Effect of sub-optimal neuromotor control on the hip joint load during level walking

Page 1

Journal of Biomechanics 44 (2011) 1716–1721

Contents lists available at ScienceDirect

journal homepage: www.elsevier.com/locate/jbiomech

Journal of Biomechanics

0021-92

doi:10.1

n Corr

Rizzoli,

fax: þ3

E-m

www.JBiomech.com

Effect of sub-optimal neuromotor control on the hip joint loadduring level walking

Saulo Martelli a,n, Fulvia Taddei a, Angelo Cappello b, Serge van Sint Jan c,Alberto Leardini d, Marco Viceconti a

a Laboratorio di Tecnologia Medica, Istituto Ortopedico Rizzoli, Bologna, Italyb Dipartimento di Elettronica, Informatica e Sistemistica, Universit �a di Bologna, Italyc Department of Anatomy, Universite Libre de Bruxelles, Belgiumd Movement Analysis Laboratory, Istituto Ortopedico Rizzoli, Bologna, Italy

a r t i c l e i n f o

Article history:

Accepted 29 March 2011Skeletal forces are fundamental information in predicting the risk of bone fracture. The neuromotor

control system can drive muscle forces with various task- and health-dependent strategies but current

Keywords:

Subject-specific musculoskeletal models

Sub-optimal neuromotor control

Hip loads

Muscle force variability

Level walking

90/$ - see front matter & 2011 Elsevier Ltd. A

016/j.jbiomech.2011.03.039

espondence to: Laboratorio di Tecnologia

Via di Barbiano, 1/10, 40136 Bologna, It

9 051 6366863.

ail address: [email protected] (S. Martelli).

a b s t r a c t

modelling techniques provide a single optimal solution of the muscle load sharing problem. The aim of

the present work was to study the variability of the hip load magnitude due to sub-optimal neuromotor

control strategies using a subject-specific musculoskeletal model. The model was generated from

computed tomography (CT) and dissection data from a single cadaver. Gait kinematics, ground forces

and electromyographic (EMG) signals were recorded on a body-matched volunteer. Model results were

validated by comparing the traditional optimisation solution with the published hip load measure-

ments and the recorded EMG signals. The solution space of the instantaneous equilibrium problem

during the first hip load peak resulted in 105 dynamically equivalent configurations of the neuromotor

control. The hip load magnitude was computed and expressed in multiples of the body weight (BW).

Sensitivity of the hip load boundaries to the uncertainty on the muscle tetanic stress (TMS) was also

addressed. The optimal neuromotor control induced a hip load magnitude of 3.3 BW. Sub-optimal

neuromotor controls induced a hip load magnitude up to 8.93 BW. Reducing TMS from the maximum to

the minimum the lower boundary of the hip load magnitude varied moderately whereas the upper

boundary varied considerably from 4.26 to 8.93 BW. Further studies are necessary to assess how far the

neuromotor control can degrade from the optimal activation pattern and to understand which sub-

optimal controls are clinically plausible. However we can consider the possibility that sub-optimal

activations of the muscular system play a role in spontaneous fractures not associated with falls.

& 2011 Elsevier Ltd. All rights reserved.

1. Introduction

Load cells and force platforms make possible accurate measure-ment of the external forces that act on our body during activities ofdaily living (Sutherland, 2005). However, the determination of theinternal forces that the same physical activity induces on ourskeleton through the joints, the ligaments, and the muscle insertionsremain difficult to quantify. But this information is of vital impor-tance in a number of research and clinical contexts. For example, therisk of fracture that a given subject faces while performing a givenmotor task depends not only on the specific bone strength, but alsoon the internal forces.

ll rights reserved.

Medica, Istituto Ortopedico

aly. Tel.: þ39 051 6366554;

The problem is affected by a dramatic indeterminacy. Even if wemodel the skeleton as a mechanism made of idealised joints,represent each major muscle bundle with a single actuator, andimpose all the physiological limits to the force expressed by eachactuator, the resulting mathematical problem has more unknownsthan equations. The best solution, when the kinematics of eachsegment has been measured experimentally, is to postulate that theneuromotor control activates the muscle fibres ensuring the instan-taneous equilibrium while minimising a cost function (Collins, 1995;Menegaldo et al., 2006; Praagman et al., 2006).

