Clinical Biomechanics 20 (2005) 822–833

www.elsevier.com/locate/clinbiomech

An algorithm for estimation of shoulder muscle forces

for clinical use
Philippe Favre *, Ralph Sheikh, Sandro F. Fucentese, Hilaire A.C. Jacob
Laboratory for Orthopaedic Research, Department of Orthopaedics, Balgrist, University of Zurich, Forchstrasse 340, Zurich 8008, Switzerland

Received 21 October 2004; accepted 20 April 2005

Abstract

Background. The shoulder joint represents an indeterminate mechanical system, making it difficult to predict individual muscle
forces required to equilibrate a given arbitrary external force. Although considerable work has been published on this matter, no
model exhibits the adaptability required for the analysis involving different positions of the humerus and for any external load.
An algorithm involving decision-making loops is developed to predict forces exerted by muscles that cross the shoulder joint in
equilibrating a given external force acting in an arbitrary direction, with the humerus in any one of 12 selected positions.
Methods. Muscle lever arms and directions of action collected from a full-size epoxy model of the shoulder joint are used together
with the external force as input. The algorithm selects an appropriate group of muscles and step by step attributes small force incre-
ments to withstand the external moment while aiming at minimising the forces involved. Each muscle force increment is stored after
every loop and eventually summed up. Stability of the glenohumeral joint is the final determining factor.
Findings. Six worked-out examples show interesting features of probable muscular activity. Muscle segmentation is of paramount
importance for spatial control. Although stability can be achieved by increasing the overall rotator cuff activity (co-contraction), this
is rarely necessary.
Interpretation. The strategy of force sharing among the muscles opens up the possibility to examine the outcome of muscle defi-
ciencies and to investigate causes of joint instability as encountered in clinical practice. Further validation of the model is still
needed, but certain clinical observations can be explained.

2005 Elsevier Ltd. All rights reserved.

Keywords: Shoulder model; Shoulder muscle forces; Muscle recruitment; Glenohumeral stability; Moment arm lengths; Direction of muscular force

1. Introduction 1990). In addition to particular muscles called upon to

It has long since been recognised that stability of the
shoulder joint must primarily be effected through mus-
cular action because the relevant joint capsule and liga-
ments are generally lax throughout the greater part of
glenohumeral rotation, and hence inactive, while the
joint surfaces themselves offer but little resistance to dis-
location (Jössel, 1880; Perthes, 1906; Gerber, 1992; Lip-
pitt et al., 1993; MacMahon et al., 1995; Kronberg et al.,
*
Corresponding author.
E-mail address: pfavre@research.balgrist.ch (P. Favre).
equilibrate an external force, shoulder joint stability
might require additional augmentation by all the mus-
cles of the rotator cuff acting together to provide what
is now commonly termed Ôconcavity compressionÕ (Lip-
pitt et al., 1993). It therefore becomes important to
understand the strategy adopted by muscles in balancing
a given external force, while preventing dislocation of
the glenohumeral joint. The complexity of the shoulder
joint, with its extensive range of motion and large num-
ber of muscles, calls for close examination of possible
muscular strategies involved, to understand some of
the disorders of this intricate articulation and to treat
these more effectively.

0268-0033/$ - see front matter
2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.clinbiomech.2005.04.007
 P. Favre et al. / Clinical Biomechanics 20 (2005) 822–833
Analysis of joint mechanics necessitates prediction of
muscular load sharing, and as long as forces within the
living muscles, or at least across the joints, have not
been directly measured while performing a given task,
any estimate of muscle force would depend on the bio-
mechanical model chosen. Invariably, one is confronted
with the problem of static indeterminacy: there are more
unknown muscle forces than equilibrium equations.
Even electromyographic (EMG) recordings fail to pro-
duce reliable force estimates because the relationship be-
tween force and EMG activity is dependent on several
unpredictable variables (Inman et al., 1952; De Luca
and Forrest, 1973; MacMahon et al., 1995; Van der
Helm, 1994). Several investigators, from the second half
of the 19th century up to recently (Fick, 1850; Fick,
1910; Pauwels, 1965; Murray et al., 2000), have used
the cross-sectional area of a muscle as being indicative
of the proportion of load taken by it in comparison to
synergists that assist in performing a common task.
Much work has already been done towards estimat-
ing the shoulder muscle forces. Many optimisation prin-
ciples have been employed to compute the load sharing
between the muscles that cross a joint. One of the first
was that the sum of all attributed muscle forces should
be a minimum (Seireg and Arvikar, 1975). Optimisation
principles of minimum energy consumption and mini-
mum muscle stress have also been implemented (Hardt,
1978; Crowninshield and Brand, 1981). Other criteria
that have been used consider a time-dependent function
related to muscular stress-endurance (Niemi et al., 1996)
or minimisation of muscular energy consumption (met-
abolic cost) during fast goal directed movements (Hap-
pee and Van der Helm, 1995).
The work published by Karlsson and Peterson (1992);
Van der Helm (1994); Happee and Van der Helm (1995)
deserve special mention. Laursen et al. developed an
EMG-based model. Others, like Johnson et al. (1996),
have contributed morphometric and coordinate data
that are indispensable for the estimation of muscle
forces about the shoulder joint. A useful biomechanical
model of the shoulder complex must realistically repre-
sent the lines of action of muscles for all three-dimen-
sional positions of the arm (Otis et al., 1994; Johnson
et al., 1996; Wickham and Brown, 1998), and although
some information on the Cartesian coordinates of sev-
eral muscle insertions about the shoulder joint have be-
come available, those of the two-joint muscles are not
included, and the problem of ‘‘muscle wrapping’’ in
many positions of the arm is not readily solved.
Most of the above mentioned computational meth-
ods have unfortunately been tailored to meet special
cases of loading (usually to balance an external moment
acting in one principal direction) and therefore do not
readily lend themselves for application in a clinical envi-
ronment where one wishes to test, for instance, the effect
of a neurological or muscular deficiency on a model in
823
which an arbitrary combination of external loading con-
ditions would apply. We therefore developed an algo-
rithm to predict muscle forces and test joint stability
from data measured off a full-size epoxy model of the
thorax, scapula and arm.
2. Method
2.1. Definitions
2.1.1. The Coordinate system
A Cartesian coordinate system referenced to the scap-
ula is used to define the line of action of forces, and the
resulting turning moments. It is defined as follows, for
the right shoulder:
• The origin is the centre of rotation of the humeral
head.
• The X-axis lies in the scapular plane and is directed
from medial to lateral.
• The Z-axis is directed caudal to cranial.
• The Y-axis follows the right-hand rule (directed
anterior).
2.1.2. Motion components, position, and directional signs
Each motion is decomposed into its elevation, flexion
and rotation components relative to the scapula, which
for the present purpose is considered to remain in a fixed
neutral position. ÔPositive elevationÕ denotes abduction,
Ônegative elevationÕ adduction; Ôpositive flexionÕ is ante-
rior flexion, Ônegative flexionÕ posterior flexion or exten-
sion; Ôpositive rotationÕ implies medial or internal
rotation, and Ônegative rotationÕ lateral or external rota-
tion. This terminology relates to a spherical system (in-
stead of a Cartesian) centred at the glenohumeral joint,
the loci of any point on the humeral axis within its total
range of motion being a globe. With the axis of this
globe running in a cranio-caudal direction, degrees of
longitude would be a measure of flexion, while degrees
of latitude express elevation. Thus, ambiguity of posi-
tion that could arise through confusion of sequence with
Eulerian angles (CodmanÕs paradox) as in a Cartesian
system is prevented.
2.1.3. ‘Given’ external moment and ‘virtual’ external
moment
The given external moment is the moment induced by
the given external force or external pure torque. The
moment is named according to the direction of move-
ment it would initiate; e.g. an abduction moment causes
abduction, etc.
The virtual external moment is defined as the differ-
ence (error) between the given external moment and
the partially exerted muscle moment, as determined at
824 P. Favre et al. / Clinical Biomechanics 20 (2005) 822–833

