of BCI research within neurorehabilitation has been
to provide assistive technology to enable communication for
the severely disabled, there have been suggestions that the
technology may have something to offer in terms of phys-
ical recovery for certain conditions through reinforcement
of damaged neural pathways , plasticity-induced corti-
cal reorganization , and triggering of functional electri-
cal stimulation . While there has been movement of BCI
research towards this area, most rehabilitation efforts have
been directed towards harnessing neural prostheses for con-
trolling robotic limbs for reaching and manipulating tasks
or control of wheelchairs. To these particular ends, great
progress has been made in terms of practicality [34, 35],
speed [36, 37], and ease of use . Such advances are con-
tinuing, however the more subtle application of the tech-
nology as a biofeedback mechanism for physical rehabili-
tation has hitherto been underdeveloped. One of the most
impressive attempts in this direction is the Brain-Orthosis-
Interface reported as a solution for chronic stroke suffer-
ers . The technology based on magnetoencephalogra-
phy methods monitors sensorimotor rhythm to derive con-
trol signals to open and close an orthotic hand coupled to
the patient’s own. In this way, the patient receives enhanced
feedback through both watching and feeling their own hand
moving. Such operant conditioning enhances the biofeed-
back process and improves neural prosthesis performance.
Such a case represents a more extreme rehabilitative appli-
cation of a BCI in that the neural prosthesis is a permanent
one. The paradigm presented in this paper casts the BCI in
the role of a temporary neural prosthetic splint that pro-
vides feedback in lieu of feedback from actual movement.
The contribution of this paper is in this context. When, if
tional forms of CIMT may be applied probably removing the
necessity for the BCI. A related concept is the provision of an
transcranial direct current stimulation of the motor cortex is
used to improve rehabilitation outcome . This can be in-
terpreted as a neural prosthetic encouraging cortical activa-
tion associated with movement.
A near infrared spectroscopy-based brain-computer inter-
face utilises an optical modality for inferring changes in
brain state. It is possible to measure changes in cerebral
blood volume and oxygenation associated with cortical ac-
tivity through the use of light in the 600–1000nm wave-
optical absorption and scattering properties of scalp, hair,
skull, and the meninges surrounding the brain allow pho-
tons of these wavelengths to penetrate in to the surface of the
cortex where they undergo scattering and absorption events
with a wide range of chromophores in the tissue. The signifi-
cant scattering component means that a small proportion of
the injected light will exit at some distance from the source
carrying information about chromophore concentration dy-
namics at the upper surface of the brain. A suitably sensi-
tive well-positioned detector can detect this photon flux and
allow noninvasive monitoring of these changes. There are a
number of chromophores in brain tissue in this band whose
optical properties are correlated with mental activation. Of
these, the most germane is haemoglobin—the oxygen carry-
ing molecule of the body. Haemoglobin exists principally in
two forms in the body: an oxidised state and a reduced state.
These two states generally referred to as oxyhaemoglobin
(HbO) and deoxyhaemoglobin (Hb) have distinctly different
absorption spectra allowing their relative concentrations to
be determined through multiple wavelength interrogation.
During concerted cortical activity, a neurovascular process
occurs whereby changes occur in cerebral blood flow, vol-
ume, and metabolic rate of consumption. This manifests it-
self principally as an increased demand for oxygen with the
local vasculature responding through flooding the cortical
area and surrounding tissue with oxygenated haemoglobin.
Usually this is accompanied by a corresponding drop in de-
oxyhaemoglobin concentration—a component thought to
be responsible for the signal recorded during fMRI. The rel-
ative changes in haemoglobin can be distinguished through
band described above and therefore changes in cortical acti-
is the basis of NIRS-BCI. A detailed review of near infrared
spectroscopy techniques for biomedical application can be
found in .
The measurement principle in more quantitative terms
can be expressed using a modified version of the Beer-
Lambert Law. The attenuation due to absorption and scat-
tering effects may be described therefore as
A = log10
= αcLB +G.