The assumption that in healthy subjects the neuromotor controlworks in fairly optimal conditions seems reasonable. Indeed, whenapplied to volunteers this approach predicts muscle activationpatterns in good agreement with electromyography (EMG) recordings(Anderson and Pandy, 2001; Erdemir et al., 2007; Heller et al., 2001).Also, the intensity of the hip load predicted is comparable to thatrecorded with telemetric instrumented prostheses (Heller et al., 2001;Stansfield et al., 2003).

Page 2

S. Martelli et al. / Journal of Biomechanics 44 (2011) 1716–1721 1717

This approach presumes that the neuromotor control chooses,among the infinite solutions available, the muscle activationpattern that optimises a certain cost function, always the sameone. But this assumption seems unrealistic for the followingcases:

–

A large variability of the internal forces can be observed for asingle subject through several repetitions of the same motiontask (Bergmann et al., 2001). The ‘‘UnControlled Manifold’’theory (Scholz and Schoner, 1999) suggests that the motorcontrol strategy focusses on the goal of the task, and that everytrajectory within the manifold of the task-equivalent config-uration of the muscle actuators is virtually possible.

–

While we move, our goal is dependent on a number of factors.Specific activation patterns were found in case of patello-femoralpain (Besier et al., 2009), in unstable conditions (Bergmann et al.,2004), in sudden motion tasks (Yeadon et al., 2010) and differentmuscles controls were found during the execution of precise andpower activities (Anson et al., 2002).

–

The way we move is also affected by our emotions. Depressionhas been found a co-factor for the risk of falling in the elders(Skelton and Todd, 2007; Talkowski et al., 2008), whereassomatisation, anxiety and depression were found intrinsicco-factors in non-specific musculoskeletal spinal disorders(Andersson, 1999).

–

Even if the optimal control assumption is acceptable fornormal subjects, does it remain acceptable when used tomodel specific patients that are known to have neuromotordeficiencies? EMG studies seem to suggest the answer is ‘‘no’’(Liikavainio et al., 2009; Mahaudens et al., 2009).

More specifically, in predicting the risk of fracture of an elderwho has a history of falls, is it still reasonable to postulate that themuscle activation follows some optimal patterns? This question isvery difficult to answer in such general terms; but as a first step asimpler question might be addressed: is the intensity of the loadsacting on the skeleton during a given motor task severely affectedby neuromotor control strategy?

The aim of the present study is to estimate, using a subject-specific musculoskeletal model, how the intensity of the hip jointreaction acting is affected when the instantaneous equilibrium isachieved in full respect of the physiological constraints, but withsub-optimal activation patterns. Sensitivity of potential sub-optimal control strategies to the muscle mechanics parameterswas also assessed.

2. Materials and methods

2.1. Data collection

A musculoskeletal model of the lower body was developed from a detailed

multiscale data collection (Testi et al., 2010). In this project the cadaver of a

81 years old woman, 167 cm height and 63 kg mass with no history of muscu-

loskeletal diseases was examined with a whole body high-resolution computed

tomography (CT) protocol. The muscle attachment on bones, the muscle length

and their superficial lines of action were digitised during dissection. The volume of

each muscle was measured by water immersion. The co-ordinates of the muscle

attachments were registered in space through a dedicated software (LhpBuilder,

SCS, Italy) into a unique multimodal data collection.

A body-matched healthy volunteer (female, 25 years old, 57 kg, 165 cm) was

subjected to a detailed gait analysis protocol during level walking (Leardini et al.,

2007; Manca et al., 2010), which provided 3D motion (Vicon Motion Capture,

Oxford UK) of the lower limb segments (sampling rate 100 Hz) and the ground

reaction forces at both feet (sampling rate 2000 Hz). The muscle activity was

recorded for the most relevant muscles of the lower limb: gluteus maximus,

gluteus medius, rectus femoris, vastus lateralis, vastus medialis, biceps femoris

long head, semitendinosus, gastrocnemius medialis, gastrocnemius lateralis,

soleus and tibialis anterior. Surface EMG signals were acquired (TelEMGs, BTS,

Italy) at the sampling rate of 2000 Hz. Digital band filtering was used to extract

the qualitative muscle control pattern (Glitsch and Baumann, 1997).