any intermediate step of the iteration procedure. It is Table 1

therefore the remaining external moment that the mus- Physiological cross-sectional area (mm2) and maximal force (N) for the
27 muscles segments
cles still have to balance. At the end of the algorithm,
the sum of all iterate muscle moments opposes in direc- Muscle segments PCSA Max force
tion and equals in magnitude the given external moment 1 SSP1 260 182
within the range of the chosen acceptable error. 2 SSP2 260 182
3 SSC1 450 315
4 SSC2 450 315
2.1.4. Protagonists and antagonists 5 SSC3 450 315
Protagonists are those muscles which exert a mo- 6 ISP1 320 224
ment that counteracts a given component of the exter- 7 ISP2 320 224
nal moment; e.g. the supraspinatus muscle would be a 8 ISP3 320 224
9 TMIN1 100 70
protagonist in equilibrating an external adduction (neg- 10 TMIN2 100 70
ative elevation) moment. Antagonists are muscles which 11 DEL1 860 602
exert a moment in the same sense as that of the external 12 DEL2 290 203
moment component. Any muscle having three compo- 13 DEL3 290 203
nents of action can well be a protagonist for one 14 DEL4 290 203
15 DEL5 430 301
external moment component, but an antagonist for 16 DEL6 430 301
another! 17 TRI 680 476
18 COR 250 175
2.1.5. Muscle segments 19 TMAJ 1000 700
The muscles of the shoulder have been segmented to 20 LAT1 290 203
21 LAT2 290 203
allow for differentiated actions. Some were divided into 22 LAT3 290 203
two, others into three parts. This was especially neces- 23 PEC1 460 322
sary in muscles with broad insertions because moment 24 PEC2 460 322
arm length and direction of tendon action vary consid- 25 PEC3 460 322
erably within a given muscle. Moreover, neuromotor 26 BI CL 310 217
27 BI CB 320 224
control strategies differ across the breadth of muscle
and all fibres in a given muscle must not necessarily con- SSP = supraspinatus, where SSP1 is the anterior part and SSP2 the
posterior.
tract simultaneously (Fick and Weber, 1877; Van der SSC = subscapularis, where SSC1 is the most cranial part, SSC3 the
Helm and Veebaas, 1991; Johnson et al., 1996; Wick- most caudal.
ham and Brown, 1998). Table 1 lists the manner in ISP = infraspinatus, where ISP1 is the most cranial part, ISP3 the most
which the muscles were segmented and also gives values caudal.
of the cross-sectional areas that were used. TMIN = teres minor, where TMIN1 is the cranial part, TMIN2 the
caudal.
DEL = deltoideus, where part 1 is the anterior, 2–4 the middle, and 5–6
2.2. Conditions to be fulfilled the posterior parts.
TRI = triceps.
2.2.1. Force boundaries COR = coracobrachialis.
The maximal force (see Table 1) a muscle can develop TMAJ = teres major.
LAT = latissimus dorsi, where LAT1 is the most cranial part and
is limited by several parameters such as its optimum LAT3 the most caudal.
length, the contraction velocity, the fibre type composi- PEC = pectoralis major, where PEC1 is the most cranial part and
tion and especially the physiological cross-sectional area PEC3 the most caudal.
(PCSA) (Fick, 1910; Karlsson and Peterson, 1992). In BICL = biceps caput longum.
this study, an average of the PCSA values found in BICB = biceps caput breve.
Wood et al. (1989), Veeger et al. (1991), and Van der
Helm (1994) has been used mainly to determine the tary virtual external moment (without overshooting)
maximum muscle force that could possibly occur, this so that the error always decreases.
being k* PCSA, in which k is a constant factor of 0.7
MN m
2 as suggested by Wood et al. (1989). 2.2.3. Stability
Moreover, since a muscle can only pull, the force it Once all muscle forces have been obtained, the result-
exerts can only be positive or null. ing joint force (joint resultant) is calculated and its spa-
tial direction computed. If the intersection of this vector
2.2.2. Convergence with the glenoid surface does not lie within the glenoid
The convergence of the algorithm is ensured as long boundaries, the articulation will dislocate. In this case,
as the virtual external moment becomes smaller after stability is gained by increasing the general rotator cuff
each iteration loop. The overall internal moment is muscle force (the so-called Ôconcavity compressionÕ).
thereby so controlled as to always oppose the momen- This entails recruiting such muscles that might, in them-
 P. Favre et al. / Clinical Biomechanics 20 (2005) 822–833
selves, be primarily antagonistic to the goal of counter-
acting the given external moment. Consequently, in such
cases more protagonistic activity and therefore more
overall muscular energy would be required.
2.3. Lever arms and muscle direction measurements
Basically, to approach the problem of determining
muscle forces which would equilibrate a given external
force acting on the humerus in any chosen position,
knowledge of the spatial direction of the muscles that
cross the glenohumeral joint and the length of their mo-
ment arms in that particular position of the humerus are
required. Furthermore, because each muscle can possi-
bly give rise to rotation components about three orthog-
onal axes simultaneously (for instance, it could abduct,
posteriorly flex, and medially rotate the humerus at
the same time), it becomes necessary to describe the cor-
responding moment arm lengths for each of these three
actions, separately.
A realistic three-dimensional model of the thorax,
scapula and proximal humerus was built, based on a
fresh cadaver adult specimen with no known shoulder-
related pathology and which showed no disorders of
the glenohumeral joint on macroscopic examination.
The fresh specimen was freed of all muscles and capsule,
taking particular care to ensure the labrum and the car-
tilaginous joint surfaces remained intact. During dissec-
tion drawings and photographs were made to document
the sites of muscle origin and insertion. Silicone rubber
moulds of the humerus and scapula were made in order
to cast epoxy resin replicas of them. In this manner
dimensionally true epoxy models of the scapula (with
Fig. 1. (a, b) The epoxy shoulder model used to determine muscle lever arms and directions of muscle action.
825
labrum) and humerus resulted. This enabled us to deter-
mine lever arm lengths (by the tendon travel method
(Fick and Weber, 1877; Shiino, 1913; Otis et al.,
1994)) and direction of muscle action (by means of
markers on the simulated tendons with the optical Mo-
tion Analysis system) without being confronted with
problems related to deterioration of biological tissues
with time. The radius of curvature of the humeral head
measured 23 mm. Also, tendons were simulated by
braided cords of thickness that corresponded with the
actually measured thickness of the tendons as encoun-
tered during dissection, thereby ensuring the determined
moment arm lengths of the corresponding muscles ap-
proach the physiological values that might be expected
through Ôwrapping aroundÕ bone protrusions (Fig. 1a
and b). For muscles that cross the joint with deep bellies,
in particular the deltoid, the cord used in the model
would represent the action of muscle fibres closest to
the joint. The moment arms obtained for segments
of the deltoid are therefore minimal values and represent
the worst possible conditions. The cords used for the
subscapularis and infraspinatus muscles were of 4 mm
diameter, those for the teres major 3 mm diameter, for
the deltoid, triceps, biceps caput longum and supraspi-
natus 2.5 mm diameter, for coracobrachialis, teres min-
or and biceps caput breve 2 mm diameter and for
pectoralis major and latissimus dorsi 1.5 mm diameter.
Points of insertion on the humerus were determined by
inspection during dissection and by the clearly defined
narrow areas of insertion detectable on the epoxy cast-
ing (Fig. 1b). Points of origin were chosen along the last
line of contact between muscle and bone, raised above
the bone surface by about 6 mm (see eyelets in
826 P. Favre et al. / Clinical Biomechanics 20 (2005) 822–833