Here A is attenuation, I0is incident light intensity, I is trans-
extinction coefficient for the absorber which is wavelength
4 Computational Intelligence and Neuroscience
dependent in this case, c is the concentration of the absorber,
L is the distance between the source and detector, B is the
differential path length factor, and G is a term to account for
Changes in haemoglobin levels are calculated then as a
superposition of the changes for each absorber species—in
this case oxy- and deoxyhaemoglobin:
Equation (2) is evaluated at two wavelengths, either side
of the isobestic point to enable separation of the two
Previous functional NIRS studies have documented
haemodynamic changes as a result of motor, cognitive, vi-
sual, and auditory activities . The device used here has
motor imagery in the sensorimotor cortex . The gen-
eral form is an increase in HbO coupled with a decrease in
Hb 3–5 seconds after the onset of movement execution or
imagery. While the idea of monitoring cerebral oxygenation
using NIRS has been around for some time, it has as yet
found only limited application in brain-computer interfac-
ing mostly due to the slow baud rate of the device. Currently,
two working devices have been reported [17, 18], however
the area is nascent and undoubtedly more in-depth investi-
gations of the efficacy of such devices will appear.
As an illustration of how CIMT might be augmented
the next section describes a simple practical embodiment in
which a NIRS-BCI is used to trigger movement of a robotic
actuator as a result of elevated motor cortical activity.
3. AN ILLUSTRATIVE EMBODIMENT
An example of how an embodiment of the concept described
in this paper might work is now given based on a syn-
chronous BCI paradigm in which the activation signal is de-
rived from bilateral cortical activity over the sensorimotor
region (SMR). Unlike most BCI experiments, however, overt
motor activity is employed by the subjects as imagined ac-
tivity is not required or indeed germane for the rehabilita-
tive therapy envisaged. Actual movement allows the exper-
imenter to determine that the motor areas must indeed be
active and hence eliminates the effect of poor engagement on
the part of the subjects in the results. A computer is used to
sual cues). Appropriate activation of the SMR during move-
ment triggers feedback in the form of movement of a robotic
The signals, collected simultaneously, are cerebral
changes in HbO and Hb, the respiration pneumogram and
the digital photoplethysmograph (PPG).
A continuous wave dual channel NIRS system is used to
interrogate the cerebral cortex on each hemisphere. The
light source comprises light emitting diodes (LEDs) at wave-
Figure 1: The Armdroid-1 robotic arm used in the feedback proto-
lengths of 760nm and 880nm (Opto Diode corp., Inc., APT-
0010/OD-880F, Calif, USA) having a narrow beam angle of
8◦and a spectral bandwidth at 50% of 30nm and 80nm, re-
spectively. The light output of each LED is modulated in the
low kilohertz range to facilitate lock-in detection at the out-
put. The LEDs are placed in direct contact with the scalp.
Avalanche photodiodes (APD), Hamamatsu C5460-01, were
used as the detector; the output of which was fed via a 3mm
diameter, 1m long, fibre optic bundle to lock-in amplifiers
(Signal Recovery, model 7265). A more detailed account of
the optical setup and other design considerations can be
found in .
For data acquisition (offline analysis), the Biopac
UIM100C interface module in tandem with a Biopac MP100
was used to collect the four analogue channels of NIRS
data (two wavelengths, two sites) from the lock-in ampli-
fiers at 16-bit resolution. In addition, two other analogue
channels of data were collected by the MP100 for respiration
and PPG (Biopac amplifiers models PPG100C and RSP100C)
with gains of 100 and 10, respectively. PPG100C settings
comprised a low-pass filter of 10Hz and high-pass filter of
0.05Hz. RSP100C settings implemented a low-pass filter of
Feedback was provided through movement of a robotic
arm (Figure 1) in sympathy with sustained elevation in HbO
levels during the motor execution tasks. The outputs of the
lock-in amplifiers was tapped to provide drive signals via
a simple 12-bit National Instruments USB-6008 DAQ at 10
samples per second. Online and real-time processings for Hb
and HbO using standard algorithms  based on (2) pro-
vided control of the robotic arm and feedback. The robotic
arm was driven using control signals from the DAQ system
allowing predetermined movement patterns to be invoked
when haemodynamic activity exceeded rest period levels.
In this work, we enlisted 5 healthy subjects (4 males, 1 fe-
male), 2 left handed and 3 right handed (determined, us-
Tomas E. Ward et al.5
Table 1: Optode locations referenced to EEG 10–20 system.
Channel 1 (left-hand side)
Channel 1 (right-hand side)
Light source location
C3 : 1.5(3π/2)
C4 : 1.5(3π/2)
C3 : 1.5(π/2)
C4 : 1.5(π/2)
Figure 2: Illustration of relative positioning of optode sources and
range was 23-25 years old (mean age 24years). One sub-
ject (female) was removed from analysis due to poor SNR
and optode placement problems. The remaining four sub-
jects underwent online feedback experiments. All remaining
had no previous experience with NIRS experiments.
Each subject was seated in a near supine position to re-
duce the effects of low-frequency blood oscillations (Mayer
wave) in a dimly lit room. The respiration monitoring device
(Biopac-TSD201) was strapped around the chest of each sub-
ject to monitor the respiratory signal during trials. The PPG
probe (Biopac-TSD200) was attached to the index finger on
the inactive limb to monitor the cardiac pulse during trials.
Subjects’ head measurements were taken to locate positions
C3 and C4. These 10–20 system positions are approximately
over primary motor cortex centres in the brain responsible
for right- and left-hand movements. The distance between
ing descriptor is available using the optode placement system
proposed in . Using this system, the optode location is
described in terms of distance and angle with respect to a
defined EEG 10–20 system landmark position which serves
as an origin. In this study, angles are referenced to Cz. This
yieldsoptodedescriptorsasinTable 1,illustratedinFigure 2.
detector to leave ample hair-free scalp. The optodes and fi-
bre optic bundles were inserted into cushioned pads in con-
tact with the subject’s scalp. The subject’s hands were placed
under restraining straps in order to facilitate isometric exer-
cise during the stimulus trials. Once positioned and instru-
mented, the subject was given instructions to follow, before
commencing the experiment. Prior to experiment each sub-
ject was informed about the nature and purpose of the ex-
perimental study and given precise instruction as to the task
required of them. To reduce artefact ,subjects were asked to
tions to breathe gently and regularly.
The paradigm for performing the overt motor task is
shown in Figure 3. An initial 30 seconds rest was followed
by alternating periods of 25 seconds of motor effort (isomet-
ric maximal voluntary contractions—MVCs of the indicated
forearm, pivoting at the elbow on a rigid support surface)
and 15-second rest. For each “experimental session,” there
were 10 stimulus periods. Each of the four subjects carried
out two sessions on each arm, thus a total of 20 stimulus pe-
riods per arm per subject. Thus, a total of 80 online trials for
each left and right arm are used in the final analysis.
Audio-visual cues indicating the task and rest periods
were presented via an LCD monitor to the subjects. Feed-
back was provided in two forms: a symbolic form which on
the LCD monitor presented itself as a change from a black
rectangle to an upwards pointing arrow when HbO levels in
excess of the previous rest period’s level were present, and
a physical action cue where movement of the robotic arm
took place under the same conditions. When the HbO lev-
els dipped below the threshold during the motor task period,
the icon reverted to the black rectangle form and motion of
the robotic arm ceased.
Raw signals from the lock-in amplifiers were sampled at
10Hz, and the Hb and HbO concentrations were calculated
in real-time, on a sample-by-sample basis. Simple moving
average filters were used in all experiments. A 10-point mov-
ing average filter was used to low-pass filter data in real time.
Once Hb and HbO concentrations were calculated, a further
moving average filter was used for classification. For the de-
tection of significant activity during the activation period,
a simple thresholding scheme was employed whereby a da-
tum was taken during the preceding rest period. This da-
tum consisted of the average HbO level during the 15 sec-
onds of the rest period. Neither Hb nor total haemoglobin
levels were used as an information signal in the online exper-
iments. The 10 point running average of the HbO signal cal-
period, significant motor cortical activity was inferred and
6 Computational Intelligence and Neuroscience
Table 2: Success rate in moving robot arm. Figures indicate the percentage of time subjects were able to keep the robot moving during each
trial. That is, subject 4 successfully moved the robot 96.7% of the time during all 10 stimulus trials for the first session of left-arm maximum
voluntary contraction (left 1).
∗Subject 3 experiments had low-light levels, thus a lower SNR. A previous X-ray has also shown that he has a relatively thick skull.
Left 1 (%)
Left 2 (%)
Right 1 (%)
Right 2 (%)
Subject average (%)
Visual and auditory
prompts for stimulus
Figure 3: Illustration of the experimental sequencing. Shaded boxes are motor task periods.
appropriate feedback was presented. In summary, activation
occurs where s[i] −r > 0,
HbO[i+ j] for i = 1,...,N, (3)
10, N is the number of samples acquired during the motor
task and r is the average HbO signal during the rest period.
So long as the stimulus moving average was greater than
the rest average, activity was sensed and the robotic arm was
Table 2 presents the results of the experiment as described.
This table shows the percentage of time that subjects were
able to move the robot during the motor activation task.
All the subjects were successful in achieving some control
of the robotic arm. For example, subject 1 was able to ac-
tivate the robot almost all the time when engaged in right
forearm movement (>95%). Subject 3 unfortunately was not
as successful as the others, only realising movement of the
robot arm just over 60% of the time (a footnote to the table
may suggest why). However, the measures presented here are
rather conservative as they indicate the percentage of time by
which the threshold was exceeded during the motor task. If
task periods where the robotic arm was activated, then the
results would be almost perfect. This of course would be a
disingenuous summary of the experiment for many reasons.
A more insightful observation of the experiments can be ob-
tained from Figure 4 which shows the averaged responses
tests during both the motor task and rest periods.
Figure 4 was produced using the Matlab-based NIRS
analysis tool HomER  and illustrates mean and standard
deviation levels that indicate consistent differences between
to order low-pass Butterworth filter with cutoff frequency of
4.POTENTIALS AND PROSPECTS
The key contribution of this paper is the presentation of the
idea that a neurocortical prosthesis may serve as a means to
extend CIMT to severe stroke sufferers as part of an ther-
apeutic regime. The very simple illustration of this idea in
Section 3 highlights well the basic operation of a NIRS-BCI
in a CIMT-like scenario. It is reasonable to suggest that even
the toy system above may provide a basic platform on which
to develop more sophisticated systems for comprehensive
studies with the intended population of stroke sufferers. The
results, which in themselves are nonsurprising in nature, are
useful for facilitating assessment of potential design issues
for more developed systems with the caveat that acquiring
good quality signals may be difficult with damaged cortex
and that even with robust signals, there is perhaps the pos-
sibility of habituation effects which may limit applicabil-
ity. Notwithstanding these concerns, the responses and ac-
tivation levels evident in Figure 4 show all the characteris-
tics expected [18, 42] of NIRS-BCI signals. While the results
show high variability, they have been calculated for real-time
biofeedback. This presents a significantly more difficult sce-
nario than offline analysis which would allow for removal
of artefact and screening of signals and would undoubt-
edly show improved figures. However, real-time control is
important for the application envisaged and the results do
Tomas E. Ward et al.7
−505 10 15 2025
05 101520 25
Figure 4: Top row shows average Hb (dashed trace) and HbO (solid trace) levels ± SD for subject 2. The bottom row shows average readings
for subject 1. The left-hand column shows activity during motor task (between vertical dashed lines) while the right-hand column shows
corresponding activity during rest. The abscissa for all plots is in seconds.
ing necessary crude, given the real-time requirements, is cer-
tainly worth more sustained development to compensate for
artefact. Online adaptive filtering is a necessary component
for a more robust system. Clearly, a better understanding
of the underlying responses may allow better integration of
other signals such as the Hb signal. The variability of the Hb
response meant that it was difficult to reliably use it as a trig-
ger signal and although discarded here, it is a useful signal to
collect for future work in improving performance. The idea
the provision of a multimodal neurocortical prothesis har-
nessing motor rhythm EEG would clearly enhance the sys-
tem further as it is probable that the compound signal would
offer greater sensitivity to weaker cortical activation and bet-
ter insight into neurological function . In addition, such
pling; the parameterisation of which may lead to greater in-
sight in the rehabilitative process. It is also worth consider-
ing if perhaps a motor-rhythm EEG BCI may work better as
a neurocortical prosthesis forthese applications independent
of any vascular response-oriented method. Only further re-
search will answer this.
The work reported in this paper clearly represents only
first steps towards extending CIMT to more severe motor
stroke patients and the authors would be first to admit that
there are very many questions unanswered which all merit
further exploration. One obvious question from the techno-
logical point of view taken here is whether or not the haemo-
dynamic signal required is as pronounced for sufferers of
stroke. In the case of cortical haemorrhagic stroke, for exam-
ple, the presence of scar tissue and haemotoma may absorb a
significant portion of the introduced near infrared light at-
tenuating the signal. Similarly constructed arguments may
be made for ischemic stroke; however in all cases, the sever-
8 Computational Intelligence and Neuroscience
ity of such effects if they occur at all clearly depends on the
site and extent of the injury. One might envisage that ini-
tial fMRI scans during attempted movement would facili-
tate the deployment of the optode configurations required
in such cases. To answer these concerns, clinical trials are
required with appropriately selected stroke patients. An in-
changes in the haemodynamic signals themselves along with
motor movement efficacy. Additional quantitative measures
such as those which might be provided through this method
would surely prove useful in measuring rehabilitative out-
come. The pioneers in this area have noted in a recent paper
that NIRS monitoring may provide a technological break-
through in terms of developing and understanding CIMT
. Techniques such as the one espoused here may make
section, the neural prosthesis advocated here is not intended
as a permanent replacement for the patient’s own nervous
system. It is envisaged that the device serve as a temporary
channel to convey some feedback for stroke sufferers who
have none. As soon as any other more conventional feedback
is available, then the prosthesis may be discarded. This phi-
use of stereotyped brain signals metabolic in origin or oth-
erwise could within this disease context produce unwanted
plasticity phenomena such as tics, obsessive thoughts, and
other aberrant neurological functioning .
This paper has highlighted the possibility of enhancing the
application of CIMT for stroke sufferers through the addi-
tion of a neuro-cortical prosthesis. Generally it is proposed
that a fruitful avenue for new research in the application of
brain computer interfaces is in their measurement of voli-
tional motor effort for biofeedback purposes in CIMT. The
little or no perceptible movement although the idea may
have utility in the broader stroke population. More specifi-
cally NIRS-based BCI are proposed as suitable candidates for
such purposes. A simplified illustration of such a system is
presented which demonstrates the basic feasibility of the ap-
proach. Testing with actual stroke sufferers is clearly the next
nificant challenges. Nevertheless we believe the concept de-
scribed inthispaperhasmeritasaspecificextension ofbrain
computer interfaces into the neurorehabilitation domain.
This work is supported by Science Foundation Ireland Grant
no. SFI/05/RFP/ENG0089. The authors would like to ac-
knowledge the contributions of Professor Barak P. Pearlmut-
ter and Dr. Ray O’Neill in discussions concerning this work.
Dr. Ward would also like to thank the reviewers and editors
for their comments, advice, and suggestions during the re-
 R. Adams, M. Victor, and A. Ropper, Adam’s & Victor’s Princi-
ples of Neurology, McGraw-Hill, New York, NY, USA, 6th edi-
 World Health Organization Global Report 2002, “World
Health Organization,” http://www.who.int/whr/2002/en/index
.html, January 2007.
 The Internet Stroke Centre, http://www.strokecenter.org/pat/
stats.htm, January 2007.
 R. W. Bohannon, “Muscle strength and muscle training after
stroke,” Journal of Rehabilitation Medicine, vol. 39, no. 1, pp.
 E. Taub, N. E. Miller, T. A. Novack, et al., “Technique to im-
prove chronic motor deficit after stroke,” Archives of Physical
Medicine and Rehabilitation, vol. 74, no. 4, pp. 347–354, 1993.
 E. Taub, G. Uswatte, and R. Pidikiti, “Constraint-induced
movement therapy: a new family of techniques with broad ap-
plication to physical rehabilitation—a clinical review,” Journal
of Rehabilitation Research and Development, vol. 36, no. 3, pp.
 W. H. Miltner, H. Bauder, M. Sommer, C. Dettmers, and E.
Taub, “Effects of constraint-induced movement therapy on
Stroke, vol. 30, no. 3, pp. 586–592, 1999.
 J. Liepert, “Motor cortex excitability in stroke before and after
constraint-induced movement therapy,” Cognitive and Behav-
ioral Neurology, vol. 19, no. 1, pp. 41–47, 2006.
 V. W. Mark, E. Taub, and D. M. Morris, “Neuroplasticity and
constraint-induced movement therapy,” Europa Medicophys-
ica, vol. 42, no. 3, pp. 269–284, 2006.
 S. L. Wolf, C. J. Winstein, J. P. Miller, et al., “Effect of
constraint-induced movement therapy on upper extremity
function 3 to 9 months after stroke: the EXCITE random-
ized clinical trial,” Journal of the American Medical Association,
vol. 296, no. 17, pp. 2095–2104, 2006.
apy for chronic stroke hemiparesis and other disabilities,”
Restorative Neurology and Neuroscience, vol. 22, no. 3-4, pp.
 P. S. Lum, C. G. Burgar, D. E. Kenney, and H. F. M. Van der
Loos, “Quantification of force abnormalities during passive
and active-assisted upper-limb reaching movements in post-
stroke hemiparesis,” IEEE Transactions on Biomedical Engi-
neering, vol. 46, no. 6, pp. 652–662, 1999.
 J. H. van der Lee, R. C. Wagenaar, G. J. Lankhorst, T. W. Vo-
gelaar, W. L. Devill´ e, and L. M. Bouter, “Forced use of the up-
per extremity in chronic stroke patients: results from a single-
blind randomized clinical trial,” Stroke, vol. 30, no. 11, pp.
 J. Chae, F. Bethoux, T. Bohinc, L. Dobos, T. Davis, and
A. Friedl, “Neuromuscular stimulation for upper extremity
motor and functional recovery in acute hemiplegia,” Stroke,
vol. 29, no. 5, pp. 975–979, 1998.
 J. Powell, A. D. Pandyan, M. Granat, M. Cameron, and D. J.
Stott, “Electrical stimulation of wrist extensors in poststroke
hemiplegia,” Stroke, vol. 30, no. 7, pp. 1384–1389, 1999.
 A. J. Butler and S. J. Page, “Mental practice with motor im-
after stroke,” Archives of Physical Medicine and Rehabilitation,
vol. 87, no. 12, supplement, pp. 2–11, 2006.
Tomas E. Ward et al.9
 S. Coyle, T. E. Ward, C. Markham, and G. McDarby, “On
the suitability of near-infrared (NIR) systems for next-
generation brain-computer interfaces,” Physiological Measure-
ment, vol. 25, no. 4, pp. 815–822, 2004.
 R. Sitaram, H. Zhang, C. Guan, et al., “Temporal classification
of multichannel near-infrared spectroscopy signals of motor
age, vol. 34, no. 4, pp. 1416–1427, 2007.
 S. Coyle, T. E. Ward, and C. Markham, “An optical brain com-
puter interface,” in Proceedings of the 2nd International Brain-
Computer Interface Workshop and Training Course, vol. 49 of
Biomedizinische Technik, pp. 45–46, Graz, Austria, 2004.
 J. W. Krakauer, “Motor learning: its relevance to stroke recov-
ery and neurorehabilitation,” Current Opinion in Neurology,
vol. 19, no. 1, pp. 84–90, 2006.
apy: answers and questions after two decades of research,”
NeuroRehabilitation, vol. 21, no. 2, pp. 93–95, 2006.
 H. I. Krebs, N. Hogan, M. L. Aisen, and B. T. Volpe, “Robot-
aided neuro-rehabilitation,” IEEE Transactions on Rehabilita-
tion Engineering, vol. 6, no. 1, pp. 75–87, 1998.
 E. J. Plautz, G. W. Milliken, and R. J. Nudo, “Effects of repet-
itive motor training on movement representations in adult
squirrel monkeys: role of use versus learning,” Neurobiology of
Learning and Memory, vol. 74, no. 1, pp. 27–55, 2000.
 J. H. van der Lee, R. C. Wagenaar, G. J. Lankhorst, T. W. Vo-
gelaar, W. L. Devill´ e, and L. M. Bouter, “Forced use of the up-
per extremity in chronic stroke patients: results from a single-
blind randomized clinical trial,” Stroke, vol. 30, no. 11, pp.
 B. T. Volpe, H. I. Krebs, N. Hogan, L. Edelsteinn, C. M. Diels,
and M. L. Aisen, “Robot training enhanced motor outcome
in patients with stroke maintained over 3 years,” Neurology,
vol. 53, no. 8, pp. 1874–1876, 1999.
T. Volpe, “Robot-aided sensorimotor arm training improves
outcome in patients with chronic stroke,” Neurology, vol. 61,
no. 11, pp. 1604–1607, 2003.
 R. Colombo, F. Pisano, S. Micera, et al., “Robotic techniques
for upper limb evaluation and rehabilitation of stroke pa-
Engineering, vol. 13, no. 3, pp. 311–324, 2005.
 B. T. Volpe, M. Ferraro, D. Lynch, et al., “Robotics and other
devices in the treatment of patients recovering from stroke,”
Current Neurology and Neuroscience Reports, vol. 5, no. 6, pp.
 M. J. Johnson, “Recent trends in robot-assisted therapy en-
vironments to improve real-life functional performance after
stroke,” Journal of NeuroEngineering and Rehabilitation, vol. 3,
no. 1, p. 29, 2006.
 L. Dipietro, M. Ferraro, J. J. Palazzolo, H. I. Krebs, B. T. Volpe,
and N. Hogan, “Customized interactive robotic treatment for
stroke: EMG-triggered therapy,” IEEE Transactions on Neural
Systems and Rehabilitation Engineering, vol. 13, no. 3, pp. 325–
 A. K¨ ubler, V. K. Mushahwar, L. R. Hochberg, and J. P.
Donoghue, “BCI meeting 2005-workshop on clinical issues
and applications,” IEEE Transactions on Neural Systems and
Rehabilitation Engineering, vol. 14, no. 2, pp. 131–134, 2006.
 B. H. Dobkin, “Brain-computer interface technology as a tool
to augment plasticity and outcomes for neurological rehabili-
tation,” The Journal of Physiology, vol. 579, no. 3, pp. 637–642,
 G.Pfurtscheller,G.R.M¨ uller,J.Pfurtscheller,H.J.Gerner,and
tion to restore hand grasp in a patient with tetraplegia,” Neu-
roscience Letters, vol. 351, no. 1, pp. 33–36, 2003.
 A. Buttfield, P. W. Ferrez, and J. D. R. Mill´ an, “Towards a ro-
bust BCI: error potentials and online learning,” IEEE Transac-
no. 2, pp. 164–168, 2006.
 Y. Wang, R. Wang, X. Gao, B. Hong, and S. Gao, “A practical
VEP-based brain-computer interface,” IEEE Transactions on
Neural Systems and Rehabilitation Engineering, vol. 14, no. 2,
pp. 234–240, 2006.
 D. J. McFarland, D. J. Krusienski, and J. R. Wolpaw, “Brain-
computer interface signal processing at the Wadsworth Cen-
ter: mu and sensorimotor beta rhythms,” Progress in Brain Re-
search, vol. 159, pp. 411–419, 2006.
 E. W. Sellers, D. J. Krusienski, D. J. McFarland, T. M. Vaughan,
and J. R. Wolpaw, “A P300 event-related potential brain-
computer interface (BCI): the effects of matrix size and in-
ter stimulus interval on performance,” Biological Psychology,
vol. 73, no. 3, pp. 242–252, 2006.
 C. Vidaurre, A. Schl¨ ogl, R. Cabeza, R. Scherer, and G.
on Biomedical Engineering, vol. 53, no. 6, pp. 1214–1219, 2006.
 N. Birbaumer and L. G. Cohen, “Brain-computer interfaces:
communication and restoration of movement in paralysis,”
The Journal of Physiology, vol. 579, no. 3, pp. 621–636, 2007.
 F. Hummel, P. Celnik, P. Giraux, et al., “Effects of non-invasive
cortical stimulation on skilled motor function in chronic
stroke,” Brain, vol. 128, no. 3, pp. 490–499, 2005.
 P. Rolfe, “In vivo near-infrared spectroscopy,” Annual Review
of Biomedical Engineering, vol. 2, pp. 715–754, 2000.
 S. Coyle, Near-infrared spectroscopy for brain computer inter-
facing, Ph.D. thesis, National University of Ireland, Maynooth,
 M. Cope and D. T. Delpy, “System for long-term measure-
ment of cerebral blood and tissue oxygenation on newborn
infants by near infra-red transillumination,” Medical and Bio-
logical Engineering and Computing, vol. 26, no. 3, pp. 289–294,
 R. C. Oldfield, “The assessment and analysis of handedness:
the Edinburgh inventory,” Neuropsychologia, vol. 9, no. 1, pp.
 S. Coyle, C. Markham, W. Lanigan, and T. Ward, “A me-
chanical mounting system for functional near-infrared spec-
troscopy brain imaging studies,” in Opto-Ireland 2005: Optical
Sensing and Spectroscopy, vol. 5826 of Proceedings of SPIE, pp.
618–627, Dublin, Ireland, April 2005.
 HomER, http://www.nmr.mgh.harvard.edu/PMI/resources/
tion of EEG, MEG and fMRI data for the solution of the neu-
roimage puzzle,” Magnetic Resonance Imaging, vol. 22, no. 10,
pp. 1471–1476, 2004.