2.2. The musculoskeletal model

The skeletal anatomy was extracted from the CT dataset using dedicated semi-

automatic software (Amiras, Mercury Computer System, Inc., USA). The biome-

chanical model of the musculoskeletal system of the lower-limb was defined as a

7-segment, 10-degree-of-freedom (DOF) articulated system actuated by 82

muscle–tendon units. Each leg was articulated by three ideal joints: a ball and

socket at the hip (3 DOF) and a hinge (1 DOF) at both the knee and the ankle

(Jonkers et al., 2008). The identification of joint parameters was based on relevant

skeletal landmarks identified on the skeletal surface (Taddei et al., 2007). All

anatomical landmarks suggested by ISB standards were identified. The local

coordinate system was computed for each segment (Wu et al., 2002). The hip

centre was defined as the centre of the sphere that best fit the femoral head

surface. The hip joint rotations were defined according to ISB standards (Wu et al.,

2002). The knee rotation axis was assumed as that connecting the medial and the

lateral epicondyles, which is normally considered a good approximation of this

axis in surgical procedures (Tanavalee et al., 2001). The ankle flexion axis was

assumed as that connecting the medial and the lateral malleoli. An expert

anatomist manually registered on the subject-specific skeletal anatomy the

muscular model of the lower extremity (Delp et al., 1990) using as reference the

muscle line-of-actions digitised during dissection (Fig. 1b). The peak force each

muscle can exert was estimated from the physiological cross section area (PCSA),

assuming the muscle tetanic stress (TMS) equal to 1 MPa (Glitsch and Baumann,

1997). The muscle mechanical properties were defined accordingly (Daggfeldt

and Thorstensson, 2003). Inertial parameters of each segment were derived from

CT data (Fig. 1a) assuming homogeneous density properties for both the hard

(1.42 g/cm3) and the soft (1.03 g/cm3) tissues (Dumas et al., 2005). Preliminary

simulations were run to solve the muscle load sharing problem using a traditional

static optimisation approach (Anderson and Pandy, 2001; Collins, 1995;

Crowninshield and Brand, 1981) (Fig. 1c). The hip load magnitude was calculated

and expressed in multiples of the ground reaction peak (GRp) to account for the

inter-subject weight and walking dynamics variability.

2.3. Musculoskeletal model validation

The predicted hip load under the hypothesis of optimal neuromotor control

was compared to those measured in various subjects using telemetric implants

(Bergmann et al., 2001). The predicted force patterns exerted by the principal

muscles were compared with the corresponding electrical activity (Glitsch and

Baumann, 1997) recorded on the body-matched volunteer during the same

motion task. The model configuration that produced the first load peak was

identified as the subject for the subsequent analysis of sub-optimal neuromotor

control conditions.

2.4. Variability of the hip load in sub-optimal motor-control strategies

The instantaneous equilibrium at the joints is achieved by every set of muscle

forces that satisfies the following equation:

B m,nUF n ¼Mm ð1Þ

where the matrix B m,n contains the muscle lever arm of each of the n muscle

action lines acting on each of the m DOFs, F n is a n dimension vector containing

the design variables, i.e. the muscle forces, and Mm is the m dimension vector that

contains the joint net moments necessary to follow the target kinematics. As

muscle force we considered only the active component, as the passive forces in the

frame of motion here considered are negligible, due to the moderate stretch of the

muscle fibre in that position. The active force component depends on a number of

factors including the current muscle length, velocity and the activation conditions

at the previous frame of motion. These dependencies of the muscle forces are

properly described by complex relationships that require a large number of

parameters to be identified (e.g. see Thelen et al., 2003) and are often affected

by considerable uncertainty. Thus, to include all possible conditions that the

neuromotor control system can potentially impose to the musculoskeletal system

we constrained the spectrum of possible muscle forces between zero (i.e.

completely inactive muscle) and the tetanic muscle forces at their optimal length

in isometric conditions (Fmax):

F n A ½0,Fmax� ð2Þ

This solution space of the non-homogeneous linear system of equations ensuring

the instantaneous equilibrium at the joints (Eq. (1)) is given by the null space of

the matrix B m,n plus a particular solution of Eq. (1) (e.g. the optimal neuromotor

control solution used in this study). This space is an m–n dimensions hyper-plane

in the n-dimensional space of the unknowns (i.e. the muscle forces). In this hyper-

plane, the domain of the physiologically plausible muscle forces consist in a

Page 3

Fig. 2. Comparison of the predicted pattern of the hip load (solid black line) with the

variability of the hip load magnitude (grey band) measured on 4 subjects through an

hip prosthesis instrumented with a telemetric force sensor (Bergmann et al., 2001).

