Effects of continuous visual feedback during sitting balance training in chronic stroke survivors
Pellegrino et al. Journal of NeuroEngineering and Rehabilitation
Effects of continuous visual feedback during sitting balance training in chronic stroke survivors
Laura Pellegrino 0
Psiche Giannoni 0
Lucio Marinelli 1 2
Maura Casadio 0
0 Department Informatics , Bioengineering , Robotics and Systems Engineering, University of Genoa , Via Opera Pia, 16145 Genoa , Italy
1 Department of Neuroscience, Ospedale Policlinico San Martino, L.go R. Benzi , Genoa 16132 , Italy
2 Department of Neuroscience, Rehabilitation , Ophthalmology, Genetics, Maternal and Child Health (DINOGMI) , University of Genoa, L.go Daneo , Genoa 16132 , Italy
Background: Postural control deficits are common in stroke survivors and often the rehabilitation programs include balance training based on visual feedback to improve the control of body position or of the voluntary shift of body weight in space. In the present work, a group of chronic stroke survivors, while sitting on a force plate, exercised the ability to control their Center of Pressure with a training based on continuous visual feedback. The goal of this study was to test if and to what extent chronic stroke survivors were able to learn the task and transfer the learned ability to a condition without visual feedback and to directions and displacement amplitudes different from those experienced during training. Methods: Eleven chronic stroke survivors (5 Male - 6 Female, age: 59.72 ± 12.84 years) participated in this study. Subjects were seated on a stool positioned on top of a custom-built force platform. Their Center of Pressure positions were mapped to the coordinate of a cursor on a computer monitor. During training, the cursor position was always displayed and the subjects were to reach targets by shifting their Center of Pressure by moving their trunk. Pre and post-training subjects were required to reach without visual feedback of the cursor the training targets as well as other targets positioned in different directions and displacement amplitudes. Results: During training, most stroke survivors were able to perform the required task and to improve their performance in terms of duration, smoothness, and movement extent, although not in terms of movement direction. However, when we removed the visual feedback, most of them had no improvement with respect to their pre-training performance. Conclusions: This study suggests that postural training based exclusively on continuous visual feedback can provide limited benefits for stroke survivors, if administered alone. However, the positive gains observed during training justify the integration of this technology-based protocol in a well-structured and personalized physiotherapy training, where the combination of the two approaches may lead to functional recovery.
Visual feedback; Trunk control; Posture; Stroke survivors; Center of pressure; Motor learning
In our daily life we maintain different postures and
automatically adjust our postural responses before starting
voluntary movements [
]. Following a stroke, some of
these abilities are often compromised . Several studies
reported an increased sway during quiet standing [
uneven weight distribution with increased weight
bearing on the unaffected limb in stance [
], a decreased
weight-shifting ability [
], and abnormalities in postural
3, 7, 8
]. For stroke survivors, good postural
control is important to support the weak side and to
reduce the effects of the altered postural tone [
impairments in sitting balance arise mainly because of
muscle weakness, loss of dexterity, sensory deficits, and
tendency to adopt compensatory strategies for avoiding
threats to balance .
A major focus of rehabilitation programs is to improve
balance and mobility for greater functional
independence. Visual feedback is largely used in rehabilitation to
improve the control of standing or sitting posture and to
train the ability to shift weight by moving the entire
body or the trunk [
]. A number of devices used in
the clinical practice provide training based on the
feedback of a cursor on a computer screen, controlled by the
position of the Center of Pressure (CoP) or of the Center
of Mass (CoM) [
1, 14, 15
Several studies showed that stroke survivors who
received rehabilitation treatments based on visual
feedback of their weight distribution on both feet or about
their CoP or CoM position, regained better standing
symmetry than those who received conventional physical
] or therapies designed to offer tactile
and verbal cues regarding postural symmetry [
and Lincoln [
] demonstrated that this improved stance
symmetry was also associated with an increased ability
to perform functional tasks. Lee et al. [
] showed that
in chronic stroke survivors, training with visual feedback
improves both static and dynamic sitting balance as well
as visual perception. However, a Cochrane review [
and two reviews on standing balance in stroke survivors
] highlighted a limited evidence of benefits of this
type of training. According to these reviews, different
studies reported small improvements [
] with a
limited long-term retention and no significant benefits
compared to conventional physical therapy [
With respect to sitting balance, there is evidence that
following a stroke, the muscles of the trunk are
], resulting in poor trunk control during
voluntary trunk and limbs movements [
few studies address the problem and the effects obtained
with exercises based on visual feedback are still unclear;
] for a review.
In the last few years, vision has been the feedback
modality most intensively investigated in the context of
optimizing augmented feedback for motor learning [
This line of research has revealed how visual feedback
strategies can either facilitate [
] or impair motor
Schmidt et al. [
] demonstrated that concurrent
feedback - i.e. feedback provided continuously during motor
task execution - can enhance performance in the
acquisition phase, but the performance gains are lost in
retention tests. This finding suggested that permanent
feedback during acquisition lead to a dependency on
Moreover, the control of posture involves vision,
proprioceptive and tactile feedback, as well as vestibular
input, and their sensorimotor integration [
]. Vision is
a dominant form of feedback [
], and one may
argue that by focusing mostly on vision, we could reduce
our attention on proprioceptive feedback , that is
fundamental in postural control. Therefore, it is
important to understand what we improve in terms of ability to
control and correctly estimate the position of our body
in space or to shift our weight, by focusing mainly on
continuous visual feedback.
Our hypothesis is that postural training with
continuous visual feedback is not effective at developing motor
programs (i.e., feedforward control) that can be executed
without reliance on feedback and applied to variety of
Thus, our specific goal was to understand if and to
what extent it is possible to improve sitting postural
stability of stroke survivors by training them to shift
their CoP while providing them with continuous visual
feedback of its position.
Here, we report an experiment where chronic stroke
survivors seated on a stool were trained to guide a point
representing their CoP in different directions toward a
fixed distance from a resting neutral position. As they
became increasingly able to control the motion of their
CoP under visual guidance, we investigated the retention
of this skill once the visual feedback was removed and
the transfer of the learned ability to different directions
as well as displacement amplitudes.
We enrolled 11 chronic stroke survivors (5 Male - 6
Female, age: 59.72 ± 12.84 years) recruited among
the outpatients of the Department of Neuroscience
of Ospedale Policlinico San Martino, Genoa, Italy.
Clinical assessment of the stroke survivors was
based on specific tests for evaluating trunk control
and balance: Berg Balance Scale (BBS) [
Impairment Scale (TIS) [
] and Nottingham
Sensory Assessment Scale (NSA) [
]. The subscale for
kinesthetic sensation of the NSA scale was used to
determine proprioceptive function. BBS and TIS
tests have been validated for use in the stroke
population and have been used to characterize balance
The exclusion criteria were: BBS < 25, TIS < 7, severe
hypovision (visual acuity with corrective lenses less than
1/10), cognitive disorders such as neglect (Albert’s Test,
]), inability to understand simple instruction
(MiniMental State Examination, MMSE > 24, ) and inability
to discriminate colors. All subjects had no severe aphasia
or problems of visual integrity and all were able to clearly
see the visual feedback provided in the computer monitor.
Table 1 summarizes the stroke subjects’ demographic
information and scores in the clinical scale.
The research conforms to the ethical standards laid
down in the 1964 Declaration of Helsinki that protects
research subjects. Each subject signed a consent form to
participate the study that conforms to these guidelines
and was approved by the local Ethical Committee (ASL 3
PH Paretic hand: (Right/Left), E Etiology, I/L Ischemic/Hemorrhagic, SL Site of lesion, DD disease duration (years), BBS Berg Balance Scale, TIS Trunk Impairment
Scale, NSA Nottingham Sensory Assessment Scale (kinesthetic section), n.a. not available
Genovese 09/04/2013). Moreover, all subjects consented
to publish individual data.
Set-up and protocol
Subjects were seated on a stool without back support
and with the hands resting on their legs. The stool had a
support for the feet and it was positioned on top of a
custom-built force platform. The platform (50X50 cm
surface) supported the entire body weight of the subjects
(Fig. 1 panel a). The signals from four load cells
positioned on the four corners of the platform (Fig. 1 panel
b) were collected at 1 kHz by using Real Time Windows
Target, Simulink, Mathworks.
The CoP coordinates were computed in real time by
using the following equation:
ðf2 þ f3−f1−f4Þ d=2
ðf3 þ f4−f1−f2Þ d=2
where fi was the force measured by the i (i = 1 to 4) load
cell and d (d = 40 cm) was the distance between two
adjacent load cells.
The estimated CoP positions were mapped to the
coordinate of a cursor (yellow circle, 5 mm radius) on a
computer monitor, positioned two meters away from the
platform at eye level. When the subject’s trunk was in
the upright posture (no bending), the cursor was at the
center of the screen. Targets were displayed as circles
(1 cm radius) against a black background.
The scale factor between the CoP displacement and
the monitor was adjusted to allow each subject to
comfortably move the cursor over the entire screen
(scale factor: 1.8 mean ± 0.2 SD; this determined a shift
of the CoP in the range of four to five cm for a displayed
cursor motion of length L = 8 cm). This calibration
procedure was performed before the experiment started.
Subjects were asked to move their trunk along the four
cardinal directions and the intermediate diagonal
directions. Then, the gain was set such that they were able to
move their CoP without difficulties in all the workspace.
This process ensured that all subjects could comfortably
reach the entire task space independent of their
individual ability and anthropometric characteristics.
At the beginning of the experiment, the experimenter
instructed the subjects to hold their trunk in the upright
position with a correct alignment of the spine, so as to
keep the cursor in the central (home) position. The
physical therapist and the experimenter controlled that
this condition was satisfied by each subject at the
beginning of each target set.
The experimenter asked participants to reach the targets
in 2 s and to be as fast and as accurate as possible. The 2 s
started when subjects left the starting position (i.e.,
distance of the cursor form its starting position > = target
radius). We explained to the subjects that, in all phases,
after these 2 s the color of the target would turn red,
indicating that the time to reach the target elapsed.
The protocol consisted of four phases: familiarization,
training, and pre- and post- training tests.
▪ Familiarization. The goal of this phase was to explain
to the subjects how and to what extent they had to move
for reaching the targets. During this phase, three targets
(Fig. 1 panel c, grey targets) positioned at L distance
(L = 8 cm) from the center of the computer screen were
presented twice (6 center-out movements) and the cursor
position was always displayed. The targets were presented
in three different directions: 0, 135, and 270 deg. Two
targets were located in the basic cardinal directions, i.e. one in
the antero-posterior direction (Fig. 1 panel c; 270 deg.,
backward) and the other in the medio-lateral direction (Fig.
1 panel c; 0 deg). The third target was on purpose selected
in a diagonal direction (Fig. 1 panel c; 135 deg.,
frontallateral direction). The main reason for this choice was that
functional movements and relative shifts of the CoP have
rarely components only in the sagittal or in frontal planes,
but often they are in directions that required combinations
of motion in both planes . The shift at 0 deg.
corresponded to the impaired side. The selection of the side for
the lateral direction was based on the impairment rather
than dominance, because we expect that the former
influence the performance more than the latter.
▪ Training. The goal of this phase was to exercise
movements of length L = 8 cm toward the three
directions presented also in the familiarization phase (Fig. 1
panel c, grey target). During the training phase, subjects
performed 8 target sets. In every set, the three peripheral
targets were presented eight times in pseudo-random
order, with the condition that each peripheral target was
not presented again before all three targets were
reached. Therefore, subjects performed a total of
8 * 3 * 8 = 192 center-out movements. A new peripheral
target was not presented to the subject if the cursor was
not in the central (home) position. This ensured that
each cursor center-out motion started from the central
target. Since we asked the subjects to minimize the error
after 2 s from their movement onset, we explained to
them that the color of the peripheral target would
change from green to red after these 2 s. If the cursor
reached the target before these 2 s, the target would not
change color, i.e. would remain green. Instead, if the
cursor was not inside the target after these 2 s, the target
would turn red and it would remain red until the cursor
reached the target. In both cases the target disappeared
after the cursor stayed inside the target for 1 s. Then a
new target would appear in the home position. Visual
feedback of the cursor was suppressed for the first 2 s of
the movement on 1/8 of the center-out trials. The order
of presentation of the no visual feedback trials was
pseudo-random, with the constraint that all peripheral
targets were presented once without visual feedback in
each movement set. The cursor disappeared as the new
target appeared. The cursor re-appeared 2 s after it
moved out of the home target. Then, if after these two
seconds, the subjects did not reach the desired target,
they should correct their error as in the other trials of
the training phase, where the visual feedback of the
cursor was available.
Subjects were aware of the presence of these no VF
trials. This suppression of the VF was applied to test
how subjects transferred the improvement in
performance to the no visual feedback condition during the
▪ Test. The goal of the test phases was to test if and to
what extent subjects were able to transfer performance
improvement obtained with visual feedback training to
conditions where they had to move (i) without visual
feedback of the cursor and (ii) to different directions and
(iii) to different displacement amplitudes
(scaling-expansion) with respect to the training phase. To reach this
goal we compared the performance without visual
feedback in the pre-training and in the post-training tests.
The pre-training and post-training test phases were
identical and consisted of 5 target sets. In each target set
14 targets were presented in random order. Eight
equispaced targets positioned at the trained distance
(L = 8 cm) from the center, three along the directions
presented both in the familiarization and in the training
phase and five located along not trained directions. The
other six targets were presented along the trained
directions, but at different distances from the home position:
three targets located at half of the trained distance from
the center L/2 = 4 cm and three targets located L/2
further with respect to the trained distance (L + L/
2 = 12 cm). Therefore, in the test phases subjects
performed a total of 5 * 14 = 70 center-out movements.
The visual feedback of the cursor was always absent.
In each movement-set the first peripheral target would
appear only when the subjects were in the home
position. In each trial, when a target was presented subjects
had 2 s to reach it. After these 2 s, the color of the target
turned red and subjects were instructed to stop moving
their CoP. After 10 ms, this red target disappeared, a
new target appeared and subjects could move again their
CoP to reach the new target as described above. We
carefully check in the data analysis that all subjects
followed the instruction to stop when the time elapsed.
The position held when a new target was presented
was considered as new starting position. The protocol
alternated one of the 15 peripheral targets and the home
target. Subjects could fail to reach correctly both the
home target and the peripheral targets. This fact was
taken into consideration in the data analysis (see below).
The experimental session lasted about 1 h. The
protocol required a minimum of 2 min break between each
phase and subjects were allowed resting when and as
long as they needed. We video recorded all subjects
while performing the experiment.
We focused on the control signal - i.e. CoP - that
determined the cursor motion. The cursor trajectories were
sampled at 100 Hz and filtered by using a 6th order
Savitzky-Golay filter with a time window of 170 ms
(equivalent cut off frequency: ~11 Hz), which allowed us
to estimate the first three time derivatives (speed,
acceleration, and jerk). The movement onset was defined as
the first time instant when the cursor speed exceeded a
threshold (10% of peak speed) . The movement
termination was defined as the last time the cursor
speed went below and remained under the same
threshold for 1 s. Another important time point considered in
the following analysis is 2 s after the cursor (visible or
invisible) left the starting target. This was the time given
to the subjects for the subject to reach the target. In all
trials after these 2 s the color of the target turned red
indicating that the time to reach the target elapsed.
During training if the subjects had not correctly
captured the target in these 2 s they could correct their
error while the test phases, no corrections were allowed.
We analyzed the following performance indicators:
– Reaching error at 2 s (RE2): the distance between
the cursor and the target calculated at the end of
the first 2 s of the cursor movement. This measure
accounts for movement accuracy (aiming) and speed
, i.e. the error at 2 s decreases if a subject moves
faster and in the correct direction.
We also looked at the reaching error after 2 s in
terms of extent error and directional error. We
considered two vectors: one connecting the starting
point of the cursor motion with the target, the other
connecting the same starting point with the cursor
position after 2 s of motion.
The directional error was computed as the angle
between these two vectors, while the extent error
was computed as the difference between their
lengths. We analyzed both the absolute error
(average of the absolute values obtained in each
trial) and the systematic error (average of the signed
values obtained in each trial). The systematic extent
error indicated if the subjects undershoot (negative
values) or overshoot (positive values) the target.
While in the trials with visual feedback, the starting
point was always inside the home target, in the blind
trials this might not be the case, i.e. the subjects
might believe to be in the home target while they
were not. In this case, they could not account for
their initial shift and they would aim at peripheral
targets located in a circle centered on their actual
starting position (instead of the home position). To
verify that the errors we observed were not mainly
caused by this mismatch of the initial positions, we
computed the direction and extent errors also
toward target locations translated such that the
center of the target space corresponded to the
cursor starting position, i.e. the location of the
cursor when the peripheral target appeared.
– Movement Duration: the time difference between
movement onset and movement termination.
– Normalized jerk index (NJ): the root mean square of
the jerk (third time derivative of the trajectory),
normalized with respect to movement amplitude
and duration ; this is a measure of smoothness
for the cursor control. We computed this indicator
considering (averaging) both the entire trajectory
from movement onset to movement termination
(JN) and the first 2 s of the movement from the
movement onset (NJ2).
Finally, we verified if in the blind target sets the
subjects were able to maintain correctly the central home
position and if they improved this ability after training.
This is particularly important because returning
correctly to the home position reflects the ability to
recognize and maintain a correct alignment of the spine
without retroversion of the pelvis  and/or a lateral
weight shift , problems often observed in stroke
We considered the position of the cursor when it had
stopped moving and the subjects assumed to be in the
home position, just before the appearance of a peripheral
To quantify the systematic and the variable errors of the
shift with respect to this position, we used the following
indicators, similar to those defined by Dukelow et al. :
– Variability: this indicator was calculated by
computing the standard deviation of the x and y
coordinates of the cursor position for each target
location and averaging these standard deviations
across all target positions for the x (Varx) and y
(Vary) coordinates. Then we combined both Varx
and Vary values as follow (Varxy):
Varxy ¼ qffiVffiffiffiaffiffirffiffix2ffiffiffiþffiffiffiffiffiVffiffiffiaffiffiffirffiffiy2ffi
– Systematic shift: this indicator is the constant error
between the position of home target and the
position of the cursor. We computed the mean error
between the position of the home target and the
corresponding cursor positions for each target
location and then we averaged the obtained values
across all target positions. We computed this
indicator for the x (Shiftx) and the y (Shifty)
coordinates and both values combined (Shiftxy) as
Shiftxy ¼ qffiSffiffihffiffiffiiffifffitffiffix2ffiffiffiþffiffiffiffiffiSffiffihffiffiffiiffifffitffiffiy2ffi
For both the variability and the systematic shift
indicators, we considered the average values of the five target
sets of the pre-training test and the five target sets of
Statistical analysis was carried out within Statsoft
environment (Statistica 7.1, Statsoft TULSA, USA).
Repeatedmeasures ANOVA compared the performance measures
across training periods, visual feedback conditions, target
directions and displacements. Specifically, to test how
performance changed during practice we ran a repeated
measure ANOVA with two factors: training (first vs last
movement set) and directions (0, 135, 270 degree;
A post-hoc analysis (Fisher’s LSD test) was used to
verify statistically significant differences among
directions after repeated measures ANOVA.
To quantify if and how the learned abilities were
transferred to the no visual feedback condition, (i) to different
target directions and (ii) to displacement amplitudes, we
ran a repeated ANOVA with two factors: training (PRE
vs POST, i.e. pre-training test vs post-training test) and
(i) directions (8 equi-spaced directions) or (ii)
displacement amplitudes (target distance from the center: L/2,
L, L + L/2), respectively.
We tested for sphericity using the Mauchly’s test;
when the sphericity was violated we applied the
To specifically test for differences between the three
trained and the five not trained directions as well as between
the trained and the untrained displacement amplitudes in
the post-training test we used planned comparisons.
Changes in performance within subject were tested by
using the Student’s t-test for paired samples.
Also, for each condition, we verified if there were
significant changes by comparing all the trials of the
pretraining test with all the trials of the post-training test or
comparing only the first three trials before and the first
three trials after training.
Finally, we verified if there were significant changes
between the pre-training test and the post-training test
with respect to the variability and systematic shifts of
the subject’s home position in the blind trials by using
the Student’s t-test for paired samples.
Effects were considered significant at the p < 0.05 level.
Performance during training with continuous visual feedback
All stroke survivors were able to improve their performance
as a result of the training except P11; this subject had no
statistically significant changes in performance for all the
indicators at the beginning and at the end of the training.
Since P11 did not change the performance during training
with visual feedback, we couldn’t test the transfer of the
improvement to other conditions. In the following we
analyzed the changes in performance due to training only for
the other ten stroke survivors. The stroke survivors had
poor performance in the first block of training, as shown by
the trajectories of a selected subject in Fig. 2.
However, with training they improved duration,
accuracy, and smoothness of the cursor trajectories (see Fig. 3
panel a: movement duration F(1,9) = 52.48, p < 0.001;
panel B: RE2 F(1,9) = 14.67, p = 0.01; panel C: smoothness
– NJ F(1,9) = 12.50, p = 0.01).
With respect to accuracy, the stroke survivors reduced
significantly their absolute extent error (Fig. 3 panel f;
training effect: F(1,9) = 19.82, p = 0.01). The systematic
extent error highlighted a tendency of the population to
slightly undershoot peripheral targets (beginning:
−0.9 ± 0.3 SE cm; end of the training −0.8 ± 0.2 SE cm)
and this strategy did not change significantly with
training (F(1,9) = 0.92, p = 0.53).
In the first target set of the training, stroke survivors
had a relevant absolute directional error, but most of them
(7 subjects out of 10) did not improve with training (Fig. 3
panel d; training effect: F(1,9) = 2.90, p = 0.12).
The analysis of the smoothness showed that stroke
survivors had a lower normalized jerk over the first 2 s
of cursor motion (JN2: Fig. 3 panel e) with respect to
the second part of the cursor trajectory (training effect:
F(1,9) = 7.27, p = 0.03), i.e. the stroke survivors made
few corrections in the first 2 s of motion although their
trajectories had a relevant directional error. Then, they
corrected the cursor position in the last part of the
trajectory by using the visual feedback.
The performance of stroke survivors depended on the
target direction in terms of duration (F(2,18) = 52.48,
p < 0.0001), smoothness (NJ: F(2,18) = 12.5, p = 0.021)
and accuracy of the cursor trajectories (RE2:
F(2,18) = 6.91, p = 0.032; directional error: F(2,18) = 0.21,
p = 0.81 and extent error: F(2,18) = 10.55, p = 0.041).
For the stroke survivors both the diagonal (135 degree)
direction and the lateral (0 degree) displacement toward the
impaired side were difficult to control. Thus, the backward
direction was significantly easier to control than the other
two directions in terms of duration (270 vs 0 deg.: p < 0.001,
270 vs 135 deg.: p < 0.001), smoothness (NJ, 270 vs 0 deg.:
p = 0.04 and 270 vs 135 deg.: p < 0.001) and accuracy (RE2,
270 vs 0 deg.: p < 0.001 and 135 vs 0 deg.: p = 0.01, extent
error: 270 vs 0 deg.: p = 0.001 and 135 vs 0 deg.: p = 0.03).
For all the indicators the rate of improvement was
equal across directions despite the differences in the
initial performance, i.e. there were not interaction effects
between direction and training.
Performance in the trials without visual feedback
We expected worse performance in post-training test
with respect to the level they reached at the end of the
training with visual feedback because of the absence of
visual feedback, and indeed the accuracy was lower (see
Fig. 4, panel a: RE2 F(1,9) = 70.97, p < 0.001; panel B:
directional error F(1,9) = 11.98, p = 0.01; panel C: extent
error F(1,9) = 29.85, p < 0.001). However, since the
stroke survivors improved their performance in the
visual feedback condition during training, to verify if they
transferred to any extent the learned ability to the no
visual feedback condition, we compared the performance
in the pre- and post-training test. This comparison
highlighted that most stroke survivors had evident
difficulties to transfer to any extent the learned abilities to
the no visual feedback condition, to different target
directions, and displacement amplitudes. Specifically, in
the post-training test, the RE2 displayed a trend of
improvement with respect to the pre-training test, but
without statistical significance neither in the trained
targets nor in targets at the different distances or directions
(Fig. 4 panel a; displacement amplitudes: F(1,9) = 8.57,
p = 0.23 and targets’ directions: F(1,9) = 10.82,
p = 0.42). The absolute directional error had no
relevant changes in the training phase, thus we expected
no improvement also in the post-training test (Fig. 4
panel b; displacement amplitudes: F(1,9) = 0.12, p = 0.73
and target directions: F(1,9) = 1.14, p = 0.31). However,
the improvement obtained for the absolute extent error
during training was not transferred to the test in absence of
VF, neither for the different target amplitudes nor
directions (Fig. 4 panel C; displacement amplitudes:
F(1,9) = 2.27, p = 0.16 and targets’ directions: F(1,9) = 3.51,
p = 0.19).
In the signed extent error, we observed that the stroke
survivors in the absence of visual feedback overshoot the
targets for L/2, L, L + L/2 distance (2.0 ± 0.2 SE cm,
2.4 ± 0.3 SE m and 3.1 ± 0.8 SE cm, respectively); but
there was no significant improvement in the
posttraining test (1.9 ± 0.2 SE cm, 2.2 ± 0.2 SE cm and
2.9 ± 0.7 SE cm, respectively) with respect to the
pretraining test (F(1,9) = 0.26, p = 0.70). Both in
pretraining and post-training test phases, this error
increased with increasing target distance. No improvement
was also observed with respect to the different target
directions (F(1,9) = 2.21, p = 0.83).
These results were also confirmed by the analysis of the
no VF trials during training, where stroke survivors had
no significant changes for any indicators, e.g. the RE2, the
absolute extent error and absolute directional error were
approximately equal in the first (5.1 ± 0.7 SE cm, 1.5 ± 0.1
SE cm and 5.87 ± 0.181 SE deg., respectively) and in the
last training set (5.0 ± 0.9 SE cm, 1.6 ± 0.1 SE cm and
5.47 ± 0.253 SE deg., respectively).
The accuracy indicators were stable in the target
sets of the pre-training test and post-training test
phases for stroke survivors (Fig. 5 panel a). The NJ2
was the only parameter that decreased (Fig. 5 panel
b) in the target sets without VF for the different
target amplitudes (F(1,9) = 14.14, p = 0.02) and
directions (F(1,9) = 31.49, p < 0.001).
Systematic shift and variability of the starting position
Stroke survivors exhibited a systematic shift (Shiftxy –
pre-training test: 2.0 ± 0.4 SE cm and post-training test:
1.9 ± 0.2 SE cm) and high variability of the starting
position (Varxy - pre-training test: 2.3 ± 0.2 SE cm and
posttraining test: 2.4 ± 0.3 SE cm). In the post-training test,
the stroke survivors did not change significantly either
the systematic shift (p = 0.75) or the variability error
(p = 0.54) with respect to the pre-training test (Fig. 6
Specifically, during pre-training test all stroke
survivors had a relevant retroversion of the pelvis and this
did not change in the post-training test (Vary
-pre-training test: −0.9 ± 0.6 SE cm and post-training test:
−0.5 ± 0.4 SE cm). Moreover, most stroke survivors (8
out of 10) had a slight shift toward impaired side during
both the pre-training test and the post-training tests
(Shiftx: 0.3 ± 0.2 SE cm).
Considering this shift of the home position, we
computed the direction and extent errors not only with
respect to the real target positions as reported above, but
also with respect to the target positions re-centered on
actual initial position of the cursor in each trial. The
absolute directional and the extent errors computed
following this method confirmed the results described in
the previous paragraph, since these indicators did not
change significantly in the post-training test with respect
to the pre-training test. Specifically, both measures did
not change significantly either for the different target
amplitudes (absolute extent error – pre-training test: L/
2 = 4.0 ± 0.3 SE cm, L = 3.9 ± 0.3 SE cm, L + L/
2 = 5.9 ± 0.6 SE cm and post-training test: L/
2 = 3.9 ± 0.2 SE cm, L = 3.8 ± 0.3 SE cm, L + L/
2 = 5.9 ± 0.4 SE cm; absolute directional error
–pretraining test: L/2 = 22 ± 2.05 SE deg., L = 14.7 ± 0.9 SE
deg., L + L/2 = 15 ± 0.9 SE deg. and post-training test:
L/2 = 21 ± 1.4 SE deg., L = 14.3 ± 0.7 SE deg., L + L/
2 = 15.2 ± 1.4 SE deg) or for the different target
directions (absolute extent error - pre-training test: 4.4 ± 0.5
SE cm and post-training test: 4.3 ± 0.3 SE cm; absolute
directional error - pre-training test: 16.2 ± 1.4 SE deg.
and post-training test: 16.4 ± 0.9 SE deg).
Most stroke survivors did not transfer the trained ability
to the no visual feedback conditions. Only three out of
ten had changes in the post-training performance with
respect to their pre-training test. These subjects were
different from the others also during training since they
reduced not only the extent, but also the directional
error. After training, one stroke survivor, P3, generalized
the learned ability for both different target directions
and displacement amplitudes. This subject had relevant
changes post training for RE2 (different amplitudes: L/2:
p = 0.03, L: p = 0.04, L + L/2: p = 0.001 and targets’
directions: p < 0.001), absolute extent error (different
amplitudes: L/2: p < 0.001, L: p = 0.03, L + L/2: p = 0.01
and targets’ directions: p < 0.001) and absolute
directional error (different amplitudes: L/2: p < 0.001, L:
p < 0.001, L + L/2: p = 0.001 and targets’ directions:
p < 0.001).
The subjects P2 and P4 had a significant decrease of
the RE2 for the targets at the same distance from the
home position, although in different directions (P2:
p = 0.004 and P4: p = 0.02) with respect to the trained
targets. Specifically, P2 reduced both the absolute extent
error and the absolute directional error (p = 0.001 and
p = 0.002, respectively) while P4 reduced only the extent
error (p = 0.003). No relevant changes were found for
the targets located at different displacement amplitudes.
Discussion and conclusion
Subjects were seated on a force platform and were
trained to reach targets positioned in three different
directions at a fixed distance from a central target, by
controlling a cursor with their CoP displacement.
During training, subjects were provided with continuous
visual feedback of the cursor.
Our goal was to study if chronic stroke survivors can
lean the task and transfer the learned ability to the
different conditions: (i) no visual feedback, (ii) other targets
directions and (iii) different displacement amplitudes.
Most chronic stroke survivors were able to perform the
task, but they strongly relied on on-line visual feedback
corrections and most of them did not transfer the
learned ability to the no visual feedback condition.
We expected that stroke survivors were able to
perform the required task and to improve their
performance in presence of visual feedback. This supported by
several studies on platform training for recovering
balance while standing [
16, 18, 22, 53, 54
], sitting [
] or during the transition between these two
conditions . In a broader view, this result is also in
agreement with the finding that even if the control and
execution of motor skills are impaired after stroke, the
ability to learn those skills is not compromised .
However, stroke survivors strongly relied on visual
feedback to complete the task during all training. They
had relevant absolute directional error, but most of them
did not cancel these errors in a predictive way as the
training proceeded, they just corrected the second part
of the cursor trajectory by using visual feedback. This
finding is supported by the hypothesis that many
subjects with hemiplegia have an excessive reliance on
visual feedback  and this is observable since the
acute stage [58, 59]. Thus, a training based only on
continuous visual feedback may decrease in a significant
manner the role of proprioceptive, tactile and vestibular
feedback in the control of posture [60–62] even when
these sensory modalities remain intact after the stroke.
This negative effect may be even stronger in the case of
proprioceptive impairment, precluding a progressive
recovery of such important sensory channel. Thus, if
subjects have postural control problems arising from these
other feedback inputs, a training based on visual cues
can lead to a reduced improvement of balance, with
respect to a training based on visual cue deprivation .
This dependence on on-line feedback has also another
important implication: it leads to modifying the behavior
by relying more on on-line corrections than applying
feedforward adjustments based on previous experience.
Scheidt and Stoeckmann  observed a similar
behavior for upper limb movements. They reported that
stroke survivors assigned a significantly low weight to
prior movement errors when planning subsequent
These findings suggest that stroke survivors have
difficulties to form an internal model of the proposed task.
The same conclusion was derived for the control of the
contralesional arm during reaching movements: Takahashi
and Reinkensmeyer  demonstrated that the
hemiparesis stroke impairs the ability to implement internal
models used for anticipatory control of arm movements.
In our experiment, when visual feedback was removed,
stroke survivors had difficulty not only repeating the task,
but also maintaining a correct sitting posture. Before
training, stroke subjects had a marked retroversion of the
pelvis and most of them tended to slightly shift their CoP
toward the affected side. During training with VF they had
a correct alignment of the spine, but after training they
assumed again the same incorrect posture.
The fact that most stroke survivors when visual
feedback was removed had difficulty maintaining a correct
sitting posture or retaining information either of the
direction or extent of their body shift, could predict also
strong limitations for the translation to daily life
activities of the postural control and balance skills learned
with this VF training.
In this respect, some studies found improvement in falls
prevention , in activities of daily living and gross
motor functions . However other works, investigating
the effects of postural training in both standing [
and sitting  positions, highlighted a difficulty for stroke
survivors to transfer to daily life activities the
improvement obtained in solving the platform-based exercises.
Also a Cochrane Review  concluded that providing
feedback from a force platform do not improve balance
and independence during functional activities.
However, stroke survivors were able to perform the
required task and to improve their performance during
training with visual feedback. These positive gains observed
during training justify the integration of this
technologybased protocol in a well-structured and personalized
physiotherapy training. The combination of the two
approaches may lead to functional recovery, as we observed
for robotic and physical therapies in previous studies .
Individual performance and limitations of the study
While the results presented and discussed were robust
across our entire stroke subject population, there were
important individual differences. One subject was not
even able to improve task performance by using the
visual feedback while another was able to transfer the
learned ability to the task performed without visual
It is worth noticing that the three subjects, who to
some extent transferred the learned ability to the no VF
condition were the only ones among stroke survivors
that also decreased the directional error with training.
There are several possible factors accounting for these
differences, such as the location of the lesions [
and age. The subject who did not learn the task was the
oldest one and had the lowest scores in the BBS and TIS
clinical scales. However, further elaborations on this
would not be warranted given the limited sample size of
our subject population.
The small sample size of this study together with the
absence of an age and a sex matched control group for
the stroke survivors are the main limitations of this
study. In a larger population, it would be possible and
interesting to investigate the individual results taking
into account the location of the lesion, the demographic
data and the individual abilities of the subjects to
proficiently use the somatosensory, vestibular, and visual
Further studies are necessary to verify if different
protocols – for example based on intermittent or terminal
feedback– could lead to more significant improvements
for the chronic stroke population.
We cannot exclude that different results could be
obtained with longer training over multiple days and with
stroke survivors in the acute stage.
This study suggests that a postural training based
exclusively on continuous visual feedback could provide
limited benefits for many stroke survivors, if administered
alone i.e. not as a part of a well-structured and
personalized physiotherapy training.
Since most subjects immediately after stroke have an
impairment of proprioceptive sensory channels and thus
are forced to exaggerate the importance of visual feedback
] in the organization of purposeful actions, it would be
important to design experimental protocols that are
capable to provide effective on-line information of the
impairment level of proprioceptive channels and,
accordingly, can induce a gradual reduction of visual feedback in
favor of the underused proprioceptive system.
Although it may appear that the main result of the study
is to refute the ability of posture training programs
focused on continuous visual feedback to provide robust
clinical gains in postural recovery of stroke survivors,
merely supporting the conclusion of the Cochrane review
, we wish to emphasize that this is only part of the
story. We wish instead to encourage therapists who are
using this training protocol to continue using it, but with
a clear understanding of its limits/drawbacks and in
particular with a creative, patient-tailored combination of
this technique with other (manual or technical)
interventions that motivate the stroke survivors subjects to better
attend the proprioceptive awareness of their body. In
general, this attitude is consistent with the suggestion of a
synergy between technology assisted therapy to
physiotherapy in the treatment of stroke survivors .
ANOVA: Analysis of variance; BBS: Berg Balance Scale; cm: Centimeter;
CoM: Center of Mass; CoP: Center of Pressure; deg.: degree; L: Length;
ms: Millisecond; NJ: Normalized jerk index; NJ2: Normalized jerk index over
the 2 s of movement; NSA: Nottingham Sensory Assessment Scale;
RE2: Reaching error at 2 s; SE: Standard error; Shift: Systematic shift; TIS: Trunk
Impairment Scale; Var: Variability error; VF: Visual feedback
The authors want to thank Professor Pietro Morasso for comments and
helpful suggestions, and Filippo Sante for the help on building the setup.
This work was supported by the Marie Curie Integration Grant
Availability of data and materials
The data sets during and/or analyzed during the current study are available
from the corresponding author upon request.
All the authors conceived the study and designed the experimental protocol.
LP and MC developed the experimental setup. LM recruited most of the
patients and evaluated lesion locations on brain scans. LP and PG run the
experimental sessions. LP and MC analyzed the results. All the authors
contribute to discuss the results and to write the manuscript. All authors
read and approved the final manuscript.
Ethics approval and consent to participate
All procedures were approved by local Institutional Boards (ASL3 Genovese)
in accord with the 1964 Declaration of Helsinki.
Consent for publication
Consent provided upon request.
The authors declare that they have no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
1. Walker C , Brouwer BJ , EG C. Use of visual feedback in retraining balance following acute stroke . Phys Therapy Pract . 2000 ; 80 : 886 - 95 .
2. Winter DA . Human balance and posture control during standing and walking . Gait Posture . 1995 ; 3 ( 4 ): 193 - 214 .
3. Ryerson S , et al. Altered trunk position sense and its relation to balance functions in people post-stroke . J Neurol Phys Ther . 2008 ; 32 ( 1 ): 14 - 20 .
4. Shumway-Cook A , Anson D , Haller S. Postural sway biofeedback: its effect on reestablishing stance stability in hemiplegic patients . Arch Phys Med Rehabil . 1988 ; 69 ( 6 ): 395 - 400 .
5. Goldie PA , et al. Maximum voluntary weight-bearing by the affected and unaffected legs in standing following stroke . Clin Biomech . 1996 ; 11 ( 6 ): 333 - 42 .
6. Dettmann MA , Linder MT , Sepic SB . Relationships among walking performance, postural stability, and functional assessments of the hemiplegic patient . Am J Phys Med Rehabil . 1987 ; 66 ( 2 ): 77 - 90 .
7. Badke MB , Duncan PW . Patterns of rapid motor responses during postural adjustments when standing in healthy subjects and hemiplegic patients . Phys Ther . 1983 ; 63 : 13 - 20 .
8. Horak FB , et al. The effects of movement velocity, mass displaced, and task certainty on associated postural adjustments made by normal and hemiplegic individuals . J Neurol Neurosurg Psychiatry . 1984 ; 47 : 1020 - 8 .
9. Sheean G. The pathophysiology of spasticity . Eur J Neurol . 2002 ; 9 ( Suppl 1 ): 3 - 9 . dicussion 53- 61
10. Sheean G , McGuire JR . Spastic hypertonia and movement disorders: pathophysiology, clinical presentation, and quantification . PM R . 2009 ; 1 ( 9 ): 827 - 33 .
11. Dean CM , Shepherd RB . Task-related training improves performance of seated reaching tasks after stroke. A randomized controlled trial . Stroke . 1997 ; 28 ( 4 ): 722 - 8 .
12. Moore S , Woollacott M. The use of biofeedback devices to improve postural stability . Phys Therapy Pract . 1993 ; 2 : 1 - 19 .
13. Jung K , et al. Weight-shift training improves trunk control, proprioception, and balance in patients with chronic hemiparetic stroke . Tohoku J Exp Med . 2014 ; 232 ( 3 ): 195 - 9 .
14. Tsaklis PV , Grooten WJ , Franzen E. Effects of weight-shift training on balance control and weight distribution in chronic stroke: a pilot study . Top Stroke Rehabil . 2012 ; 19 ( 1 ): 23 - 31 .
15. Dault MC , et al. Effects of visual center of pressure feedback on postural control in young and elderly healthy adults and in stroke patients . Hum Mov Sci . 2003 ; 22 ( 3 ): 221 - 36 .
16. Winstein CJ , et al. Standing balance training - effect on balance and locomotion in Hemiparetic adults . Arch Phys Med Rehabil . 1989 ; 70 ( 10 ): 755 - 62 .
17. Ustinova KI , et al. Impairment of learning the voluntary control of posture in patients with cortical lesions of different locations: the cortical mechanisms of pose regulation . Neurosci Behav Physiol . 2001 ; 31 ( 3 ): 259 - 67 .
18. Matjacic Z , Hesse S , Sinkjaer T. BalanceReTrainer: a new standing-balance training apparatus and methods applied to a chronic hemiparetic subject with a neglect syndrome . NeuroRehabilitation . 2003 ; 18 ( 3 ): 251 - 9 .
19. Sackley CM , Lincoln NB . Single blind randomized controlled trial of visual feedback after stroke: effects on stance symmetry and function . Disabil Rehabil . 1997 ; 19 ( 12 ): 536 - 46 .
20. Lee SW , Shin DC , Song CH . The effects of visual feedback training on sitting balance ability and visual perception of patients with chronic stroke . J Phys Ther Sci . 2013 ; 25 ( 5 ): 635 - 9 .
21. Howe T , et al. Exercise for improving balance in older people . J Aging Phys Act . 2012 ; 20 : S132 - 3 .
22. Geurts ACH , et al. A review of standing balance recovery from stroke . Gait Posture . 2005 ; 22 ( 3 ): 267 - 81 .
23. Van Peppen RP , et al. Effects of visual feedback therapy on postural control in bilateral standing after stroke: a systematic review . J Rehabil Med . 2006 ; 38 ( 1 ): 3 - 9 .
24. Sihvonen S , et al. Fall incidence in frail older women after individualized visual feedback-based balance training . Gerontology . 2004 ; 50 ( 6 ): 411 - 6 .
25. Wolf B , et al. Effect of a physical therapeutic intervention for balance problems in the elderly: a single-blind, randomized, controlled multicentre trial . Clin Rehabil . 2001 ; 15 ( 6 ): 624 - 36 .
26. Lichtenstein MJ , et al. Exercise and balance in aged women: a pilot controlled clinical trial . Arch Phys Med Rehabil . 1989 ; 70 ( 2 ): 138 - 43 .
27. Chen IC , et al. Effects of balance training on hemiplegic stroke patients . Chang Gung Med J. 2002 ; 25 ( 9 ): 583 - 90 .
28. Bohannon RW , Cassidy D , Walsh S. Trunk muscle strength is impaired multidirectionally after stroke . Clin Rehabil . 1995 ; 9 ( 1 ): 47 - 51 .
29. Tanaka S , Hachisuka K , Ogata H . Trunk rotatory muscle performance in poststroke hemiplegic patients . Am J Phys Med Rehabil . 1997 ; 76 ( 5 ): 366 - 9 .
30. Tanaka S , Hachisuka K , Ogata H . Muscle strength of trunk flexion-extension in post-stroke hemiplegic patients . Am J Phys Med Rehabil . 1998 ; 77 ( 4 ): 288 - 90 .
31. Sigrist R , et al. Augmented visual, auditory, haptic, and multimodal feedback in motor learning: a review . Psychon Bull Rev . 2013 ; 20 ( 1 ): 21 - 53 .
32 Sulzenbruck S , Heuer H . Type of visual feedback during practice influences the precision of the acquired internal model of a complex visuo-motor transformation . Ergonomics . 2011 ; 54 ( 1 ): 34 - 46 .
33 Proteau L. Visual afferent information dominates other sources of afferent information during mixed practice of a video-aiming task . Exp Brain Res . 2005 ; 161 ( 4 ): 441 - 56 .
34 Schmidt RA . Frequent Augmented Feedback Can Degrade Learning - Evidence and Interpretations. Tutorials in motor . Neuroscience . 1991 ; 62 : 59 - 75 .
35 Horak FB . Postural orientation and equilibrium: what do we need to know about neural control of balance to prevent falls? Age Ageing . 2006 ; 35 ( suppl 2 ): ii7 - ii11 .
36 Asakawa K , et al. Effects of ocular dominance and visual input on body sway . Jpn J Ophthalmol . 2007 ; 51 ( 5 ): 375 - 8 .
37 Berthoz A , et al. The role of vision in the control of posture during linear motion . Prog Brain Res . 1979 ; 50 : 197 - 209 .
38 Dornan J , Fernie GR , Holliday PJ . Visual input: its importance in the control of postural sway . Arch Phys Med Rehabil . 1978 ; 59 ( 12 ): 586 - 91 .
39 Allum JH , et al. Proprioceptive control of posture: a review of new concepts . Gait Posture . 1998 ; 8 ( 3 ): 214 - 42 .
40 Berg KO , Wooddauphinee SL , Williams JI . Measuring balance in the elderly - validation of an instrument . Can J Public Health-Revue Canadienne De Sante Publique . 1992 ; 83 : S7 - S11 .
41 Verheyden G , et al. The trunk impairment scale: a new tool to measure motor impairment of the trunk after stroke . Clin Rehabil . 2004 ; 18 ( 3 ): 326 - 34 .
42 Lincoln N , Jackson J , Adams S. Reliability and revision of the Nottingham sensory assessment for stroke patients . Physiotherapy . 1998 ; 84 ( 8 ): 358 - 65 .
43 Blum L , Korner-Bitensky N . Usefulness of the berg balance scale in stroke rehabilitation: a systematic review . Phys Ther . 2008 ; 88 ( 5 ): 559 - 66 .
44 Albert ML . A simple test of visual neglect . Neurology . 1973 ; 23 ( 6 ): 658 - 64 .
Folstein MF , Folstein SE , McHugh PR . "Mini-mental state". A practical method for grading the cognitive state of patients for the clinician . J Psychiatr Res . 1975 ; 12 ( 3 ): 189 - 98 .
Knott M , Voss DE . Proprioceptive neuromuscular facilitation (pattern and techniques) . Am J Med Sci . 1957 ; 234 ( 4 ): 490 .
Nyberg L , Gustafson Y. Fall prediction index for patients in stroke rehabilitation . Stroke . 1997 ; 28 ( 4 ): 716 - 21 .
Danziger Z , Mussa-Ivaldi FA . The influence of visual motion on motor learning . J Neurosci . 2012 ; 32 ( 29 ): 9859 - 69 .
Teulings HL , et al. Parkinsonism reduces coordination of fingers, wrist, and arm in fine motor control . Exp Neurol . 1997 ; 146 ( 1 ): 159 - 70 .
Karthikbabu S , et al. Pelvic alignment in standing, and its relationship with trunk control and motor recovery of lower limb after stroke . Neurol Clin Neurosci . 2017 ; 5 ( 1 ): 22 - 28 .
Karthikbabu S , et al. A review on assessment and treatment of the trunk in stroke: a need or luxury . Neural Regen Res . 2012 ; 7 ( 25 ): 1974 - 7 .
Dukelow SP , et al. Quantitative assessment of limb position sense following stroke . Neurorehabil Neural Repair . 2010 ; 24 ( 2 ): 178 - 87 .
Barclay-Goddard R , et al. Force platform feedback for standing balance training after stroke . Cochrane Database Syst Rev . 2004 ; 4 : CD004129 .
Bonan IV , et al. Reliance on visual information after stroke. Part I: balance on dynamic posturography . Arch Phys Med Rehabil . 2004 ; 85 ( 2 ): 268 - 73 .
Engardt M , Ribbe T , Olsson E. Vertical ground reaction force feedback to enhance stroke patients' symmetrical body-weight distribution while rising/ sitting down . Scand J Rehabil Med . 1993 ; 25 ( 1 ): 41 - 8 .
Cheng PT , et al. The sit-to-stand movement in stroke patients and its correlation with falling . Arch Phys Med Rehabil . 1998 ; 79 ( 9 ): 1043 - 6 .
Winstein CJ , Merians AS , Sullivan KJ . Motor learning after unilateral brain damage . Neuropsychologia . 1999 ; 37 ( 8 ): 975 - 87 .
Bonan I , et al. Visual dependence after recent stroke . Ann Readapt Med Phys . 2006 ; 49 ( 4 ): 166 - 71 .
Yelnik AP , et al. Postural visual dependence after recent stroke: assessment by optokinetic stimulation . Gait Posture . 2006 ; 24 ( 3 ): 262 - 9 .
Henriques DY , Cressman EK . Visuomotor adaptation and proprioceptive recalibration . J Mot Behav . 2012 ; 44 ( 6 ): 435 - 44 .
Sainburg RL , et al. Effects of altering initial position on movement direction and extent . J Neurophysiol . 2003 ; 89 ( 1 ): 401 - 15 .
In Rehabilitation Robotics , 2007 . ICORR 2007 . IEEE 10th International Conference on. 2007 . IEEE.
Bonan IV , et al. Reliance on visual information after stroke. Part II: effectiveness of a balance rehabilitation program with visual cue deprivation after stroke: a randomized controlled trial . Arch Phys Med Rehabil . 2004 ; 85 ( 2 ): 274 - 8 .
Scheidt RA , Stoeckmann T. Reach adaptation and final position control amid environmental uncertainty after stroke . J Neurophysiol . 2007 ; 97 ( 4 ): 2824 - 36 .
Takahashi CD , Reinkensmeyer DJ . Hemiparetic stroke impairs anticipatory control of arm movement . Exp Brain Res . 2003 ; 149 ( 2 ): 131 - 40 .
Sackley C , Baguley B . Visual feedback after stroke with the balance performance monitor: two single-case studies . Clin Rehabil . 1993 ; 7 ( 3 ): 189 - 95 .
Winstein CJ . Knowledge of results and motor learning-implications for physical therapy . Phys Ther . 1991 ; 71 ( 2 ): 140 - 9 .
Palluel-Germain R , Jax SA , Buxbaum LJ . Visuo-motor gain adaptation and generalization following left hemisphere stroke . Neurosci Lett . 2011 ; 498 ( 3 ): 222 - 6 .