Table 2 Table 3
The 12 investigated positions of the humerus with respect to the Lever arm in mm, and unit vectors of tendon direction in which muscle
scapula forces act on the humerus for the 27 muscle segments; humerus in 30
Position Elevation () Flexion ()a Rotation ()b elevation in the scapular plane, in neutral rotation
Elevation Flexion Rotation X Y Z
1 30
30 0
2 60
30 0 SSP1 22.27 5.47 5.48
0.98
0.14
0.16
3 80
30 0 SSP2 23.07 0.00
6.25
0.98
0.14
0.16
4 0 0 0 SSC1 5.90 8.98 15.56
0.95
0.30
0.07
5 30 0 0 SSC2
1.77 5.64 21.35
0.93
0.34
0.14
6 60 0 0 SSC3
12.77 1.43 16.40
0.95
0.25
0.21
7 80 0 0 ISP1 14.46 0.00
15.90
0.96
0.03
0.29
8 30 30 0 ISP2 7.47
3.01
19.55
0.94
0.08
0.32
9 60 30 0 ISP3 0.00
3.01
21.47
0.93 0.07
0.35
10 80 30 0 TMIN1
12.87 0.00
17.21
0.91
0.28
0.31
11 30 0
60 TMIN2
13.80
1.45
12.85
0.91
0.28
0.31
12 60 0
60 DEL1 4.85 25.82 0.00
0.50 0.10 0.86
a
Positive values are anterior to the scapular plane, negative values DEL2 22.90 12.84 0.00
0.79
0.12 0.60
posterior. DEL3 29.11
7.13 0.00
0.80
0.07 0.60
b
Negative values denote lateral rotation. DEL4 10.25
26.0 0.00
0.50
0.21 0.84
DEL5
11.00
31.46 0.00
0.60
0.23 0.77
DEL6
28.60
25.88 0.00
0.68
0.12 0.72
Fig. 1a). Broad flat muscles were segmented in 2–3 parts TRI
32.55
11.48 0.00
0.60
0.03
0.80
and the centreline of each segment was regarded as the COR
3.40 28.66 0.00
0.52 0.17 0.84
line of action. Finally, by virtue of the ÔlabrumÕ, the TMAJ
41.55
12.81 0.00
0.92
0.29
0.26
LAT1
28.22
5.73 6.95
0.87
0.18
0.45
humeral head was centred to the extent that no transla- LAT2
52.63
5.73 7.86
0.66
0.02
0.75
tory motion in any direction could be detected while the LAT3
66.51
5.73 5.09
0.42 0.07
0.90
humerus was rotated within the glenoid cavity, thus PEC1
8.69 28.68 6.16
0.76 0.45 0.47
eliminating possible errors from this source when deter- PEC2
17.20 45.81 7.55
0.59 0.75 0.30
mining moment arm lengths. Data were collected for 12 PEC3
34.40 63.00 12.25
0.32 0.89
0.33
BI CL 17.95 17.27 5.70
0.80
0.60
0.02
discrete positions of the humerus with respect to the BI CB
2.85 31.54 0.00
0.51 0.02 0.86
scapula, as shown in Table 2. In each of the 30, 60
and 80 levels of elevation, three positions of flexion (9
positions), and in 0 elevation only one position of flex- finally fulfil several boundary conditions, some of which
ion (humerus in the scapular plane). Because rotation are anatomically and physiologically set.
was maintained in the neutral position in all the afore- The algorithm for shoulder forces estimation (ASFE)
mentioned cases, two additional cases were investigated involves decision-making iteration loops (see Fig. 2). In
to explore at least two positions in the important act of the main loop, an appropriate group of muscles is first
throwing, or pitching, in which lateral rotation of 60 chosen solely depending on the sign (direction) of the
was chosen in 30 and 60 of elevation in the plane of moment components they could possibly exert, for a
the scapula. Changes in position of the scapula relative given position of the humerus. A small arbitrary force
to the thorax have not been taken into consideration is initially attributed to this set of muscles in proportion
for the present purpose but this would only involve to their cross-sectional area. This approximation is nec-
the pectoralis major and latissimus dorsi muscles. The essary to only start the process of iteration, i.e. to express
corresponding values given in the tables can, however, a first set of equations (one for each of the three internal
be readily corrected once the position of the scapula rel- orthogonal moment components) which, after further
ative to the thorax has been defined. One example of the modification through sub-routines to suit specific bound-
muscle lever arms and directions of action can be seen in ary conditions (admission of only positive muscle forces,
Table 3, the remaining cases can be found in the elec- re-arrangement of individual muscle contributions to
tronic version of this article in Tables A–L. meet directional requisites and to minimise magnitude
of forces involved) leads to an iterate that is a better
2.4. The algorithm approximation of the muscle forces required. Also, by
choosing only small increments of muscle force in each
2.4.1. Synopsis of the algorithm loop and, if necessary by even clipping an already esti-
Because each muscle commonly gives rise to turning mated muscle force, it is ensured that at no stage the
moments in all three orthogonal planes simultaneously, muscular moment overshoots the external one. At any
choice of muscles and the magnitude of force demanded intermediate stage, although the given external moment
from each to equilibrate a given external moment be- components are still not balanced, the amount of imbal-
comes an elaborate issue. Solving the problem therefore ance has decreased. By treating the issuing imbalance,
requires a system of arithmetical operations that have to also referred to as ÔerrorÕ as a new set of external moment
 P. Favre et al. / Clinical Biomechanics 20 (2005) 822–833 827

lated in magnitude and direction (positive or negative)

External force and
humerus position with reference to the scapular Cartesian system.
(Input)
2.4.4. The main iteration loop
External moment
Suitable muscle segments are selected, equilibrium
Loop calculation equations solved, and forces attributed to these, as
follows:
Muscle recruitment
2.4.4.1. Muscle recruitment. The main criterion for mus-
F(i) is reduced
cle recruitment is the direction of muscle action. How-
Equilibrium equation
for antagonists ever, since each muscle segment generally exerts
moments in all three orthogonal planes simultaneously,
Average forces
selection of essential muscle segments requires grouping
If there exist i into two categories, as follows:
such that F(i) < 0 Force redistribution
If for all i,
F(i) > 0 (A) Muscle segments that can exert a significant
Force limitation moment to oppose the greatest of the three exter-
nal moment components, while totally disregard-
F(i) is reduced Internal moment ing the two lesser ones.
for antagonists calculation (B) Muscle segments that can exert moments which
Condition 2 oppose all three external components
Error calculation simultaneously.
Condition 1

Virtual moment From category A all muscles with a high potential

If Error reaches
= (large moment magnitude) to exert a protagonistic func-
acceptable limit
Error
Joint resultant
force
If not stable, ground
level activation is
Joint stability
introduced / raised
Solution
Fig. 2. Structure of algorithm. Condition 1: One or two internal
moment components do not oppose the corresponding virtual moment
components. Condition 2: All three internal moment components
oppose the corresponding virtual moment components.
components, called virtual external moment compo-
nents, the process described above is repeated until the
issuing error converges to an acceptable level. Each thus
determined muscle force is stored after every loop and
finally summed up to give the final estimate.
2.4.2. Input
The input consists of the given external force and the
humerus position. The external force is defined by its
point of application and directional components within
the local Cartesian coordinate system. A choice of one
of the 12 discrete humerus positions is presently
possible.
2.4.3. External moment calculation
From the input data, components of the given exter-
nal moment for elevation, flexion and rotation are calcu-
tion in opposing the greatest component of the external
moment are primarily chosen. Such muscle segments are
chosen when their individual maximum muscle moment
in the direction of interest is greater than the average of
all maximum muscle moments in the same direction
multiplied by a threshold factor. If the threshold value
be high, few muscles with high potential would be re-
cruited to counteract the external moment, whereas
low values lead to several muscles of lower potential
sharing the task, resulting in finer tuning of the muscular
moment.
Since it would be logical to assume that neuromotor
control strategy is such that muscles are loaded to the
least extent in performing a given task, it follows that
the issuing magnitude of the joint resultant would also
be as low as possible. Tests have shown that a minimum
joint resultant force issues with a threshold factor be-
tween 0.2 and 1.4 and therefore, the optimal threshold
factor which leads to the lowest value of resultant force
is sought by iteration.
To category B belong those muscles that do not exhi-
bit a sufficiently large moment potential in the major
direction, but which nevertheless favour convergence
in all three directions simultaneously. This category
has been found crucial to arrive at an acceptable final
solution by virtue of the larger number of muscles (or
muscle segments) available. Although these have little
moment potential to equilibrate the major virtual exter-
nal moment component they are indispensable for equil-
ibrating the remaining two components of external
moment.
828 P. Favre et al. / Clinical Biomechanics 20 (2005) 822–833
The strategy used to solve the problem is therefore to
first focus on the greatest external moment component
which calls for a choice of muscles that strongly oppose
this in the first place and to then select other muscles
that will counteract the two remaining moment compo-
nents, too.
If the three given external moment components be of
the same amplitude, the programme will continuously
switch its focus from one component to another, ensur-
ing convergence of all three components alternately.
2.4.4.2. Equilibrium equations. To obtain equilibrium a
system of three equations, each for moments in one of
the three orthogonal planes, must be satisfied. For each
plane we have:
X greatest virtual external moment component. But, if
ðinternal momentsÞ þ ðvirtual external momentÞ ¼ 0
where
X X
ðinternal momentsÞ ¼ ðF i  leveri Þ
Fi being the force exerted by the muscle segment ÔiÕ
and leveri its lever arm, where i = 1,2,3, . . . , 27. The
27Fi are the unknowns of the equation.
To overcome the problem of indeterminacy we first
make use of the physiological cross-sectional area,
PCSA (Table 1). Assuming that all protagonists would
exert force simultaneously in proportion to their PCSA,
one muscle ÔuÕ is chosen and all other muscle forces re-
lated to it. The PCSA values link together all muscle
forces, and thus decrease the number of unknowns from
27 to 1:
Qi ¼ PCSAi =PCSAu
where Qi is the quotient between the PCSA of the muscle
ÔiÕ and the PCSA of the chosen reference muscle
P ÔuÕ.
The sum of all internal muscle moments, (Fi * le-
veri), can then be written as:
X attributed muscle forces during the first few loops can
ðinternal momentsÞ ¼ F u  ½Q1  lever1 þ Q2
 lever2 þ    þ 1  leveru
þ    þ Q27  lever27 
Inserting Eq. (4) in Eq. (1) and redistributing, one
equation containing only one single unknown Fu for
each of the three orthogonal planes is thus obtained.
From the hypothesis that all active muscles develop a
force proportional to their PCSA, used as a first approx-
imation towards a solution, each muscle force Fi then
follows from Eq. (5):
F i ¼ Qi  F u
The above procedure delivers three values of force for
each muscle, one for each plane!
2.4.4.3. Force averaging. The three different results for
every muscle segment arrived at above are dealt with
by attributing the average of the three values to the cor-
responding muscle. This constitutes a compromise that
would not satisfy any of the single given equations but
is the best approximation on the whole.
2.4.4.4. Force redistribution for prevention of negative
force values and antagonistic action. Negative muscle
force values that issue from the equations are not real,
although muscle moments can still be positive or nega-
tive. Given a first choice of muscle forces Fi, the muscles
of category A may introduce undesirable antagonistic
effects in the other two directions. Indeed, the total mo-
ment of all selected muscles will always oppose the
the antagonists exert together a greater moment than
ð1Þ that of the protagonists in at least one of the two
remaining directions, this would also lead to an increase
of misbalance in the corresponding direction. In the case
of such antagonistic action, the solution of the equations
ð2Þ may even lead to negative muscle force values. This is
dealt with as follows:
The protagonists must in any case exert a greater
moment than that of the antagonists. To do so, an in-
ner loop is inserted before the equilibrium equations in
the next iteration step to redistribute the forces that
were first selected on the basis of PCSA, giving less
weight to the antagonists until the moment exerted by
the protagonists exceeds that of the antagonists. In this
way, solely positive muscle force values are obtained
and overall protagonistic action is ensured in all three
directions.
ð3Þ 2.4.4.5. Prevention of overshooting. Because only the
signs of the muscle moment components are controlled
to oppose that of the virtual external moment compo-
nents and no special considerations have been made to
regulate the amplitude of the muscle components, the
reach excessively high values when the virtual external
moment is high and might result in overshooting the gi-
ven external moment. This can become problematic for
ð4Þ convergence and also represents a waste of energy by
demanding over-activity from muscles that are less opti-
mal. To overcome this problem, all forces are multiplied
in each loop with a constant shrinking factor of 0.05 thus
making the incremental steps very small, which leads to
a smoother iteration process, as shown in Fig. 3.
2.4.4.6. Force limitation. If the sum of iterative contri-
butions of any single muscle reaches its maximal force
value, k* PCSA, where k = 0.7 MN m
2, the muscle is
ð5Þ
turned off for the rest of the simulation. Other muscles
that have not reached their limit must take over to resist
the virtual external moment.
 P. Favre et al. / Clinical Biomechanics 20 (2005) 822–833
Without force increment reduction
Muscle moment amplitude
Iteration steps
Fig. 3. Effect of force increment reduction through a Ôshrinking factorÕ to prevent overshooting.
2.4.4.7. Error calculation. At the end of each loop, the
overall three orthogonal muscle moment components
have been calculated and the difference between these
and the corresponding virtual external moment compo-
nents determined. These differences (error1, error2, er-
ror3) are an indication of the remaining amount of
imbalance. The total error is quantified by an amplitude
as follows:
2 2 2 1=2
amplitude ¼ ½ðerror1Þ þ ðerror2Þ þ ðerror3Þ 
The closer the muscle and external moments ap-
proach each other, the smaller is this amplitude. The
program stops when the amplitude of the error reaches
a chosen set value.
2.4.5. Resulting force and rotator cuff ground level
activation for joint stability
Once the main loop is completed and all final individ-
ual muscle forces have been determined, the resulting
joint force is calculated by adding all muscle forces plus
the external force and is checked to fall within the gle-
noid boundaries. Minimisation of muscle forces finally
precipitates in a minimisation of the joint resultant
amplitude.
The glenohumeral joint will not dislocate as long as
the humeral joint reaction force (the resultant) acts with-
in the boundaries of the glenoid. This can be enforced
through co-contraction of the rotator cuff muscles (con-
cavity compression) (Van der Helm, 1994; Happee and
Van der Helm, 1995). This superimposed force exerted
simultaneously by all the rotator cuff muscles (supraspi-
natus, infraspinatus, subscapularis and teres minor) is
referred to as the ground level activation and is in some
cases indispensable for joint stability. If stability is not
reached at the end of the process described before,
ground level activation is introduced, or raised, and
the algorithm restarted. However, a minimal ground
level activation force is sought in order to minimise
the required muscular energy. In this study, only active
stabilisation is taken into consideration and passive
mechanisms such as adhesion/cohesion, ligamentous
stabilisation or suction produced under the action of a
destabilising force, etc. are not taken into account.
829
With force increment reduction
Muscle moment amplitude
External moment
Iteration steps
The optimisation criterion may therefore be described
as the minimisation of muscle forces leading to a stable
glenohumeral joint, that is, a joint resultant of minimal
magnitude which acts within the boundaries of the
glenoid.
3. Worked-out examples
Some worked-out examples are now given involving
muscular moments of abduction, adduction, posterior
flexion and anterior flexion with the humerus elevated
30 in the scapular plane and in neutral rotation. The
external moment to be overcome in all these cases is
9 N m, which, in the following examples, would corre-
spond to a 15 N force acting tangentially, at a distance
of 0.6 m from the centre of the humeral head, i.e. at the
end of an outstretched arm. No weight has been attrib-
uted to the arm itself. The particular directions of the
external force were chosen to compare the issuing results
with some of the estimates reported in literature. How-
ever, a force in any arbitrary direction and spatial posi-
tion could have been used.
Furthermore, for the same humerus position as de-
scribed above, two cases of pure torque of 9 N m about
the humeral axis have been examined, one of medial and
one of lateral rotation. The results are shown in Fig. 4.
3.1. Muscular abduction
This corresponds to an external force that gives rise
to adduction about the glenohumeral joint. The anterior
and middle part of the deltoid, especially the latter
(DEL3, DEL4), are particularly active, representing
about 1/2 of the joint resultant, followed by the biceps
(long head), supraspinatus and infraspinatus. The gleno-
humeral resultant force amounts to 450 N.
3.2. Muscular adduction
The external force tends to abduct the arm. Teres ma-
jor (TMAJ) takes the greater part in load sharing, with
the triceps, posterior part of the deltoid, latissimus dorsi,
830 P. Favre et al. / Clinical Biomechanics 20 (2005) 822–833
Fig. 4. Forces exerted by the 27 muscle segments for 6 cases of loading with the humerus in 30 elevation in the scapula plane and in neutral rotation.
The inserted field shows the intersection of the resultant with the glenoid, the right side corresponding to the anterior direction and the left side to the
posterior direction.
caudal parts of the pectoralis major and infraspinatus,
biceps (short head), coracobrachialis and teres minor
contributing. The magnitude of the resultant amounts
to 204 N.
3.3. Muscular posterior flexion
The external force to be balanced acts anteriorly. The
force exerted by the middle portion of the deltoid
(DEL3, DEL4) reaches a very high value, amounting
to about 2/3 of the total muscle force. The other muscles
involved are the teres major, latissimus dorsi, the poster-
ior part of the deltoid, triceps and infraspinatus. The
joint resultant exhibits a value of 466 N.
3.4. Muscular anterior flexion
The external force to be balanced acts posteriorly.
The anterior part of the deltoid (DEL1) becomes the
major contributor while the whole pectoralis major,
the middle deltoid segments DEL2 and DEL3, teres
minor, the cranial part of infraspinatus, supraspinatus,
subscapularis, biceps (short head), and coracobrachialis
assist. In this particular case, the co-activation of the
rotator cuff had to be raised to 1.6% of each rotator cuff
muscle maximal force in order to increase Ôconcavity
compressionÕ which was required to prevent dislocation
of the joint. The joint resultant amounts to 316 N.
3.5. Muscular medial rotation
An external turning moment of 9 N m was applied to
rotate the humerus in a lateral direction. The main mus-
cle which equilibrates this is the subscapularis (SSC1,
SSC2, SSC3) with a force amounting to about 3/4 of
the joint resultant. The middle part of the deltoid
(DEL3, DEL4) and the latissimus dorsi assist. The joint
resultant amounts to 642 N.
3.6. Muscular lateral rotation
An external turning moment of 9 N m was applied to
rotate the humerus in a medial direction. This is equili-
brated mainly by the infraspinatus (ISP2, ISP3) and
teres minor. Coracobrachialis, biceps (short head) and
supraspinatus assist to a small extent. The joint resultant
amounts to 478 N.
4. Discussion
We have not found any method described in the liter-
ature which would allow us to readily estimate the forces
acting across the glenohumeral joint when an arbitrary
combination of external loading conditions occur.
Although we also have not reached the point where this
can be accomplished for every possible position that can
 P. Favre et al. / Clinical Biomechanics 20 (2005) 822–833
be adopted by the humerus in relation to the scapula, we
have chosen a total of 12 particular positions often
encountered in daily life.
The algorithm presented is not a mathematically
exact method of arriving at a final solution but, like
any iterative method, is a search for an acceptable pos-
sibility within given constraints: tensile muscular activ-
ity, direction of muscle action, maximum attributable
muscle force, minimal energy requirements and position
of joint resultant with respect to glenoid boundaries.
The ASFE offers the advantage of very short calcula-
tion time, in generally less than 5 min, with a commonly
available PC when run with the well-known and wide-
spread Matlab
software.
A major simplification of the model employed is that
scapula motion in relation to the thorax has not been ta-
ken into account, the former remaining in its anatomi-
cally neutral position. Therefore, all elevation angles
mentioned in this presentation are referred strictly to
the scapula. This would certainly affect the estimated ac-
tion of the thoraco-humeral muscles when the humerus
is elevated beyond about 30 relative to the scapula, but
this was considered acceptable for the present purpose,
since only latissimus dorsi and pectoralis major are in-
volved. All other muscles link the humerus to the scap-
ula, so that movement of the latter does not change the
directions of action of the muscles as defined in the cho-
sen scapular coordinate system.
The factor k = 0.7 MN m
2 (Wood et al., 1989) has
been chosen because it lies in the middle of the consider-
able range of 0.4–1 MN m
2 (Crowninshield and Brand,
1981) found in the literature. However, this factor is of
major importance because it limits the maximal external
force that can be balanced and needs to be more pre-
cisely determined in further studies.
For muscles that cross the joint with deep bellies, in
particular the deltoid, the cord used in the model would
represent the action of muscle fibres closest to the joint.
The moment arms obtained for segments of the deltoid
are therefore minimal values and should in fact be re-
placed by values greater than those given in the corre-
sponding tables. However, because of the extreme
variability in the morphology of these muscle bellies,
and want of relevant details with respect to the path ta-
ken by the centroids of cross-section in various humeral
positions, these minimal values (worst conditions) have
been presently made use of in the algorithm. Further
work is required to produce more realistic data for del-
toid action.
Recruitment and dosage of muscular force to balance
a given external force is the goal of the algorithm; how-
ever, two major problems are encountered: (a) there are
more unknown muscle forces than equilibrium equa-
tions, and (b) each muscle (or muscle segment) most
often exerts forces in all three orthogonal directions of
a chosen Cartesian system simultaneously, not all of
831
which are conducive to equilibrating the external force
about a spherical joint. Segmentation of muscles in-
creases the number of unknown muscle forces, making
condition (a) more difficult to fulfil, but on the other
hand this was found to facilitate equilibration of a given
external force by having more selective variables, thus
making (b) simpler to comply with, through allowing
for finer regulation. The individual role played by each
muscle segment becomes obvious and this also vividly
illustrates the necessity to break down muscles, espe-
cially those with broad insertions, into component parts.
As already observed by Jössel (1880), muscular units,
although anatomically designated as one, must not be
functionally considered as one single motor unit, but
made up of several, and this is probably how fine control
of the neuromuscular system indeed functions (Wick-
ham and Brown, 1998).
The worked-out examples (Section 3) illustrate the
possibility of application of this algorithm clearly. A
comparison of the results obtained for the 6 different
loading cases shows immediately which external loads
are more difficult to cope with than others. For the same
external load amplitude, but in different directions, the
joint resultant can vary between 204 N (muscular adduc-
tion against an external abducting force) and 642 N
(muscular medial rotation against a turning moment in
the lateral direction).
Direct comparison of the estimates of muscle activity
obtained through the ASFE with those obtained by oth-
ers is difficult because of methodological and geometric
differences. Nevertheless, the middle and anterior del-
toid and supraspinatus muscles, known to be primary
movers in abduction (Kronberg et al., 1990), are shown
to play a predominant role for this activity in our model,
too. Incidentally, the biceps (long head) also appears to
play a significant role in stabilising the humeral head, an
observation shared by Karlsson and Peterson (1992),
Van der Helm (1994) and David et al. (2000). Laursen
et al. (1998) and Kronberg et al. (1990) measured
EMG activity from the infraspinatus during abduction
and considered this as a stabilizer.
In the case of adduction, others have attributed the
latissimus dorsi and pectoralis major a leading role in
this function. Although all these muscles do indeed
come into play, we observe a very important role taken
by teres major, corresponding to the findings of Pearl et
al. (1992), together with high activity on the part of the
triceps. The lower magnitude of the joint resultant
(204 N) in the worked-out example for adduction indi-
cates that the external force applied is far below what
can be coped with.
Posterior flexion surprisingly called upon the middle
part of the deltoid (DEL3 and DEL4) to play a domi-
nant role. The middle and the posterior deltoids were de-
scribed as prime movers for posterior flexion in the
EMG study by Kronberg et al. (1990). Actually in our
832 P. Favre et al. / Clinical Biomechanics 20 (2005) 822–833
model, these muscle segments bring a modest contribu-
tion to posterior flexion directly, but they are of major
value because of their high abduction component which
is required to neutralise the adduction moments of all
other important posterior flexors involved (teres major,
latissimus dorsi, DEL5, DEL6, and triceps), as can be
seen in Table 3.
Anterior flexion also incorporates some interesting
features: While the anterior part of the deltoid (DEL1)
plays a leading role, as would be expected (Kronberg
et al., 1990), and pectoralis major (Pearl et al., 1992),
coracobrachialis together with the biceps (Van der
Helm, 1994) are also known to be anterior flexors, the
substantial activity of teres minor has not been de-
scribed in this context before. Although teres minor does
not exhibit any flexion potential in this case, it is called
upon to compensate the unwanted medial rotation side-
effect of the anterior flexors (see Table 3). Interestingly,
anterior subluxation of the glenohumeral joint in this
case necessitates an increase of the general rotator cuff
activity which, however, entails simultaneous action of
antagonists that increase the magnitude of the joint
resultant somewhat.
Medial rotation was seen not to incorporate the ac-
tion of pectoralis major, a muscle described as an impor-
tant medial rotator by Kronberg et al. (1990). This is
because the very high anterior flexion activity of this
muscle makes it unsuitable. Subscapularis on the other
hand is much more efficient, in combination with latiss-
imus dorsi (both cited as prime movers by Kronberg
et al., 1990) and the middle part of the deltoid. The latter
can, in our study, more easily compensate the unwanted
anterior flexion effect of subscapularis.
Lateral rotation required the action of infraspinatus
and teres minor, as also reported in the literature (Bas-
majian and De Luca, 1985; MacConaill and Basmajian,
1977). In this humerus position, together with the pos-
terior segment of supraspinatus, they require some com-
pensation from coracobrachialis and the short-head
biceps to neutralise a posterior flexion tendency.
The good agreement concerning muscle recruitment
between the results of our study and other simulations
studies as well as EMG measurements show that the
recruitment criteria described in this study (groups A
and B) lead to meaningful results.
The examples show that in most situations, rotator
cuff co-activity is not required for stability. The resultant
mainly falls within the glenoid boundaries without
necessitating any co-contraction. This also ensures that
the magnitude of the resultant is kept as low as possible.
It must again be emphasised that as in any model
which endeavours to predict muscle forces, the solution
arrived at is strongly determined by the given boundary
conditions. In this model we have seen that the amount
of segmentation of each muscle plays an important role,
apart from the commonly known factors such as mo-
ment arm lengths, direction of muscle action, etc. How-
ever, it has been shown that with this algorithm, any
external force (within reasonable limits of magnitude)
can be considered, leading to an estimation of muscle
forces by searching for a solution in which these are a
minimum, resulting in a joint resultant of lowest possi-
ble value.
Finally, the ASFE has been used to analyse clinical
observations of instability with torn or weakened shoul-
der muscles and supports the observations made (article
in preparation).
5. Conclusion
A useful algorithm has been devised which is capable
of predicting muscle forces for certain given positions of
the humerus in equilibrating an external force acting in
an arbitrary direction. Good agreement between some
worked-out examples using this algorithm and other
simulation studies as well as EMG measurements has
been found. The strategy of force sharing among the
muscles opens up the possibility to reflect on the out-
come of muscle deficiencies and on causes of joint insta-
bility as encountered in clinical practice.
Further work would aim at examining the influence
of scapular rhythm, and the formulation of algebraical
functions that would allow determination of the re-
quired anthropometrical data (moment arm lengths
and directions of muscle action) for a general position
of the humerus within its physiological range of motion,
without the necessity to refer to tables containing data
that apply only for discrete positions. Also, additional
validation of the model is needed.
Acknowledgements
We are most grateful for the unwavering support gi-
ven by Prof. Christian Gerber M.D., Chairman of the
Orthopaedic Department of Balgrist, University of Zur-
ich, during the long course of this project. We also wish
to acknowledge the arduous contributions of Dr. Susan-
na Schärli, Mr. Daniele Müller, Dr. Reto Bertani and
Mr. Jordan Velikov, who collaborated with us during
early phases of this project, and especially the assistance
of Mr. Hans-Rudolf Sommer of the Biomechanics Lab-
oratory, who made the epoxy models and attended to
the measuring equipment.
Appendix A. Supplementary data
Supplementary data associated with this article can
be found, in the online version at doi:10.1016/
j.clinbiomech.2005.04.007.
 P. Favre et al. / Clinical Biomechanics 20 (2005) 822–833
References
Basmajian, J.V., De Luca, C.J., 1985. Muscles Alive: Their Functions
Revealed by Electromyography. Williams and Wilkins, Baltimore.
Crowninshield, R.D., Brand, R.A., 1981. A physiologically based
criterion of muscle force prediction in locomotion. J. Biomech. 14,
793–801.
David, G., Magarey, M.E., Jones, M.A., Dvir, Z., Turker, K.S.,
Sharpe, M., 2000. EMG and strength correlates of selected
shoulder muscles during rotations of the glenohumeral joint. Clin.
Biomech. 15 (2), 95–102.
De Luca, C.J., Forrest, W.J., 1973. Force analysis of individual
muscles acting simultaneously on the shoulder joint during
isometric abduction. J. Biomech. 6, 385–393.
Fick, A., 1850. Statische Betrachtung der Muskulatur des Oberschen-
kels. Zeitsch. F. Rationelle Med. 9, 94–106.
Fick, A.E., Weber, E., 1877. Anatomisch-mechanische Studie über die
Schultermuskeln, 2. Teil. Verh Phys-Med Gesell, Würzburg.
Fick, R., 1910. Handbuch der Anatomie und Mechanik der Gelenke,
II.Teil. Gustav Fischer, Jena.
Gerber, C., 1992. Latissimus dorsi transfer for the treatment of
irreparable tears of the rotator cuff. Clin. Orthop. 275, 152–160.
Happee, R., Van der Helm, F.C.T., 1995. The control of shoulder
muscles during goal directed movements, an inverse dynamic
analysis. J. Biomech. 28, 1179–1191.
Hardt, D.E., 1978. Determining muscle forces in the leg during normal
human walking—an application and evaluation of optimisation
methods. J. Biomech. Eng. 100, 72–78.
Inman, V.T., Ralston, H.J., Saunders, J.B., Feinstein, B., Wright Jr.,
E.W., 1952. Relation of human electromyogram to muscular
tension. Electroencephalogr. Clin. Neurophysiol. Suppl. 4, 187–
194.
Johnson, G.R., Spalding, D., Nowitzke, A., Bogduk, N., 1996.
Modelling the muscles of the scapula: Morphometric and coordi-
nate data and functional implications. J. Biomech. 29, 1039–1051.
Jössel, J.G., 1880. Ueber Rezidive bei Humerusluxationen. Deutsche
Zeitsch. f. Chir. 13, 167.
Karlsson, D., Peterson, B.J., 1992. Towards a model for force
predictions in the human shoulder. Biomechanics 25, 189–199.
Kronberg, M., Németh, G., Broström, L.-A., 1990. Muscle activity
and coordination in the normal shoulder. Clin. Orthop. 257, 76–85.
Laursen, B., Jensen, B.R., Németh, G., Sjøgaard, G., 1998. A model
predicting individual shoulder muscle forces based on relationship
between electromyographic and 3D external forces in static
position. J. Biomech. 31, 731–739.
Lippitt, S.B., Vanderhooft, J.E., Harris, S.L., Sidles, J.A., Harryman,
D.T., Matsen, F.A., 1993. Glenohumeral stability from concavity-
833
compression: A quantitative analysis. J. Shoulder Elbow Surg. 2,
27–35.
MacConaill, M.A., Basmajian, J.V., 1977. Muscles and Movements: A
Basis for Human Kinesiology. Williams and Wilkins Co,
Baltimore.
MacMahon, P.J., Debski, R.E., Thompson, W.O., Warner, J.J.P., Fu,
F.H., Woo, S.L., 1995. Shoulder muscle forces and tendon
excursions during glenohumeral abduction in the scapular plane.
J. Shoulder Elbow Surg. 4, 199–208.
Murray, W.A., Buchanan, T.S., Delp, S.L., 2000. The isometric
functional capacity of muscles that cross the elbow. J. Biomech. 33,
943–952.
Niemi, J., Nieminen, H., Takala, E.-P., Viikari-Juntura, E., 1996. A
static shoulder model based on a time-dependent criterion for load
sharing between synergistic muscles. J. Biomech. 29, 451–460.
Otis, J.C., Jiang, C.-C., Wickiewicz, T.L., Peterson, M.G.E., Warren,
R.F., Santner, T.J., 1994. Changes in the moment arms of the
rotator cuff and deltoid muscles with abduction and rotation.
J. Bone Joint Surg. 76-A, 667–676.
Pauwels, F., 1965. Der Schenkelhalsbruch. In: Pauwels, F. (Ed.),
Gesammelte Abhandlungen Zur Funktionellen Anatomie Des
Bewegungsapparates. Springer, Berlin, pp. 1–138.
Pearl, M.L., Perry, J., Torburn, L., Gordon, L.H., 1992. Detailed
biomechanical analysis of shoulder muscle function. Clin. Orthop.
284, 116–127.
Perthes, G., 1906. Über Operationen der habituellen Schulterluxation.
Deutsche Zeitsch. f. Chir. 85, 199–227.
Seireg, A., Arvikar, R.J., 1975. The prediction of muscular load
sharing and joint forces in the lower extremities during walking.
J. Biomech. 8, 89–102.
Shiino, K., 1913. Ueber die Bewegungen im Schultergelenk und die
Arbeitsleistung der Schultermuskeln. Arch. f. Anat. u. Physiol.
Van der Helm, F.C.T., 1994. Analysis of the kinematic and dynamic
behaviour of the shoulder mechanism. J. Biomech. 27, 527–550.
Van der Helm, F.C.T., Veebaas, R., 1991. Modelling the mechanical
effect of muscles with large attachment sites: Application to the
shoulder mechanism. J. Biomech. 24, 1151–1163.
Veeger, H.E.J., Van der Helm, F.C.T., Van der Woude, L.H.V.,
Pronk, G.M., Rozendal, R.H., 1991. Inertia and muscle contrac-
tion parameters for musculoskeletal modelling of the shoulder
mechanism. J. Biomech. 24, 615–629.
Wickham, J.B., Brown, J.M., 1998. Muscles within muscles: The
neuromotor control of intra-muscular segments. Eur. J. Appl.
Physiol. Occup. Physiol. 78 (3), 219–225.
Wood, J.E., Meek, S.G., Jacobsen, S.C., 1989. Quantitation of human
shoulder anatomy for prosthetic arm control—I. Surface model-
ling. J. Biomech. 22, 273–292.