Fig. 1. From CT images and cadaver dissection to the subject-specific musculoskeletal simulation: extracting the skeletal anatomy and the inertial parameters from CT data (a),

registering the muscular model by using the dissected fibre path as reference (b), fusing motion data (i.e. kinematics and ground forces) registered on the body-matched

volunteer.

S. Martelli et al. / Journal of Biomechanics 44 (2011) 1716–17211718

bounded portion (the hypersimplex hereinafter) where the boundaries are

identified by the constraints (Eq. (2)). Preliminary simulations suggested that

the shape of the hypersimplex was acicular; i.e. that the domain width along the

direction connecting the two neuromotor control strategies producing the max-

imum (HRmax) and the minimum (HRmin) hip load was much larger than the width

along all the other directions. Therefore, 103 particular solutions of Eq. (1) were

obtained by uniformly sampling the vector connecting HRmax and HRmin (the

principal force vector hereinafter). For each particular solution 102 perturbations

were calculated by adding random combinations of the base vectors of the null

space. Thus, the resulting 105 samples of the hypersimplex included the optimal

and the sub-optimal neuromotor control conditions. Differences between the hip

load magnitude produced by the optimal and the sub-optimal neuromotor

controls were calculated and expressed in multiples of the individual body weight

(BW) (see Appendix A for details).

2.5. Sensitivity of sub-optimal joint reactions to changes of the tetanic muscle

stress (TMS)

The peak force a muscle can exert is conventionally estimated through two

different parameters: the muscle physiological cross section area (PCSA) and TMS

(Daggfeldt and Thorstensson, 2003). But while several methods can be adopted in

extracting reliable information on the PCSA (Jolivet et al., 2009), for the TMS

estimation the problem is more complex and the literature (Buchanan et al., 2004)

reports a very large range (0.35–1.37 MPa). We repeated the estimation of the hip

load boundaries spanning the entire TMS range.

3. Results

3.1. Musculoskeletal model validation

When the subject-specific model was solved imposing the optimalcontrol condition, the hip load was predicted in reasonably goodagreement with reported measurements (Bergmann et al., 2001)(Fig. 2). The major differences were found during stance-to-swing;

the out-of-bounds and the higher load rate of predictions wereprobably due to the different healthy conditions of the body-matchedvolunteer and of the implanted subjects from the reference study(Bergmann et al., 2001). In fact, the walking dynamics of the body-matched volunteer induced a peak load on the recorded groundreaction of 1.33 BW while this value was always approximately 1 BWon the reference study (Bergmann et al., 2001). The frame corre-sponding to the first peak of the hip load (2.8 GRp) was selected forthe subsequent analysis of the hip load variability in sub-optimalmotor-control conditions. This peak was found at 17% of the gaitcycle, consistently with earlier studies (Bergmann et al., 2001;Heller et al., 2001).

The force patterns predicted for the major muscle groups werefound in acceptable agreement with recorded and expected (Neneet al., 2004; Shiavi, 1985) firing patterns, particularly at 17% gait(Fig. 3). A major discrepancy was the near-zero force predicted forthe gluteals and the quadriceps muscles close to heel-strike where aconsistent electrical activity was recorded. The moderate loadspredicted on the vastii during the stance-to-swing phase was

Page 4

Fig. 3. For each muscle, on the top the predicted forces and below the recorded electrical activity of the muscle during a complete gait cycle. The shaded region represent

the swing phase while the non-shaded region represent the stance phase. The electrical activity is represented by the rectified EMG signal and superimposed, it is visible

the EMG envelope (Glitsch and Baumann, 1997). The black bars in the bottom represent the expected firing patterns (Nene et al., 2004; Shiavi, 1985). The dash-line

indicate the frame of interest for this study.

S. Martelli et al. / Journal of Biomechanics 44 (2011) 1716–1721 1719

neither recorded nor expected (Nene et al., 2004; Shiavi, 1985). Last,a consistent electrical activity was recorded for the tibialis anteriorduring mid stance when a negligible muscle force was predicted.

3.2. Variability of the hip load in sub-optimal motor-control

strategies

At 17% of the gait cycle the two extreme sub-optimal neuro-motor control strategies on the principal force vector induced thelargest variations of the hip load magnitude; the distance betweenthe highest (HRmax¼7.1 BW) and the lowest (HRmin¼3.3 BW) hipload value was 3.8 BW. Perturbing intermediate neuromotor con-trol solutions, changes on the hip load magnitude were alwaysbelow 1 BW.

3.3. Sensitivity of sub-optimal joint reactions to changes of TMS

At 17% of the gait cycle the equilibrium at the joints was foundfor TMS values ranging from 0.57 to 1.37 MPa; when TMS was inthe range 0.35–0.57 MPa, the excessive muscle weakness did notguarantee the dynamic balance of the motion. AssumingTMS¼0.57 MPa, the hip load ranged between 3.77 and 4.26 BW

(Fig. 4). This means 13% higher than the lower boundary of thejoint load. When TMS was increased to the maximum value(1.37 MPa) by uniform steps of 0.1 MPa the lower boundary ofthe hip load magnitude initially decreased from 3.77 to 3.3 BWand no significant changes were found as TMS increased from0.77 to 1.37 MPa. On the contrary, the variation of the upperboundary of the hip load was significant: for the lowest TMS valuethe maximum hip load magnitude was 4.26 BW while for thehighest TMS value it was 8.93 BW, the 275% higher than thecorresponding lower boundary of the joint reaction (Fig. 4).

4. Discussion

The aim of the present study was to estimate, using a subject-specific musculoskeletal model, how the intensity of the jointreaction acting on the femur through the hip is affected when theinstantaneous equilibrium is achieved in full respect of the physio-logical constraints, but with sub-optimal activation patterns.

When the model was solved imposing the optimal neuromotorcontrol the predicted muscle forces were in an acceptable agree-ment with the recorded electrical activities. Some discrepancies can

Page 5

Fig. 4. Field of possible hip load due to TMS changes (dark grey region). The lowest hip load magnitude was slightly sensitive to TMS changes (3.3–3.77 BW) while the highest

hip load magnitude increased from 4.26 to 8.93 BW as TMS increased. The regression lines intersect in a TMS region very close to the lower boundary of the published TMS

(0.37 MPa, (Buchanan et al., 2004)) although the numerical optimisation algorithm did not reach convergence for TMS values below 0.57 MPa (light grey region).

S. Martelli et al. / Journal of Biomechanics 44 (2011) 1716–17211720

be likely attributed to the limits of the optimisation technique.Close to heel strike, the underestimation of the gluteals and thevastii forces were likely due to the tendency of the method tounderestimate the muscle co-contractions when rapid changes onthe joint net moments occur (Yeadon et al., 2010). The moderatevastii force (o80 N) predicted during stance-to-swing was notexpected. However similar force patters were predicted in similarstudies (Xiao and Higginson, 2010) suggesting that the complexactivation pattern of the quadriceps (Nene et al., 2004) can beroughly described by optimisation techniques. The last majordiscrepancies can be attributed to cross-talk. The significant activityrecorded during early stance and late swing by the EMG sensor ofthe rectus femoris was probably due to the vastii activity during thesame phase of motion (Nene et al., 2004) and, the activity recordedby the tibialis anterior sensor during mid stance was probably dueto the high activity of the triceps surae. However a generalagreement was found between predicted force patterns and theexpected firing patterns, and the major discrepancies were notpresent at 17% gait, the frame afterwards considered for theevaluation of sub-optimal neuromotor conditions. The predictedhip loads were consistent with published measurements(Bergmann et al., 2001) showing that the total sum of hip muscleforces is reasonable. Thus, in the authors’ opinion, all these limita-tions should not invalidate the generality of the conclusions.

The subsequent analysis explored 105 sub-optimal neuromo-tor control conditions showed that sub-optimal neuromotorcontrols can drastically increase the hip load intensity up toapproximately 9 BW; this load is of the same order of magnitudeof the fracture load of the femoral neck measured in cadavericstudies (Cristofolini et al., 2007). The results seem to support theexistence of a principal force axis on the hypersimplex. WhenTMS was set to 1.37 MPa, sampling the muscle force solutionsalong the principal force vector we found variations on the hipload up to 5.63 BW while random perturbations of the inter-mediate solutions induced much smaller variations. Within thehypersimplex the optimal neuromotor control solution wasslightly sensitive to TMS changes in agreement with Redl et al.(2007). However, when sub-optimal controls are allowed, thesensitivity of the model predictions to TMS changes is muchhigher. For instance, the maximum hip load increased from 4.26to 8.93 BW as TMS increased from 0.57 to 1.37 MPa.

To the authors’ knowledge this is the first study that exploresthe effect of sub-optimal control on the forces acting on theskeleton during motion. Nonetheless the comparison of

intermediate outcomes with published works supports the relia-bility of our results. Indeed, the gait kinematic and kinetics in ourmodel were consistent with relevant reports (Heller et al., 2001)and, the firing pattern of principal muscles was consistent withexpected normal patterns (Shiavi, 1985).

The variability of the hip load during level walking in the samesubject during repetitions is below 16% (Bergmann et al., 2001).While this value is much smaller than the one reported here, weshould not forget that the variability over a repeated taskrepresents the aleatory uncertainty (uncertainty arising becauseof natural, unpredictable variation in the performance of thesystem under study) of the neuromotor control. The rangewe reported in this work represents the epistemic uncertainty

(uncertainty due to a lack of knowledge about the behaviour ofthe system), and in particular to the epistemic uncertainty relatedto the degradation of the neuromotor control.

The large variability of possible sub-optimal control strategies isconsistent with the known potential of the neuromotor controlsystem of activating muscles following various strategies (Bergmannet al., 2004; Besier et al., 2009). In epidemiology studies onspontaneous osteoporotic fractures, there is a fraction of thepopulation for which the decrease of bone density appears insuffi-cient to explain the fracture event (Yang et al., 1996). But thisobservation could be easily explained if we accept that thedegradation of the neuromotor control not only increases the riskof falling, but also produces overloads during normal physiologicalactivities.

The present study is affected by some limitations. The simula-tions were run deriving the data from different sources: acadaveric study and in-vivo recordings on a body-matchedvolunteer. While this inevitably induces some inaccuracies, theimpossibility of collecting all necessary data on living subjects,does balance what we would determine if many of these para-meters had to be estimated from the literature.

The study was limited to a single instant of the walking cycle.However, it is hard to see how the extension to the entire cyclecould produce different conclusions. If the variability of the hipload in a single instant is so big, over the entire cycle it can onlybe equal or bigger.

The method we developed to sample the solution hyperspacedoes not guarantee that the sampling covers all the hypersimplex.But again, even if the sampling were missing entirely a region ofthe hypersimplex, this could only produce a variability equal orlarger than that reported here.

Page 6

S. Martelli et al. / Journal of Biomechanics 44 (2011) 1716–1721 1721

Probably the most important limitation of this study is that wehave no way to know which of these sub-optimal solutions isclinically plausible. At our current level of understanding there isno way to know how far the neuromotor control can degradefrom the optimal activation pattern; the answer to this questionrequires further studies.

Nevertheless the results presented here seem relevant forbiomechanics research. The great variability of the intensity ofthe hip load observed for sub-optimal solutions suggests that thedetermination of the internal forces transmitted to the skeletonby postulating some optimal control is probably in many casestoo optimistic. It seems more reasonable to imagine probabilisticapproaches to this problem, where the probability associated toeach sub-optimal solution is estimated from quantifications ofneuromotor condition of the patient.

The present study suggests that the deviation of the optimalneuromotor control could drive the intensity of the internal forcesacting on the skeleton to considerably higher levels that thosepredicted in optimal conditions. If this is true, we can consider thepossibility that spontaneous fractures might be due to skeletaloverloading produced by the sub-optimal control of the muscles’activation.

Conflict of interest statement

There is no potential conflict of interests related to this study.None of the authors received nor will receive direct or indirectbenefits from third parties for the performance of this study.

Acknowledgments

Data were produced during the EU-funded project LHDL (IST-2004-026932) and freely available at www.physiomespace.com

Appendix A. Supplementary data

Supplementary data associated with this article can be foundin the online version at doi:10.1016/j.jbiomech.2011.03.039.

References

Anderson, F.C., Pandy, M.G., 2001. Static and dynamic optimization solutions forgait are practically equivalent. J. Biomech. 34, 153–161.

Andersson, G.B.J., 1999. Epidemiological features of chronic low-back pain. Lancet354, 581–585.

Anson, J.G., Hasegawa, Y., Kasai, T., Latash, M.L., Yahagi, S., 2002. EMG dischargepatterns during human grip movement are task-dependent and not modu-lated by muscle contraction modes: a transcranial magnetic stimulation (TMS)study. Brain Res. 934, 162–166.

Bergmann, G., Deuretzbacher, G., Heller, M., Graichen, F., Rohlmann, A., Strauss, J.,Duda, G.N., 2001. Hip contact forces and gait patterns from routine activities.J. Biomech. 34, 859–871.

Bergmann, G., Graichen, F., Rohlmann, A., 2004. Hip joint contact forces duringstumbling. Langenbecks Arch. Surg. 389, 53–59.

Besier, T.F., Fredericson, M., Gold, G.E., Beaupre, G.S., Delp, S.L., 2009. Knee muscleforces during walking and running in patellofemoral pain patients and pain-free controls. J. Biomech. 42, 898–905.

Buchanan, T.S., Lloyd, D.G., Manal, K., Besier, T.F., 2004. Neuromusculoskeletalmodeling: estimation of muscle forces and joint moments and movementsfrom measurements of neural command. J. Appl. Biomech. 20, 367–395.

Collins, J.J., 1995. The redundant nature of locomotor optimization laws.J. Biomech. 28, 251–267.

Cristofolini, L., Juszczyk, M., Martelli, S., Taddei, F., Viceconti, M., 2007. In vitroreplication of spontaneous fractures of the proximal human femur. J. Biomech.40, 2837–2845.

Crowninshield, R.D., Brand, R.A., 1981. A physiologically based criterion of muscleforce prediction in locomotion. J. Biomech. 14, 793–801.

Daggfeldt, K., Thorstensson, A., 2003. The mechanics of back-extensor torqueproduction about the lumbar spine. J. Biomech. 36, 815–825.

Delp, S.L., Loan, J.P., Hoy, M.G., Zajac, F.E., Topp, E.L., Rosen, J.M., 1990. Aninteractive graphics-based model of the lower extremity to study orthopaedicsurgical procedures. IEEE Trans. Biomed. Eng. 37, 757–767.

Dumas, R., Aissaoui, R., Mitton, D., Skalli, W., de Guise, J.A., 2005. Personalizedbody segment parameters from biplanar low-dose radiography. IEEE Trans.Biomed. Eng. 52, 1756–1763.

Erdemir, A., McLean, S., Herzog, W., van den Bogert, A.J., 2007. Model-basedestimation of muscle forces exerted during movements. Clin. Biomech.(Bristol, Avon) 22, 131–154.

Glitsch, U., Baumann, W., 1997. The three-dimensional determination of internalloads in the lower extremity. J. Biomech. 30, 1123–1131.

Heller, M.O., Bergmann, G., Deuretzbacher, G., Durselen, L., Pohl, M., Claes, L., Haas,N.P., Duda, G.N., 2001. Musculo-skeletal loading conditions at the hip duringwalking and stair climbing. J. Biomech. 34, 883–893.

Jolivet, E., Daguet, E., Bousson, V., Bergot, C., Skalli, W., Laredo, J.D., 2009.Variability of hip muscle volume determined by computed tomography. IRBM30, 14–19.

Jonkers, I., Sauwen, N., Lenaerts, G., Mulier, M., Van der Perre, G., Jaecques, S., 2008.Relation between subject-specific hip joint loading, stress distribution in theproximal femur and bone mineral density changes after total hip replacement.J. Biomech. 41, 3405–3413.

Leardini, A., Sawacha, Z., Paolini, G., Ingrosso, S., Nativo, R., Benedetti, M.G., 2007.A new anatomically based protocol for gait analysis in children. Gait Posture.

Liikavainio, T., Bragge, T., Hakkarainen, M., Karjalainen, P.A., Arokoski, J.P., 2009.Gait and muscle activation changes in men with knee osteoarthritis. Knee 17,69–76.

Mahaudens, P., Banse, X., Mousny, M., Detrembleur, C., 2009. Gait in adolescentidiopathic scoliosis: kinematics and electromyographic analysis. Eur. Spine J.18, 512–521.

Manca, M., Leardini, A., Cavazza, S., Ferraresi, G., Marchi, P., Zanaga, E., Benedetti,M.G., 2010. Repeatability of a new protocol for gait analysis in adult subjects.Gait Posture 32, 282–284.

Menegaldo, L.L., de Toledo Fleury, A., Weber, H.I., 2006. A ‘cheap’ optimal controlapproach to estimate muscle forces in musculoskeletal systems. J. Biomech.39, 1787–1795.

Nene, A., Byrne, C., Hermens, H., 2004. Is rectus femoris really a part ofquadriceps?: Assessment of rectus femoris function during gait in able-bodiedadults. