150
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 43, NO. 2, FEBRUARY 1996
oactivation
lectrically Stimulated
scles
Bing He Zhou, Member, IEEE, Richard V. Baram,* Member, IEEE, Moshe Solomonow, Senior Member, IEEE,
Leslie J. Olivier, 111, Giang T. Nguyen, and Robert D. D’Ambrosia
anterior translation and internal rotation of the tibial plateau
with respect to the femoral condyles [5].This can be avoided
by applying small amounts of tension in the antagonist flexor
(hamstrings) musculature [4], [18]. These studies suggested the
use of antagonistic co-contraction in applications of functional
electrical stimulation (FES) as well, to prevent the early onset
of joint arthropathy due to impaired muscle coordination.
Generally, applications of FES mainly have involved the
prime mover, or the agonist muscle group required to perform
a given movement, without much emphasis on the role of
the antagonistic musculature. With the current development of
FES systems which may in the future see prolonged functional
use by patients, the utilization of antagonist co-contraction
needs to be considered in order to avoid long term problems
which may appear due to increased joint laxity and nonphysiological contact stress distributions in the patients’ joints.
In an evaluation of stimulation tracking tasks, Durfee [6]
utilized a bidirectional control strategy which consisted of proportional pure flexor activity for flexion tasks and proportional
pure extensor activity for extension tasks, and coactivation
modes determined by the subject who controlled the tracking
task. No preprogrammed concurrent antagonist activity was
elicited. Possible advantages of this strategy include reducing
fatigue by minimizing overall activity and simplifying controller design. These advantages may come at the expense
of joint laxity at low force levels, nonphysiological articular
contact stress distributions, and unevenly sloped transitions
I. INTRODUCTION
between movements in opposite directions. Preprogrammed
URING normal movement, low level activity from the antagonistic activity using a co-contraction map in which
antagonist musculature was shown to distribute articular the antagonist muscle was engaged over a prescribed cocontact pressure [11, prevent joint subluxation [ 2 ] , provide stimulation range was also reported [7]. This strategy required
dynamic braking at high angular velocities [3], regulate joint a linear decrease in antagonist force as the agonist force
stiffness [l5] and torque, and compensate for changes in increased. When compared to the absence of co-contraction,
loading conditions [8], [16]. A particular issue of interest from this approach has the potential advantages of smoother tranthe orthopaedic standpoint is that of joint stability, which refers sitions between opposing movements (transition from flexion
to antagonists muscles’ tendency to help maintain joints in to extension), decreased joint laxity, and improved resistance
their correct relative alignment. In this regard, reports have to external disturbances.
shown that isolated knee extensor force results in undesirable
During muscular contractions in normal humans, the pattern
Manuscript received September 19, 1994; revised September 6, 1995. This of antagonistic activity has been described as an increasing
work was supported by the National Science Foundation under Grant BCS- function of agonist activity or net joint torque [8]. With
9207007. Asterisk indicates corresponding author.
increased joint torque output, the antagonist’s activity inThe authors are with the Bioengineering Laboratory, Louisiana State
University Medical Center, Department of OrthopaedicSurgery, New Orleans, creases, presumably to counteract the agonist’s increasingly
LA 70112 USA.
destabilizing influence on the joint, its ligaments, and articular
*R. V. Baratta is with the Bioengineering Laboratory, Louisiana State surfaces. In contrast, the strategies employed in electrical stimUniversity Medical Center, Department of Orthopaedic Surgery, 2025 Gravier
ulation systems use generally decreasing antagonist activity
Street, Suite 400, New Orleans, LA 70112 USA.
Publisher Item Identifier S 0018-9294(96)01049-X.
to simplify the control problem [7]. Therefore, fundamental
Abstract-The performance of various coactivation strategies to
control agonist-antagonist muscles in functional electrical stimulation (FES) applications was examined in a cat model using the
tibialis anterior and soleus muscles to produce d e isometric
dorsiflexion and plantarflexion torques, respectively. Three types
of coactivation strategies were implemented and tested. The
first strategy was based on coactivation maps described in the
literature as consisting of decreasing antagonistic activity as the
input command to the agonist was increased. The second type of
strategy was based on the physiologic coactivation data collected
from normal subjects exhibiting joint stabilizationduring the full
range of contractions. These strategies included scaled increasing
antagonist activity and therefore joint stiffness with increasing
agonist input command. A third strategy was devised which
at low force levels mimicked the strategies described in the
literature and at high force levels resembled strategies exhibited
by normal subjects. The three strategies were evaluated based on
their ability to track a linear or sinusoidal input command and
their efficiency of torque transmission across the joint. Coactivation strategies using increasing antagonist activity resulted in
decreased maximal joint torque and efficiency, decreased signal
tracking capability for linear inputs, and increased harmonic
distortion for sinusoidal inputs. Peak efficiency and tracking ability appeared when a moderate degree of antagonist activity was
engaged near the neutral joint position. Signal tracking quality
improved with earlier engagement of the antagonist muscles. Our
results suggest that strategies combining low-level coactivation as
described in the physiological literature and previous FES studies
could satisfactorilyaddress the issues of controllability,efficiency,
and long-term joint integrity.
0018-9294/96$05.00 0 1996 IEEE
______
ZHOU et al.: EVALUATION OF ISOMETRIC ANTAGONIST COACTIVATION STRATEGIES OF ELECTRICALLY STIMULATED MUSCLES
Pelvic
Clamp
Extensor Flexor
Stimulation Stimulation
Electrode Electrode
Fig. 1. Diagram of the experimental setup. The ankle joint is aligned with
the axis of a pivoting armature to which the foot is fastened securely. This
m a t u r e is, in turn, connected via a metal rod to a force transducer. A
transcondylar pin holds the femur rigidly in conjunction with a pelvic clamp.
Electrodes on the common peroneal and tibial nerves stimulate the TA and
SOL to produce dorsiflexion and plantarflexion, respectively.
differences exist between the co-contraction strategies of the
aforementioned FES systems and those recorded from normal
intact humans.
Given these antithetical requirements, it is the objective
of this study to assess the isometric performance of cocontraction strategies based on data recorded from normal
humans and on the FES paradigms of Durfee [6] and Chizeck
et al. [7] in order to develop strategies which address the
conflicting issues of optimal controllability and long term joint
integrity. These results will be directly applicable to the design
and development of optimal FES systems intended to restore
functional movement to paralyzed limbs while preserving the
integrity of joints, ligament and articular cartilage.
11. METHODS
A. Preparation
Four adult cats were anesthetized with a-chloralose (60
mg/kg). The sciatic nerve was exposed through a posterior
thigh incision, and the common peroneal nerve was exposed
through an anterior lateral shank incision. All nerve branches
except those innervating the soleus (SOL) and tibialis anterior
(TA) muscles were cut. These muscles were chosen because
of their similar maximal torques and elongation ranges. After
the shank incision was sutured closed, two tripolar nerve cuff
electrodes were placed through the thigh incision: one was
wrapped around the common peroneal nerve, the other around
the tibial nerve. These electrodes were later connected to the
stimulation system. A pin placed through the femoral condyles
and a pelvic clamp attached to a rigid platform achieved secure
proximal fixation with the hip and knee joints at 90" of flexion.
The ankle joint was secured in 90" of flexion to an isometric
torque-measuring armature as shown schematically in Fig. 1.
B. Instrumentation
Stimuli were delivered by a four-channel computercontrolled stimulation system that was developed as a
two-muscle version of a system described previously [9], [19].
151
Briefly, an IBM PC generated four analog output signals which
controlled four stimulators. Two of those stimulators were
voltage-controlled oscillators. They controlled the firing rate
of each muscle's motor units with pulses of 100-pS duration
and supramaximal amplitude, at repetition rates which could
range from 1 to 200 pulses per second (p/s). The other two
stimulators were responsible for orderly recruitment of motor
units via high frequency (600 p/s) blocking with variable
amplitude. Thus, this stimulation system was capable of
independently eliciting motor unit control strategies of the two
muscles under investigation in a near physiological manner
[9], 1191. CO-contraction and motor unit control strategy maps,
as well as calibration parameters were implemented through
software in the control computer. A representative diagram
of this system is shown in Fig. 2. Isometric joint torque
was measured by a Grass FT-10 force transducer which was
connected with a steel rod to the rotating armature where
the cat's ankle was secured. The measurement system had
a constant moment arm, and the ankle joint was set at 90";
Therefore, joint torque was related linearly to the measured
force. The torque data was sampled by an IBM-AT computer at
a rate of 50 sampleds. The output torque and four stimulation
input signals were also displayed on a Gould 260 polygraph.
C. Calibration
Initial calibration of the stimulation parameters was performed independently for both muscles. For each muscle, the
firing rate at which maximal isometric torque was achieved
was determined by 3-s trials. The first of these trials was
performed at 40 p/s. Subsequent trials were performed with
progressive increases of 3 p/s until the last trial showed no
visible torque increase over the previous trial. The initial
firing rate was determined by similar trials which started
at 5 p/s and progressively increased by 2 p/s until it was
observed that the torque produced by each stimulus pulse
did not return to its pretrial baseline between twitches. These
calibration trials set the upper and Iower limits of firing rate
for the experimental procedure. Trials were then conducted
to identify the maximal and minimal recruitment stimulus
intensities required to achieve maximum and minimum torque.
These stimulation intensity levels corresponded to a point just
above threshold of the largest motor units and just below
threshold of the smallest motor units [9], [19]. Throughout
calibration and the recorded experimental trials, three-minute
rest periods were strictly observed in order to minimize the
effects of fatigue on the data.
D. Co-Contraction Strategies
Three types of co-contraction strategies based on those
described by Durfee [6] and Chizeck et al. [7], coactivation
data recorded from normal subjects, and combinations of both
were examined. In implementing these strategies, however,
one must bear in mind that the stimulation technique utilized
in this study had a residual torque level at minimal activation
which was approximately 3-5% of maximal. While this could
be viewed as a potential disadvantage, in the behaving human
or animal, a background level of muscle tone is always present
due to baseline motor unit activity and passive muscular tissue
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 43, NO. 2, FEBRUARY 1996
152
Force
l.v---
I
I
'Command
Input b o - C o n t r a c t i o v
1
(W
I
I
Torque
I
I
I
I
Extensor
Activation
-
VCO)
Control
Strategy/
I
Dorsiflexor
Fig. 2. Block diagram of the agonist-antagonist stimulation system. A command input is generated in the control computer which uses the prescribed
co-contraction map to convert this sequence into activation signals for flexor and extensor muscles. Control strategy and calibration procedures determine
the stimulation parameters necessary to generate the prescribed muscle activation. In this particular study, a control strategy using concurrent motor unit
recruitment and firing rate up to 100% activation was employed. These signals were then used to control firing rate and recruitment control stimulators which
activate each muscle in a near physiological manner. The torque output of flexor and extensor are transduced through the joint into torque.
stiffness. The relative differences in co-contraction strategy
can be delineated despite this residual activity. One must,
however, consider that this residual torque results in increased
muscle activity for a given net joint torque, thus reducing the
joint's efficiency.
The input command signal resided in the computer software.
In the first part of the protocol, a step plantadexion input was
given, and one s allowed for the transient response to settle.
Then, the trial input command consisted of a linear ramp that
started at maximum plantarflexion torque and progressed to
maximum dorsiflexion torque over a 6-s period. In the second
part, an 8-s sinusoidal input command at 0.4 HZ ranging
from maximum plantarflexion torque to maximum dorsiflexion
torque was utilized. Muscle activation was computed on the
basis of the command signal and the coactivation map. It then
was used by the control strategy and calibration subroutines to
determine the actual stimulation parameters processed by the
stimulation hardware and delivered to each muscle.
1) Co-Contraction Strategies of Previous FES Studies: The
preprogrammed strategies used by Chizeck et al. [7] and
Durfee [6] have one common characteristic: the activation map
of each muscle is strictly increasing. Durfee's [6] strategy
consisted of pure proportional agonist activity, as shown in
Fig. 3(a). Chizeck et al.'s [7] strategy consisted of bilinear
activation, starting at a prescribed antagonist input. In this
study, the strategy was modified into a single linear map
because of our system's ability to provide a linear torque
versus activation curve. The strategy was defined as percent
overlap according to the point where antagonistic activity
was first elicited. In Fig. 3(b), the antagonist overlap is 50%
for both muscles. This means that if the command input
changes linearly from maximal flexion torque (- 1) to maximal
extension torque (l), the flexor starts fully active and decreases
its activity linearly, reaching zeroat the 50% of extension
command. At a 50% flexion torque input command, the
extensor is minimally activated, linearly increasing its activity
until the maximum extension torque point. In Fig. 3(c), the
antagonist overlap is 100% for both muscles. In this case, the
flexor starts fully active, and decreases its activity linearly,
reaching zero at maximum extension torque. At maximum
flexion torque, the extensor is minimally activated, linearly
increasing its activity until the maximal extension point torque.
In this strategy example, coactivation is present throughout
the entire contraction cycle. Durfee's paradigm [6] can be
considered a version of this strategy, with a 0% overlap;
throughout this report it is defined as such for simplicity. Five
overlap levels were investigated: 0%, 25%, 50%, 75%, and
100%. During voluntary contractions, differences in muscle
strength and stability requirements may dictate asymmetric antagonist strategies [2], [SI, [ 131. In practical FES applications,
asymmetric antagonist activity may be used as well. In this
study, however, antagonist activity was kept equal for both
muscles because it would be impractical to explore all the
possible permutations of overlap for the two muscles.
2) Co-Contraction Strategies of Normal Humans: In contrast to the strategies based on the practical control problems
of FES, the coactivation strategies recorded from normal
humans show that as increased net joint torque is required,
the antagonist's activity increases, albeit at a much lesser
rate than the agonist [SI, [14]. The rationale is based on
the fact that with increased destabilizing agonist activity, the
need for antagonist activity increases. A strategy based on
t h ~ sconcept is shown on Fig. 3(d), where the slopes of both
antagonists are 20% of unity. In the flexion region of the
command domain, the flexor activity is equivalent to the input
requirement, while the extensor is scaled to 20% of the input
command. Similarly, in the extension region of the command
domain, the extensor activity is scaled to a factor of one
with respect to the input command, whereas the flexor is
scaled to 20%. The slope of the antagonist linear activation
function is defined as antagonist gain. Antagonistic activity
depends on many variables, including development [lo], skill
[21 and [ l l ] , orientation with respect to the gravity vector 181,
and other factors, but it rarely exceeds 25% of the muscle's
~
ZHOU er al.: EVALUATION OF ISOMETRIC ANTAGONIST COACTIVATION STRATEGIES OF ELECTRICALLY STIMULATED MUSCLES
-1.0 -0.5 0.0 0.5
1.0
-1.0 -0.5
(a)
0.0 0.5
I
-1.0 -0.5 0.0 0.5
1.0
(b)
1.0
-1.0 -0.5 0.0
(C)
/,
0.5
1.0
(d)
\V
\
-1.0 -0.5 0.0 0.5
1.0
(e)
Fig. 3. This figure represents the input-output relationships of a variety
of co-contraction maps, or each activation function with changing input
command. On panel (a) 0% co-contraction is used, with no flexor activity
in extension and no extensor activity in flexion. (b) Illustrates a 50% overlap
strategy where the antagonist is engaged at 50% of the agonist command. (c)
Shows both muscles engaged throughout the contraction with one muscle
increasing its contraction intensity as the other decreases (d) Illustrates a
strategy based on physiologic principles, where the antagonist activity is
scaled to 20% of the agonist. (e) A compromise strategy is used where a
50% overlap is combined with 20% antagonist. This strategy is based on a
comparison of the two strategies and the selection of the one which would
result in the higher activity level. This type of strategy, when compared with
FES or physiology-based paradigms, results in increased joint stiffness, and
increased control with improved stability at the cost of reduced net joint
torque and joint efficiency.
maximal force (mostly in the range of 5-15%). Therefore,
several proportional co-contraction gains were examined: 0%,
5%, lo%, 20%, and 30%. The antagonist co-contraction gain
also was kept symmetric in these trials.
3) Combined CO-Contraction Strategies: Since sound control reasons exist for the strategies used by Durfee [6] and
Chizeck et al. [7] while the need remains for physiological
antagonist enhanced joint stability, a strategy combining aspects of both was devised. The combination superimposed
a physiologically based strategy on an FES based strategy
and chose the higher antagonist activity of the two. Fig.
3(e) shows a combination of the 50% overlap and 20%
of antagonist gain strategies. Near the neutral position, the
strategies based on FES systems are utilized. Near the extremes
of the input command domain, physiologically based strategies
dominate. The crossover point between strategies is prescribed
153
by the intersection of the lines defined by the two strategies.
Combinations of all the strategies used in the prior two sections
were studied in the trials with linear input command. Each
strategy was designated according to the percentage of overlap
and the percentage of proportional antagonist activity. Those
with 0% overlap represent the physiology based strategies, and
those with 0% antagonist gain are the strategies based on the
reports by Durfee and Chizeck et al. [6], [7], respectively.
A total of 25 trials with linear input command were performed in random order for each cat in order to test all possible
combinations of antagonist strategies. In the sinusoidal trials
only strategies resembling coactivation data of human subjects
and the strategies of Durfee [6] and Chizeck [7] were used,
resulting in 10 trials performed in random order.
E. Analysis
The data were normalized with respect to the SOL maximal
torque, dividing the torque value by the maximal torque
obtained from the SOL during the first trial without TA
activity. This procedure was used because attempts at using
both the TA and SOL maximal torques would result in shifts
of the neutral position. In most cases, the ratio of SOL to TA
maximal torque at the ankle joint was about 1.5, thus resulting
in consistent normalization of extension torque as well.
The analysis of the data was based first on the ability to
duplicate the pattern of the input signals without the benefit
of feedback, and secondly on the efficiency, or loss of net
torque due to antagonist co-contraction. Each muscle pair’s
ability to follow a straight line from maximum dorsiflexion to
maximum plantarflexion under each paradigm was measured.
The information was evaluated based on the correlation coefficient obtained from linear regression of the torque signal
versus time during the interval between the command sequence
initiation to termination. Further evaluation of the linear input
trials was based on the standard error from the regression
line. In sinusoidal contractions, the quality of the output signal
was evaluated through the calculation of harmonic distortion,
which is the amount of power in multiples of the base 0.4-Hz
frequency divided by the power at the base frequency.
Torque transmission efficiency was defined as the proportion
of torque in each trial transmitted to the joint relative to
the sum of torques from both muscles. This parameter was
calculated according to
/IT1 dt
E f f = /(lTal
+ ITtl) dt’
Where T is the net torque as a function of time in each
contraction, T, and Tt are the SOL and TA torque functions
of time without antagonist activity recorded during the initial
trials.
This efficiency €ormula compares the torque deviation from
neutral to the independent activation of SOL and TA with
0% antagonist and 0% overlap with no residual torque. Cocontraction and residual torque have deleterious effects on the
efficiency, because net joint torque is reduced in comparison
to the absence of antagonist activity.
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 43, NO. 2, FEBRUARY 1996
154
-
6
Experimental
3
0
3
6
0
3
2
1
4
5
6
Time (s)
Fig. 4. This graph illustrates the effect of simultaneously stimulating the TA
and SOL. The top and bottom traces are the results of trials where the SOL and
TA were stimulated with the electrode leads disconnected for the other muscle.
Positive torque represents plantarflexion, and negative represents dorsiflexion.
The dark middle trace was obtained by using the same stimulation sequence,
but with both muscles active. The dotted middle line was obtained by the
algebraic addition of the independent torque signals. The closeness of the
two middle lines indicates that the stimulation channels were independent (no
cross-talk).
111. RESULTS
A. Stimulator-Muscle Independence
The first task was to assess the independence of the two
muscles when acted upon by the stimulation system. The
first two experimental trials were performed by applying a
0% antagonist strategy with 0% overlap, disconnecting first
the TA and then the SOL. For the SOL, the torque pattern
initially was maximal, and decreased over a 3-s interval to
a minimum which was held for 3 s. For the TA, the initial
torque was minimal, and when the soleus torque reached its
minimum, the TA was activated to its maximum over a 3-s
period. The first two trials also were used for the denominator
in the efficiency calculation. The third trial was performed with
both muscles active and was then compared to the algebraic
sum of the first two trials. In every case the addition of the
first two trials essentially was identical to the third trial, as
shown in Fig. 4. Once the correct functioning of the system
was assured, experimental trials ensued.
B. Linear Input Command
The means of all normalized trial conditions with linear
input command are shown in Fig. 5 which displays the changes
in output waveform that occurred upon varying the antagonist
control strategies. The transition points where antagonist activity was elicited are apparent from the changes in the slopes
of the traces.
The results of the correlation analysis for these trials are
tabulated in Table I(a) and shown graphically in Fig. 6(a).
From the standpoint of the antagonist gain, the correlation
had a general decrease with increasing antagonist activity in
the vicinity of 0.98 for 0% antagonist gain and it decreased
to the vicinity of 0.94 at 30% antagonist gain. A general
increase in correlation with increased overlap was also noted.
The correlation coefficient is a measure of the deviation of
points from a fitted line in relation to their departure from
their mean. Because the trials with higher antagonist gain
had less peak torque and smaller departure from the neutral
torque position, the correlation coefficient tended to decrease.
In view of this, a standard error (SE) analysis was performed to
quantify each strategy’s ability to track a line. This analytical
parameter was less sensitive to changes in the regression slope
and more sensitive to deviation about a straight line; the results
are tabulated in Table I@) and shown in Fig. 6(b). Beyond
25% of overlap, the SE decreased from a peak of over 0.11
normalized torque units to as low as 0.05. A strong tendency
across different antagonist gains was not observed.
The results of torque transmission efficiency are tabulated
in Table I1 and shown in Fig. 7. As a function of overlap, an
optimum occurred near the middle of the overlap range, and
a systematic decrease of efficiency with increased antagonist
gain was noted. Peak torque transmission efficiency was 54%
with 0% antagonist gain and 50% overlap. The peak torque
decreased to 34% with 30% antagonist gain.
C. Sinusoidal Input Command
In the trials using a sinusoidal input, the emphasis was on the
effect of increasing the overlap and the percent of antagonist
activity. The effect on the raw signal of increasing the overlap
percentage is shown in Fig. 8(a). The most noticeable feature
is an inflection observed at 0% overlap that disappeared as the
overlap was increased. The effect of increasing the antagonist
activity is shown in Fig. 8(b). An increase in the prominence
of this inflection and a decrease in peak torque was noted with
increasing antagonist gain. The results of harmonic distortion
analysis are shown in Fig. 9(a) with respect to increasing
overlap and in Fig. 9(b) with respect to the antagonist activity.
With increasing overlap, the total harmonic distortion was
approximately 2%, with little systematic change across the
overlap range. In contrast, the antagonist gain had a detrimental effect on the sinusoidal output quality, increasing the
average harmonic distortion from 2.5% at 0% antagonist gain
to 6.6% at 30% antagonist gain.
IV. DISCUSSION
The main objective of this study was to assess the signal
tracking and torque transmission characteristics of various
co-contraction strategies for functional use in restorative neuroprosthetic systems using FES. Several findings arose: First,
despite small differences in correlation, standard error, and
harmonic distortion analyses, good signal tracking capability
could be obtained with any of the tested paradigms. Second,
increasing amounts of antagonist gain resulted in decreased
maximal torque and decreased torque transmission efficiency.
The signal tracking ability of the muscle pair was dependent
on the antagonist gain [as shown in Fig. 5(d) and (e)]. Fig. 5
clearly illustrates that increased antagonist gain (e.g., slope
155
ZHOU et al.: EVALUATION OF ISOMETRIC ANTAGONIST COACTIVATION STRATEGIES OF ELECTRICALLY STIMULATED MUSCLES
g 1.0
8
.*g 1.0
.
I
%
eI
F
a-
9
c)
0.5
s&
s&
r
i2
1
.Y
0.5
a -
r
0.0
0.0
1
.a
.
I
I
m
51 .g
g 'Ea 0.5
0.5
2:
9
e
a"
1.0
1.0
El
0
1
2
3
4
5
6
0
1
2
3
4
5
6
5
6
Time (s)
Time (s)
(a)
(b)
a
1.0
.s
8
e
gs
0.5
PF4
0.0
1
.a
CI
Ld
z" .s8a 0.5
1.0
0
1
2
3
4
5
0
6
1
2
3
4
Time (s)
Time (s)
(4
(c)
0
1
2
3
4
5
6
Time (s)
(e)
Fig. 5. Mean normalized torques for each condition tested. The graphs are organized by overlap, with (a)-(e) corresponding to 0'70, 25% 50%, 75%, and
100% overlap, respectively. Among the panels, inflections in the torque traces occur near the transition points where the antagonist is engaged or disengaged.
Within each panel, the relative loss of peak-to-peak torque with increasing antagonist gain also is apparent.
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 43, NO. 2, FEBRUARY 1996
156
TABLE I(a)
CORRELATION
COEFFICENTS
ANTAGONIST
GAIN
Overlap
0%
5%
10%
20%
30%
0%
25 %
50%
75%
100%
0.9741 f 0.0131
0.9705 f 0.0170
0.9767 k 0 0167
0.9874 f 0.0077
0.9878 f 0.0099
0.9755 f 0.0103
0.9689 f 0.0216
0.9740 f 0.0196
0.9850 f 0.0108
0.9858 f 0.0138
0.9754 f 0.01 10
0.9644 f 0.0239
0.9689 f 0.0207
0.9784 f 0.0165
0.9830 f 0.0219
0.9586 f 0.0273
0.9504 f 0.0266
0.9571 f 0.0316
0.9701 f 0.0263
0.9761 f 00289
0.9329 f 0.0525
0.9338 f 0.0369
0.9447 f 0.0406
Oi9634 f 0.0328
0.9657 f 0.0341
20%
30%
TABLE I@)
STANDARD ERROR
ANTAGONIST
GAIN
Overlap
0%
5%
0%
25 %
50%
75%
100%
0.0871 f 0.0197
0.11 11 f 0.0607
0.0938 f 0.0523
0.0635 f 0.0184
0.0517 f 0.0213
0.0824 f 0.0152
0.1060 f 0.0622
0.0922 f 0.0526
0.0644 f 0.0238
0.0512 f 0.0236
10%
0.0786
0.1051
0.0902
0.0666
0.0463
f 0.0160
f 0.0587
f 0.0477
f 0.0263
f 0.0199
0.0847
0.0994
0.0872
0.0637
0.0493
f 0.0171
f 0.0442
f 0.0492
f 0.0315
f 0.0194
0.0919 f 0.0227
0.0949 f 0.0364
0.0801 f 0.0432
0.0583 f 0.0287
0.0514 f 0.0180
TABLE II
TORQUE
TRANSMISSION E m m a
ANTAGONIST
GAIN
Overlap
0%
5%
10%
20%
30%
0%
25 %
50%
75%
100%
45.79 f 11.04
53.35 f 11.98
54.21 f 8.60
50.95 f 8.24
46.38 f 12.63
44.46 f 10.98
50.71 f 10.13
51.70 f 8.56
48.62 f 8.69
44.56 f 11.47
42.74 f 9.59
48.08 f 8.70
47.33 f 5.95
43.88 f 6.31
39.57 f 8.11
39.97 f 8.63
41.21 f 6.20
41.15 f 5.32
37.67 f 6.06
35.57 f 8.00
34.64 f 7.51
34.87 f 5.83
34.32 f 4.27
32.08 f 6.71
32.18 f 8.80
from 0-5% range, 0-10% range, etc.) yielded larger departure
from the desired linear output elicited by the linear input
for all overlap conditions. Larger antagonist gains resulted in
increased output nonlinearity at the beginning and end of each
movement and the consequent deterioration of signal tracking.
Larger antagonist gains also reduced the absolute maximal
torques available to the joint.
Physiological studies with normal human subjects revealed
that antagonist activity ranges from 2-10% of its maximal
activity when the same muscle acts as agonist [ 2 ] , [3], [SI,
and [14]. Only in rare cases did the antagonist activity exceed
10%. It is, therefore, not surprising that strategies with up to
10% antagonist gain (slope) yielded satisfactory results from
the signal tracking standpoint. Furthermore, 10% antagonist
gain was sufficient to stabilize the joint when subjected to
large agonist force [4], [5]. Under such conditions, applying
antagonist force in the range of 0-10% of its maximal for
0-100% force from the agonist is recommended. This provided
good input tracking capabilities and sufficient stabilizing force
to the joint. In addition, low antagonist forces improved the
efficiency of the joint as shown in Fig. 7 and described below.
The regression analysis displayed overall improvement in
tracking with increasing overlap and decreasing antagonist
gain. However, it is of note that the lowest mean correlation
coefficient was 0.93, which is acceptable for physiologic
applications. Inspection of the experimental traces of ramp
and sinusoidal tracking revealed that an inflection occurred
near the transition points where the SOL became inactive or
where the TA was activated. These inflections were notable
especially for the 25% and 50% overlap. They may have been
somewhat masked in the 75% overlap strategies because of
increased joint stiffness caused by larger muscle torques. At
the beginning and end of each contraction, the torque traces
also rounded off. In the 100% overlap trials, no transition
occurred and only the rounding effect was noticeable. In trials
with no overlap, both transitions occurred simultaneously.
Therefore, a marked inflection was not obvious. It is suspected
that if the agonist-antagonist pair was strongly unbalanced
from the maximal torque standpoint, a distinct transition
inflection would have occurred. The trends shown by the
correlation analysis with varying overlap also held true for
the standard error analysis.
The largest errors were generally found at 25% overlap,
and they diminished markedly with increasing overlap. The
trends with varying antagonist gain were not as clear from the
standard error as from the correlation analysis. The correlation
analysis showed a clear loss of output quality with antagonist
gain increase, while inspection of the traces and their mean did
not show such a marked deterioration of the signal. This apparent discrepancy may be explained by the mathematical basis
ZHOU et al.: EVALUATION OF ISOMETRIC ANTAGONIST COACTIVATION STRATEGIES OF ELECTRICALLY STIMULATED MUSCLES
157
60
c
c
98
96
94
-ma,
a,
1
E'U
/
c
.-0
.-U)
U)
Antagonist:
40
E
U)
0
n
t
50
0
0%
5%
10%
921
.--+--+
c
2
Ia,
I
3
3o
g
I-
90
20
25
0
50
75
100
0
25
50
75
100
Percent Overlap
Percent Overlap
(a)
0.12
I
I
Fig. 7. Torque transmission efficiency results. The mean torque transmission
efficiency is shown for each co-contraction strategy. Efficiency generally is
highest in the middle of the overlap range, whereas a definite trend to decrease
with increasing antagonist activity also is apparent.
with a given input command. In this analysis, although the
trends were not strictly monotonic, they showed less deviation
from the straight line with increasing antagonist activity.
Two main observations can be made from the torque transmission efficiency analysis: First, there is a clear trend of
decreasing efficiency with increasing antagonist gain; Second,
Antagonist:
an optimum level exists near the middle of the overlap range.
0 0%
The decrease of efficiency with increasing antagonist gain was
0
5%
expected, because the net joint torque decreased with increased
10%
antagonist countertorque. This result raises one of the most
difficult questions for the FES system designer: maximal
0
20%
torque efficiency (i.e., agonist activation only) is conducive to
0 30%
0.04
I
I
I
minimal fatigue, yet it may be accompanied by long-term joint
0
25
50
75
100 problems associated with muscle activity imbalance. Studies
exploring the co-contraction patterns in skilled workers and
Percent Overlap
in elite basketball or volleyball players as well as subjects
with normally active lifestyles showed decreased flexor co(b)
contraction
in the athlete group [ 2 ] , [ll]. The athletes were
Fig. 6. Signal quality analysis results. (a) The mean correlation coefficients
for each co-contraction strategy. Correlation coefficients generally increase as deemed to be at increased risk of ligamentous injury and were
a function of overlap and decrease with increasing antagonist gain. (b) Depicts placed on a hamstrings exercise regimen which resulted in
the mean standard error in each category. This analysis is less sensitive to
differences in slope and more sensitive to deviation from the straight line. The increased coactivation during extension. Thus, the amount of
strong trend of standard error is to decrease with increasing overlap while a physiologic coactivation can vary according to the level and
weak tendency is to decrease with increasing antagonist activity.
type of activity. It could be argued that in a neuroprosthetic
application without the benefit of reflexive protection mechof the correlation coefficient which compared the deviation of anisms [18], the level of coactivation should be relatively
a set of points from a regression line to the deviation of those high to minimize the possibility of joint pathology due to
points from their mean. Because the initial and final torques imbalanced muscle activity. From the overlap perspective, a
decreased with increasing antagonistic activity, the deviation strategy with overlap between 25 to 50% may be optimal.
The sinusoidal traces revealed an interesting feature of the
from the mean in these cases was less and led to smaller correlation coefficients. Standard error analysis was limited to the relationship between agonist and antagonists during bidirecdeviation from the regression line. This was more interesting tional movements. With 0% overlap, a noticeable inflection
from the control standpoint as it pointed to the expected error occurred during the transition from dorsal to plantar flexion,
.
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 43, NO. 2, FEBRUARY 1996
158
K
.-X0
6
a,
E
m
+K
3
m
E
0
S
.-X0
a,
3
!E
$2
0
n
6
!
7
2
,
I
3
4
5
6
4
5
6
"
(a)
I
2
3
Time (s)
0)
Fig. 8. Torque traces resulting from sinusoidal input command in one preparation. (a) Shows the variation of torque with changes in the overlap. No large
variation in signal quality occurs except for an inflection in the transition from dorsiflexion to plantarffexion at 0% overlap. (b) Shows the variation in signal
with increasing antagonist gain. The aforementioned inflection becomes more pronounced and the net torque decreases with increasing antagonist gain.
whereas no such inflection was found in the transition from
plantar to dorsal flexion. This probably was due to delay in
activating the soleus which exhibits predominantly slow twitch
characteristics [12]. The TA, with a predominantly fast twitch
population [12], did not have such a prolonged delay and
produced a smoother transition. This was probably aided by
slower deactivation of the soleus contractile mechanism and
the use of overlap eliminated this inflection. This suggests
that overlap may be used to achieve smoother transitions.
Otherwise, the quality of the sinusoids was not greatly affected
by overlap, as shown by the harmonic distortion analysis. The
antagonist gain, however, had a greater effect on the signal
by reducing the peak torque and increasing the harmonic
distortion. Since overlap was not used in any of the sinusoidal trials, the inflection during dorsiflexion-plantarflexion
transition was seen in all cases and became more marked
with increased antagonist gain. This was probably because
of the TA's increasing opposition role during the onset of
plantarflexion, which becomes more pronounced before the
start of SOL torque production.
In a practical application, the transition and stability requirements as well as the differences in strength and response speed
in any given joint may necessitate asymmetric co-contraction
strategies. For example, the knee stability-enhancing role of
the hamstrings is much more important than that of the
quadriceps [l], [4], and [SI,and the knee extensors usually
generate more torque than the flexors. Therefore, it would be
expected that the percent of hamstrings coactivation be higher
than that of the quadriceps in an FES knee control application.
Different overlap strategies also may be required to ensure
smooth movement transitions between fast and slow muscles.
Future work will be aimed at determining the influence of
coactivation strategies on signal tracking in the moving ankle,
and the long-term effects of coactivation on joint orthopaedic
integrity. Other roles of antagonistic activity such as compensating for load changes [16], [17] or maintaining isometric
joint stiffness [E]can also be envisioned in neuroprosthetic
applications.
In summary, it has been shown that agonist-antagonist coactivation control strategies can be implemented using principles
based on normal physiologic behavior in healthy humans in
combination with practical solutions for control problems in
FES applications. It is suggested that a strategy with 25 to SO%
overlap and antagonist gain in the range of 0 to 10% may pro-
ZHOU er al.: EVALUATION OF ISOMETRIC ANTAGONIST COACTIVATION STRATEGIES OF ELECTRICALLY STIMULATED MUSCLES
10
0% Antagonist
8
6
4
2
0
I
I
I
I
I
0
25
50
75
100
Overlap Percent
(a)
10
0 % Overlap
T
8
6
4
2
0
0
I
I
I
I
I
I
5
10
15
20
25
30
159
antagonist muscle controllers,” IEEE Trans. Biomed. Eng., vol. 36, pp.
309-321, 1989.
[7] H. J. Chizeck, N. Lan, L. S. Palmieri, and P. E. Crago, “Feedback
control of electrically stimulated muscle using simultaneous pulse width
and stimulus oeriod modulation,” ZEEE Trans. Biomed. Enn., vol. 38,
pp. 1224-1234, 1991.
r81
_ _ M. Solomonow, A. Guzzi, R. Baratta, H. Shoii, and R. D’Amhrosia,
“EMG-force model of the elbow antagonistic muscle pair: The effect
of ioint uosition, gravitv and recruitment,” Am. J. Phys. Med., vol. 65,
pp.- 223-244, 1986.
191. R. V. Baratta, M. Ichie, S.-K. Hwang, and M. Solomonow, “Orderly
.
stimulation of skeletal muscle motor units with tripolar nerve cuff
electrodes,” IEEE Trans. Biomed. Eng., vol. 36, pp. 836-843, 1989.
[lo] I. Fenyes, C. Gergely, and S. Toth, “Clinical and electromyographic
studies of spinal reflexes in premature and full term infants,” J. Neurol.,
Neurosurg. Psychiatry, vol. 23, pp. 6 3 4 8 , 1960.
[ 11] R. Person, “Electromyographic study of coordination of the activity of
human antagonist muscles in the process of developing motor habits,”
(Russian Text), J. Vys’cei Nervn. Dejut., vol. 8, pp. 17-27, 1958.
[12] M. A. Ariano, R. B. Armstrong, and V. R. Edgerton, “Hindlimb muscle
population of five mammals,” J. Histochem. Cytochem., vol. 21, pp.
51-55, 1973.
[I31 J. Sanchez, M. Solomonow, R. V. Baratta, and R. D’Ambrosia, “Control
strategies of the elbow antagonist muscle pair during two types of
increasing isometric contractions,” J. Electromyography Kinesiol., vol.
2, pp. 232-241, 1993.
[14] M. Solomonow, R. V. Baratta, B.-H. Zhou, and R D’Ambrosia, “EMG
coactivation pattems of the elbow antagonist muscles during slow
isokinetic movement,” Exp. Neurol., vol. 100, pp. 470-477, 1988.
[15] N. Hogan, “Adaptive control of mechanical impedance by coactivation
of antagonist muscles,” IEEE Trans. Automat. Contr., vol. 29, pp.
681-690, 1984.
[16] M. Latash, “Control of fast elbow movements: A study of electromyographic pattems during movements against unexpectedly decreased
inertial load,” Exp. Bruin Res., vol. 98, pp. 145-152, 1994.
[I71 M. Levin, A. Feldman, T Milner, and Y. Lamarre, “Reciprocal and
coactivation commands for fast wrist movements,” Exp. Brain Res.,
vol. 89, pp. 669477, 1992.
[18] M. Solomonow, R. V. Baratta, B. Zhou, H. Shoji, W. Bose, C. Beck, and
R. D’Ambrosia, “The synergistic action of the acl and knee muscles in
maintaining joint stability,” Amer. J. Sports Med., vol. 15, pp. 207-218,
1987.
[19] B. Zhou, R. V. Baratta, and M. Solomonow, “Manipulation of muscle
force with various firing rate and recruitment control strategies,” IEEE
Trans. Biomed. Eng., vol. 34, pp. 6-18, 1987.
Antagonist Gain (%)
(b)
Fig. 9. Total harmonic distortion analysis results. (a) Displays harmonic
distortion as a function of antagonist overlap and shows little variation. (b)
Depicts the harmonic distortion versus antagonist gain, indicating a gradual
loss of signal quality with increasing antagonist activity.
vide good signal tracking with maintenance of long-term joint
integrity without great loss of torque transmission e f f i c i e n c y .
REFERENCES
[I] K. N. An, S. Himeno, H. Tsumura, T. Kawait, and E. Chao, “Pressure
distribution on articular surfaces: application to joint stability evaluation,” J. Biomech., vol. 23, pp. 1013-1020, 1990.
[2] R. V. Baratta, M. Solomonow, B.-H. Zhou, G. Letson, R. Chuinard,
and R. D’Amhrosia, “Muscular coactivation: The role of the antagonist
musculature in maintaining knee stability,”Am. J. Sports Med., vol. 16,
pp. 113-122, 1988.
[3] S. Hagood, M. Solomonow, R. Baratta, B.-H. Zhou, and R. D’Ambrosia,
“The effect of joint velocity on the contribution of the antagonist
musculature to knee stiffness and laxity,” Am. J. Sports Med., vol. 18,
pp. 182-187, 1990.
[4] S. Hirokawa, M. Solomonow, Z. Luo, Y. Lu, and R. D’Ambrosia, “Muscular co-contraction and the control of knee stability,” J. Electromyog,
Kinesiol., vol. 1, pp. 199-208, 1991.
[5] S . Hirokawa, M. Solomonow, Y. Lu, R. V. Baratta, and R. D’Ambrosia,
“Anterior-posterior displacement of the tibia elicited by quadriceps
contraction,” Am. J. Sports Med., vol. 20, pp. 299-306, 1992.
[6] W. Durfee, “Task-based methods for evaluating electrically stimulated
Bing He Zhou (M’89) graduated in 1970 from the
Department of Electronic Engineering, University
of Science and Technology of China (USTC) in
Beijing, China.
From 1970 to 1978, he worked as an Electronics
Engineer at the Beipiao Broadcasting Station in
Liaoning Province. In 1978, he joined the faculty of
the Department of Electronic Engineering at USTC,
where he was an Associate Professor of Electronic
and Biomedical Engineering and the Vice Director
of the Institute of Biomedical Engineering. From
1985 to 1987, he was a Visiting Research Professor in the Bioengineering
Laboratory at Louisiana State University Medical Center (LSUMC) in New
Orleans, where he worked with the Laboratory staff on various studies related
to the analysis and control of the neuromuscular system, electromyography,
and instrumentation design. Currently, he is a Visiting Research Professor in
the BioengineeringLaboratory at LSUMC. His teaching and research interests
focus on analog and digital electronics, biomedical electronics, digital signal
processing, and microcomputerized medical instrumentation.
Dr. Zhou is a Committee Member of the International Union of Radio Science (USRI), the Commission of Electromagnetics in Biology and Medicine
(Commission K), and the Chinese Biomedical Electronic Society. He is also
a Senior Member of the Chinese Electronic Society, as well as a member of
the Chinese Biomedical Engineering Society, the Chinese Computer Society,
and the IEEEEngineering in Biology and Medicine Society. He received the
Zhang Zhongzhi Award for excellent teaching and research activities at USTC
in 1989, and first-place awards for most outstanding academic paper from
the Chinese Biomedical Electronic Society (1991) and the Anhui Biomedical
Engineering Society (1992).
160
B E E TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 43, NO. 2, FEBRUARY 1996
Richard V. Baratta (S’87-M’88) received the
B.S.E. degree (magna cum la&) in mathematics
and biomedical engineering (1984), the MSc.
degree in biomedical engineering (1986), and the
Ph.D. degree (1989) from Tulane University, New
Orleans, LA.
Since 1983, he has been affiliated with the
Bioengineering Laboratory at Louisiana State
University Medical Center, Baton Rouge, where he
presently serves as Assistant Professor and Director
of Rehabilitation Engineering. His major research
interests focus on the applications of numerical analysis, signal processing,
and systems engineering in the analysis and control of the neuromuscular
svstem.
Dr. Baratta is a member of Tau Beta Pi and Alpha Eta Mu Beta.
Giang T. Nguyen received the B.Sc. degree (magna
cum laude) in biomedical engineering in 1990 from
Tulane University, New Orleans, LA. He is a second year medical school student at Louisiana State
University Medical School, New Orleans, LA. He
has been involved in research in the bioengineering
laboratory at LSU Medical Center from 1992 to
1995. His research interests include neuromuscular
physiology and orthopedic biomechanics.
Robert D. D’Amhrosia received the M.D. degree
from the University of Pittsburgh, Pittsburgh, PA,
in 1964.
Following a one-year internship at the University
of Colorado, Denver, he served in Southeast Asia
Moshe Solomonow (M’79-SM’83) received the
from 1965 to 1967 as a Flight Surgeon and ComPh.D. degree in engineering systems and neuromander of the 8th TAC Hospital. After his residency
science from the University of Caliiomia, Los AItraining at the University of Pittsburgh, he became
geles.
affiliated with the University of California at Davis
In 1983, following faculty appointments at the
as an Associate Professor. He became Professor
University of California and Tulane University,New
and Chief of Orthopaedic Surgery at Lousiana State
Orleans, LA, he joined the Department of Orthopaedics at Louisiana State University Medical University Medical Center and affiliated hospitals, New Orleans, in 1976. He
was also a Consultant for the NCAA Basketball Tournament (1987-1993).
Center, New Orleans, where he is currently Professor of orthopaedic Surgery, Physiology, and Bio- His research interests focus on sports medicine and biomechanics as related
to joint and muscular deficiencies.
physics, as well as the Director of Bioengineering.
Dr. D’Ambrosia is the Editor of the journals Orthopedics and Orthopedics
He is also Director of the Paraplegic Locomotion Section at the Rehabilitation
Institute of New Orleans. He has served as a Consultant for the NIH, International, and he is a member of the American Academy of Orthopedic
NSF, Veteraus-Administration, Louisiana Department of Health and Human Surgeons’ Board of Directors, the Medical Director of LSU Universitf
Resources. The Louisiana Center for Cerebral Palsy, and numerous industrial Hospital’s Department of Rehabilitation, a member of AOA and AAOS, and
firms. His research interests focus on the biomechanics of movement and a Past President of both the Academic Orthopedic Society and the Louisiana
Orthopedic Association. He is also listed in The Best Doctors in America.
sensory-motor control of the extremities.
Dr. Solomonow is the Editor-in-Chief of 7he Joumal of Electromyography and Kinesiology. He has also served as a member of the EEEON
EMBS AD COM and as an Associate Editor of the B E E TRANSACTIONS
BIOMEDICAL
ENGINEERING.
He received the Crump Award for Excellence in
Biomedical Engineering at UCLA in 1977.
Leslie Joseph Oliver, I11 is working toward the
B.S. degree in mechanical engineering at Rice UNversity, Houston, TX.
Mr. Oliver is a member of the American Society
of Mechanical Engineers, as well as Secretary of
the Rice chapter of the Tau Beta Pi engineering
honors society. His interests include programming
and systems engineering.