Content uploaded by Andrea Turolla
Author content
All content in this area was uploaded by Andrea Turolla
Content may be subject to copyright.
Assessment and Treatment of the Upper
Limb by Means of Virtual Reality in Post-
Stroke Patients
Lamberto PIRON
a
, Andrea TUROLLA
a
, Michela AGOSTINI
a
, Carla ZUCCONI
a
,
Paolo TONIN
a
, Francesco PICCIONE
a
and Mauro DAM
b
a
I.R.C.C.S. San Camillo Hospital, Venice, Italy
b
Department of Neuroscience, University of Padua, Padua, Italy
Abstract. The disability deriving from stroke impacts heavily on the economic
and social aspects of western countries because stroke survivors commonly
experience various degrees of autonomy reduction in the activities of daily living.
Recent developments in neuroscience, neurophysiology and computational science
have led to innovative theories about the brain mechanisms of the motor system.
Thereafter, innovative, scientifically based therapeutic strategies have initially
arisen in the rehabilitation field. Promising results from the application of a virtual
reality based technique for arm rehabilitation are reported.
Keywords. Stroke, Rehabilitation, Motor Learning and Control, Augmented
Feedback, Virtual Reality
Introduction
Stroke is a leading cause of death and disability for men and women of all ages, classes,
and ethnic origins worldwide. Several epidemiological surveys were conducted on
cerebro-vascular disease, especially in the United States, where 500,000 new strokes
occur each year causing 100,000 deaths and leaving residual disability for 300,000
survivors. Moreover, approximately 3 million Americans have survived a stroke with
some degree of residual disability [1, 3].
Within 2 weeks after stroke, hemiparesis is present in 70-85% of patients and a
percentage, between 40 to 75%, is completely dependent in their activities of daily
living [4]. There is a lack of epidemiological data for European countries, although in
the United Kingdom the Oxfordshire Community Stroke Project (1983) reported an
annual incidence of 500 new cases in a 250,000 people community, with a peak in
people older than 75 years [5]. In a recent study conducted in Norway, a total annual
incidence of 2.21 strokes per 1000 people was reported. This rate is congruent with
other European countries showing that there are no regional variations within Western
Europe [6].
The estimates of the total cost of stroke are very variable in relation to the
difficulty of calculating the indirect cost resulting from disability and mortality. A 1993
estimate placed the total annual cost of stroke at $30 billion in the United States, of
which $17 billion are direct costs (hospital, physician, rehabilitation, equipment) and
$13 billion are indirect costs (lost productivity) [7].
Support us – Ask your institution/library/department to order our books or to apply for series
membership - ISSN 1566-7677
To order a book
contact by fax or
by e-mail the offices
below:
IOS Press, Inc.
4502 Rachael Manor drive
Fairfax, VA 22032
U.S.A.
Tel.: +1 703 323 5600
Fax: +1 703 323 3668
sales@iospress.com
IOS Press
Nieuwe Hemweg 6B
1013 BG Amsterdam
The Netherlands
Tel: +31 20 688 3355
Fax: +31 20 620 3419
info@iospress.nl
IOS Press,
c/o Ohmsha, Ltd.
3-1 Kanda Nishiki-cho
Chiyoda-ku
Tokyo 101
Japan
Fax: +81 3 3233 2426
(Books only)
IOS Press/
Gazelle Book Services Ltd
White Cross Mills
Hightown
Lancaster LA1 4XS
United Kingdom
Tel.: +44 1524 68765
Fax: +44 1524 63232
sales@gazellebooks.co.uk
Advanced Technology in Rehabilitation
Empowering Cognitive, Physical, Social and Communicative Skills through Virtual Reality, Robots,
Wearable Systems and Brain-Computer Interfaces
Volume 145 Studies in Health Technology and Informatics
Edited by: Gaggioli, A., Keshner, E.A., Weiss, P.L., Riva, G.
June 2009, approx. 310 pp., hardcover
ISBN: 978-1-60750-018-6 NEW
Price: US$167 / €115 / £81
Enacting Intersubjectivity
A Cognitive and Social Perspective on the Study of Interactions
Volume 10 Emerging Communication: Studies on New Technologies and Practices in Communication
Edited by: F. Morganti, A. Carassa and G. Riva
May 2008, approx. 280 pp., hardcover
ISBN: 978-1-58603-850-2 NEW
Price: US$161 / €115 / £81
From Communication to Presence
Cognition, Emotions and Culture towards the Ultimate Communicative Experience
Festschrift in honor of Luigi Anolli
Volume 9 Emerging Communication: Studies on New Technologies and Practices in Communication
Edited by: G. Riva, M.T. Anguera, B.K. Wiederhold, F. Mantovani
September 2006, 323 pp., hardcover
ISBN: 978-1-58603-662-1
Price: US$161 / €115 / £81
Global Data Management
Volume 8 Emerging Communication: Studies on New Technologies and Practices in Communication
Edited by: R. Baldoni, G. Cortese, F. Davide and A. Melpignano
July 2006, 376 pp., hardcover
ISBN: 1-58603-629-7
Price: US$161 / €115 / £81
The Hidden Structure of Interaction
From Neurons to Culture Patterns
Volume 7 Emerging Communication: Studies on New Technologies and Practices in Communication
Edited by: L. Anolli, G. Riva, S. Duncan Jr. and M.S. Magnusson
May 2005, 304 pp., hardcover
ISBN: 1-58603-509-6
Price: US$161 / €115 / £81
Ambient Intelligence
The Evolution of Technology, Communication and Cognition Towards the Future of Human-Computer
Interaction
Volume 6 Emerging Communication: Studies on New Technologies and Practices in Communication
Edited by: G. Riva, F. Vatalaro, F. Davide and M. Alcañiz
January 2005, 316 pp., hardcover
ISBN: 1-58603-490-1
Price: US$161 / €115 / £81
Being There
Concepts, Effects and Measurements of User Presence in Synthetic Environments
Volume 5 Emerging Communication: Studies on New Technologies and Practices in Communication
Edited by: G. Riva, F. Davide and W.A. IJsselsteijn
2003, 344 pp., hardcover
ISBN: 1-58603-301-8
Price: US$161 / €115 / £81
Say not to Say: New Perspectives on Miscommunication
Volume 3 Emerging Communication: Studies on New Technologies and Practices in Communication
Edited by: L. Anolli, R. Ciceri and G. Riva
2001, 288 pp., hardcover
ISBN: 1-58603-215-1
Price: US$161 / €115 / £81
Towards CyberPsychology
Mind, Cognition and Society in the Internet Age
Volume 2 Emerging Communication: Studies on New Technologies and Practices in Communication
Edited by: G. Riva and C. Galimberti
2001, 326 pp., hardcover
ISBN: 1-58603-197-x
Price: US$161 / €115 / £81
Communications Through Virtual Technologies
Identity, Community and Technology in the Communication Age
Volume 1 Emerging Communication: Studies on New Technologies and Practices in Communication
Edited by: G. Riva and F. Davide
2001, 292 pp., hardcover
ISBN: 1-58603-162-7
Price: US$161 / €115 / £81
The main cost of stroke survivors is related to their residual motor disabilities that
interfere with personal, social and/or productive activities. Surprisingly, there are few
therapeutic approaches to restore lost functions. Nowadays the available rehabilitative
therapies are currently working to develop treatments that are closely related to motor
learning principles.
Recently the development of tools for quantitative analysis of motor deficits gave
the opportunity to increment the amount of data in clinical practice to better study
human motor behavior with consequent important practical implications. First of all, it
will be possible to infer the anatomical structures that modulate the different elements
of motor control. Furthermore, it may help to better characterize motor deficits and, as
a consequence, to plan individually modified therapeutic approaches. Finally, the
quantitative analysis of movement may allow to monitor pharmacological therapies (i.e
drugs interacting with the central neurotransmitters levels) that could modify the
human motor behavior [8, 9].
1. Rationale
1.1 Neurophysiology of motor learning
Research on the physiological underpinnings of movement dynamics has traditionally
focused most extensively on the primary motor cortex (M1) pointing out that neurons
in M1 are modulated by external dynamic perturbations. Some investigators [10]
indicate that several premotor areas feed M1 which then projects to the spinal cord.
These areas are intensely interconnected with each other, with a parallel contribution to
the control of movement [11].
Other work on primates demonstrated that several cortical cells in motor and pre-
motor areas responded selectively to kinematic variations during motor adaptation
tasks. These cells, clearly identified in the monkey SMA, are involved in the process of
kinematics-to-dynamics transformation, hence in new motor task learning [11]. Doya et
al. [11, 12] proposed other correlations between the motor learning problem and
circuitry at the cortical level, suggesting that different brain areas are involved in three
different kinds of learning mechanisms: supervised learning, reinforcement learning
and unsupervised learning.
The cerebellum is supposed to be involved in real-time fine tuning of movement
by means of its feed-forward structure based on massive synaptic convergence of
granule cell axons (parallel fibers) onto Purkinje cells, which send inhibitory
connections to deep cerebellar nuclei and to inferior olive. The circuit of the cerebellum
is capable of implementing the supervised learning paradigm which consists of error
driven learning behaviors. Reinforcement learning is based on the multiple inhibitory
pathways of the basal ganglia that permit the reward predicting activity of dopamine
neurons and change of behavior in the course of goal directed task learning. The
extremely complex anatomical features of the cortex suggest that information coding is
established by an unsupervised learning paradigm in which the activity is determined
by the Hebbian-rule of synaptic updating. In this paradigm the environment provides
input but gives neither desired targets nor any measure of reward or punishment [11,
12]. Recent neurophysiologic studies demonstrated that some natural complex systems
have discrete combinatory architecture that utilizes finite numbers of primary elements
to create larger structures, such as motor primitives in spinal cord [13]. Poggio and
Bizzi [14] hypothesized a hierarchical architecture where the motor cortex is endowed
with functional modular structures that change their directional tuning during
adaptation, visuo-motor learning, exposure to mechanical load and reorganization after
lesions, i.e. the circuit of interneurons as central pattern generators, unit burst
generators, and spinal motor primitives contributing to motor learning. In the latter case
the force fields stored as synaptic weights in spinal cord may be viewed as representing
motor field primitives from which, through linear superimposition, a vast number of
movements can be fashioned by impulses conveyed by supraspinal and reflex pathways
[14]. Computational analysis [15] verifies that this proposed mechanism is capable of
learning and controlling a wide repertoire of motor behaviors. This hypothesis suggests
that the cortical lesion induced by a stroke could modify the hierarchical architecture
with negative influences on learning and controlling new motor behaviors.
1.2 Neurophysiopathology of stroke lesion
From the physiopathologic perspective, much evidence demonstrated that the location
of the stroke lesion is related to upper limb motor deficit severity. Specifically it is
argued that patients with cortical stroke have a better motor outcome than patients with
subcortical stroke. Furthermore, patients with mixed cortical plus subcortical stroke
tended to improve more than patients with pure subcortical stroke despite the expected
larger size of mixed lesions. Although subcortical strokes are normally smaller than
cortical strokes, they are more likely to involve primary (from M1) and secondary
motor pathways (from SMA and premotor area, PMA). The descending fibers from
primary and secondary motor areas converge in the internal capsule maintaining their
somatotopic distribution. Consequently, even small subcortical lesions produce
devastating motor effects. The probability of upper limb motor recovery after stroke is
hence linked strictly with the anatomical lesion: 75% for patients with lesions restricted
to the cortex (MI, PMA, SMA); 38.5% for those with subcortical or mixed cortical plus
subcortical lesions not affecting the posterior limb of the internal capsule (PLIC); and
3.6% for those with involvement of the PLIC plus adjacent corona radiata, basal
ganglia or thalamus [16].
1.3 Computational approach to the upper limb rehabilitation.
The computational approach to the motor system is a powerful analysis, in the field of
neuroscience, which offers the opportunity to unify the experimental data in a
theoretical framework. In the computational perspective, the motor behavior is intended
as the manifestation of an engineering system, whose basic task is to manage the
relationships between motor commands and sensory feedback. This management is
necessary for two reasons:
1. it ensures that our movements achieve their goals;
2. it enables us to learn by experience to make more accurate and effective
movements.
Recently, Han et al. developed a computational model for bilateral hand use in arm
reaching movements to study the interactions between adaptive decision making and
motor learning after motor cortex lesion [17]. This model combines a biologically
plausible neural model of the motor cortex with a non-neural model of reward-based
decision making and physical therapy intervention. The model demonstrated that in the
damaged cortex, during therapy, the supervised learning rules ensured that
underrepresented directions of movement were “repopulated”, thereby decreasing
average reaching errors.
The authors suggested that after stroke, if no therapy is given, plasticity due to
unsupervised learning may become maladaptive, thereby augmenting the stroke’s
negative effect. They also indicated that there is a threshold for the amount of therapy
based on three types of learning mechanisms (unsupervised, supervised and
reinforcement) required for the recovery process; below this threshold motor retraining
is “in vain”. In other words, there is an absent or exiguous use of the arm exhibiting the
“learned non-use” phenomenon. In the absence of supervised or reinforcement learning,
subsequent motor performance worsens with any amount of rehabilitation trials. On the
contrary, if unsupervised learning is not present, motor performance improves with any
amount of rehabilitation trials in the late period.
1.4 Virtual reality as an emerging therapy
Virtual Reality (VR) is an innovative technology consisting of a computer based
environment that represents a 3-D artificial world. VR has been already applied in
many fields of human activity. New computer platforms permit human-machine
interactions in real time, therefore the possibility of using VR in medicine has arisen.
The present level of technical advances in the computer interface allows the
development of VR systems as therapeutic tools in some neurological and psychiatric
pathology. For example, stroke survivors may undergo rehabilitative therapeutic
procedures with different VR systems [18, 19]. The use of a VR-based system coupled
to a motion tracking tool allows us to study the kinematics of arm movement in the
restorative process after stroke. Furthermore, the possibility of modifying the artificial
environment, where the patients could interact, may exploit some of the mechanisms of
motor learning.
We know from physiological studies that humans perform a large variety of
constrained and unconstrained movement in a smooth and graceful way because the
CNS enables us to rapidly solve complex computational problems. One hypothesis is
that the CNS needs little information in order to adapt movements to the changing of
the external requirements, providing that it already contains preprogrammed algorithms
for function [20]. These algorithms produce regularities in biological movements that
are not in any way implied by the motor task. Accordingly with this view, a given
movement can be characterized by variant and invariant elements. For instance, the
variant part of a reaching movement is the distance of the targets (corresponding to the
amplitude of the movement). The invariant part consists of straight paths with a bell-
shaped speed profile in all movements [21, 22].
In our laboratory, we experimented with a VR based setting for the assessment and
treatment of arm motor deficit in patients after stroke. We compared a VR based
(reinforced feedback in virtual environment, RFVE) and traditional physical therapy
technique (conventional therapy, CT) in the treatment of arm motor impairments in
post-stroke patients. The studied population met the following inclusion criteria: a
single ischemic stroke in the region of the middle cerebral artery at least six months
before the study (proven by means of CT scan or MRI); conventional physical therapy
treatment received in the early period after stroke; mild to intermediate motor
impairments of the arm assessed as a Fugl-Meyer Upper Extremity score (F-M UE)
between 20 and 60, at baseline [23]. Clinical history or evidence of memory
impairments, neglect, and apraxia or aphasia interfering with verbal comprehension
were all considered exclusion criteria.
The experimental intervention was the RFVE treatment and the control procedure
consisted of conventional physical therapy treatment. Both therapies were oriented
towards upper extremity motor rehabilitation. In the first one, the subject was requested
to perform different kinds of motor tasks while the movement of the entire
biomechanical arm system’s end-effector was simultaneously represented in a virtual
scenario by means of motion-tracking equipment. The equipment included a computer
workstation connected to a 3D motion-tracking system (Polhemus 3Space FasTrak,
Vermont, U.S.A) and a high-resolution LCD projector which displayed the virtual
scenarios on a large wall screen. The electromagnetic 3D motion-tracking sensor was
positioned on a manipulable object (rubber ball, polystyrene cube etc.) held by the
subject, or, alternatively, was attached to a glove worn by the patient in cases of severe
grasping deficits. The physical therapist could create numerous virtual motor tasks for
the arm through the use of flexible software, developed at the Massachusetts Institute
of Technology (Cambridge, MA, U.S.), which processes the motion data coming from
the end-effector receiver. The therapist selected the characteristics and the complexity
of the motor tasks in order to suit each patient’s arm deficit. In the virtual scenario, the
therapist determined the starting position and the characteristics of the target, such as
target orientation, for each task or the addition of other virtual objects to increase the
task’s complexity. A simple reaching movement could accomplish some tasks, while
others required more complicated movements, such as putting the envelope in the
mailbox, hitting the nail, or pouring the glass in the carafe. The subject moves the real
envelope, hammer, or glass and sees on the screen the trajectory of the corresponding
virtual object toward the virtual mailbox, nail, or carafe.
During the RFVE therapy, patients were asked to perform motor tasks according to
constraints specified beforehand by the therapist. Subjects were given information
about their arm movements during the performance of motor skills (knowledge of
performance, KP) by the movement of the end-effector’s virtual representation. The
therapist’s movement and trajectory could also be displayed in the background of the
virtual scene in order to facilitate the subject’s perception and adjustment to motion
errors (learning by imitation) [24]. Moreover, knowledge of the results (KR) regarding
motor task correctness was supplied to patients in the form of standardized scores and
by displaying arm trajectory morphology on the screen. Initially, the above mentioned
KP and KR were provided at a frequency of more than 90% and were gradually
decreased as performance improved.
In the CT group the subjects were asked to perform specific exercises for the upper
limb with a strategy of progressive complexity. First, the patients were requested to
control isolated motions without postural control, with physical therapist support if
necessary, then postural control was included and, finally, complex motion with
postural control was practiced. For example, patients were asked to touch different
targets arranged upon a horizontal plane in front of them; to manipulate different
objects; to follow trajectories displayed on a plane; to recognize different arm positions.
The physical therapists chose the exercises in relation to functional assessments
and patient needs.
The aim of this study was to compare the RFVE and CT approaches towards the
treatment of arm motor impairments in post-stroke patients. We hypothesized that a
rehabilitation technique based on motor learning rules, specifically the kinematic
information about arm movements in a virtual environment, could significantly
improve the motor outcome scores better than CT therapy. Before and after the
treatment, the degree of motor impairment and independence in daily living activities
were evaluated in both groups with the F-M UE score and the Functional Independence
Measure scale (FIM) [25]. At the same evaluation times, for all of the patients, we
determined the mean duration (MD) in sec, mean linear velocity (MLV) in cm/sec, and
the number of sub-movements (SM) in 36 motor trials organized into four tasks.
The patients’ starting position was the same in all of the trials. The different
orientation of the target (horizontal, vertical and diagonal on the subject’s frontal plane)
determined the complexity of the movement in terms of involving the activation of
different muscles. The patients were randomly assigned into the 2 groups and both
groups underwent the therapy for 1 hour treatment sessions daily, 5 days a week for 4
weeks.
Analyzing clinical (F-M UE and FIM) variables, we found, in both groups, a
statistical significance within the groups for the F-M UE (p-values<0.00, p-
values<0.016, respectively) and for the FIM (p-values<0.00, p-values<0.009,
respectively) scales. The robust regression analysis revealed that the F-M UE values
after the treatment were systematically higher in the RFVE patients than in the CT
subjects (β =-4.26, p-value<0.005). We observed the same result also for the FIM
values after the treatment (β =-4.59, p-value<0.02). The kinematic (MD, MLV, SM)
parameters changed significantly after the treatment only in the experimental group (p-
value = 0.01, 0.00 and 0.02 respectively), in contrast to those of the control subjects (p-
value = 0.18, 0.11 and 0.15 respectively). Finally none of patients who underwent the
RFVE therapy complained of any discomforts due to interaction with the virtual world,
such as cybersickness, altered eye-motor coordination or postural disequilibrium,
thereby demonstrating that this VR-application is safe for neurological patients.
Our results confirm that late therapy may improve motor performance as suggested
in others studies using different rehabilitation techniques [26, 27]. The kinematic
results were coherent with the RFVE rationale based on the amplification of kinematic
feedback to promote motor recovery; furthermore the improvement in motor
performance occurred concurrently with kinematic variation. In our opinion, the higher
results achieved with RFVE treatment were connected with the rationale of the VR
based technique which exploits the motor learning mechanisms.
2. Conclusion
In our VR setting, patients were given information about their arm movements during
the performance of motor skills (KP) that consisted of the representation of their end-
effector, and “virtual teacher” movement which showed the actual kinematics of the
hand path in order to practice “learning by imitation”. The teacher, as other relevant
feedback, realizes an ideal environment to implement new predictors or to modify
disrupted forward models. These mechanisms are developed by means of amplification
of the actual state. On the other side, new or better controllers can be developed by
means of different sensorimotor context presented in every scenario, as by the
utilization of graphic models that reproduce the visual objects’ appearance, giving
coherent contextual information. Furthermore, instructions imparted by the therapist
during the experimental procedure and the virtual representation of the correct
movement contributed to providing information about motor performance, thereby
exploiting so-called “supervised learning”. Moreover, the object’s trajectories
displayed on-screen allowed patients to evaluate the accuracy of their movement (KR),
thereby promoting the identification of successful motor strategies through the “trial
and error” paradigm. A second kind of KR provided to patients was a reward delivered
when the task performance score surpassed a pre-established threshold. These two
phenomena contributed to generating the basis for the “reinforcement learning”
mechanism.
In our experience, the synergistic activity of supervised, reinforcement and
learning by imitation facilitates faster development of the kinematic internal models
essential for motor learning. The opportunity for supplying patients with a
measurement of motor performance generated an auto-competitive stimulus for
progressively improving the correctness of arm trajectories session by session. The
above aspect, combined with the novelty and the originality of the VR-based therapy,
motivated the patients to enthusiastically participate in the rehabilitation sessions.
References
[1] J.H. Chesebro, V. Fuster, J.L. Halperin, Atrial fibrillation-Risk marker for stroke, The New England
Journal of Medicine 323 (1990), 1556-1558.
[2] R.D. Abbott, Y. Yin, D.M. Reed, et al., Risk of stroke in male cigarette smokers, The New England
Journal of Medicine 315 (1986), 717-720.
[3] M.L. Olsen, Autoimmune disease and stroke, Stroke (4) (1992), 13-16.
[4] B. Dobkin, Neurologic rehabilitation, Contemporary Neurology Series, 1995.
[5] Oxfordshire Community Stroke Project, Incidence of stroke in Oxfordshire : first year of experience of
a community stroke register, British Medical Journal 287 (1983), 713-717.
[6] H Ellekjaer, J Holmen, B Indredavik, and A Terent, Epidemiology of stroke in Innherred, Norway,
1994 to 996. Incidence and 30-day case-fatality rate, Stroke (1997), 2180-2184.
[7] PORT Study, Duke University Medical Center, Durham, NC, 1994.
[8] R.W.V. Flynn, R.S.M. MacWalter, A.S.F. Doney, The cost of cerebral ischaemia, Neuropharmacology
55 (3) (2008), 250-266.
[9] D.M. Feeney, A.M. De Smet, S. Rai, Noradrenergic modulation of hemiplegia: facilitation and
maintenance of recovery, Restorative Neurology and Neuroscience 22 (2004), 175-190.
[10] L.B. Goldstein, Neurotransmitters and motor activity: effects on functional recovery after brain injury,
NeuroRx 3 (2006), 451-457.
[11] C. Padoa-Schioppa, Li Chiang-Shan Ray, and E. Bizzi, Neuronal activity in the supplementary motor
area of monkeys adapting to a new dynamic environment, Journal of Neurophysiology 91 (2004), 449-
473.
[12] K. Doya, What are the computations of the cerebellum, the basal ganglia and the cerebral cortex?,
Neural networks 12 (1999), 961-974.
[13] K. Doya, Complementary roles of basal ganglia and cerebellum in learning and motor control, Current
Opinion in Neurobiology 10 (6) (2000), 732-739.
[14] E. Bizzi, F.A. Mussa-Ivaldi, S. Giszter, Computations underlying the execution of movement: a
biological perspective, Science 253 (1991), 287-291.
[15] T. Poggio, E. Bizzi, Generalization in vision and motor control, Nature 431 (2004), 768-774.
[16] F.A. Mussa-Ivaldi, Computational Intelligence in Robotics and Automation in Proc. 1997 IEEE (IEEE
Computer Society, Los Alamitos, California, Int. Symp., 1997, 84-90.
[17] N. Shelton, M.J. Reding, Effect of lesion location on upper limb motor recovery after stroke, Stroke 32
(1) (2001), 107-112.
[18] C.E. Han, M.A. Arbib, N. Schweighofer, Stroke rehabilitation reaches a threshold, PLoS Computational
Biology 4(8) (2008), e1000133.
[19] M.L. Aisen, H.I. Krebs, N. Hogan, F. McDowell, and B.T. Volpe, The effect of Robot assisted therapy
and rehabilitative training on motor recovery following stroke, Archives of Neurology 54 (1997), 443-
446.
[20] E. Todorov, R. Shadmehr, and E. Bizzi, Augmented feedback presented in a virtual environment
accelerates learning of a difficult motor task, Journal Motor Behavior 29 (1997), 147-158.
[21] P. Morasso, Spatial control of arm movements, Experimental Brain Research 42 (1981), 223-227.
[22] W. Abend, E. Bizzi, and P. Morasso, Human arm trajectory formation, Brain 105 (1982), 331-348.
[23] C.A.M. Stan Gielen, Movement dynamics, Current Opinion in Neurobiology 3 (6) (1993), 912-916.
[24] A.R. Fugl-Meyer, L. Jaasko, L. Leyman, S. Olsson, S. Steglind, The post-stroke hemiplegic patient. 1. a
method for evaluation of physical performance, Scandinavian Journal of Rehabilitation Medicine 7 (1)
(1975), 13-31.
[25] A.E. Russon, Learning by imitation: a hierarchical approach, Behavioral and Brain Sciences 21 (5)
(1998), 667-684.
[26] R. Keiith, C. Granger, B. Hamilton, G. Sherwin, The functional independence measure: a new tool for
rehabilitation, in M.G. Eisenburg, R.C. Grzesiak (Eds), Advances in clinical rehabilitation. Springer
Publishing Co., New York, 1987, pp. 6-18.
[27] M. Dam, P. Tonin, S. Casson, et al., The effects of long-term rehabilitation therapy on poststroke
hemiplegic patients, Stroke 24 (8) (1993), 1186-1191.
[28] P.T. Tangeman, D.A. Banaitis, A.K.Williams, Rehabilitation of chronic stroke patients: changes in
functional performance, Archives of Physical Medicine and Rehabilitation 71 (11) (1990), 876-880.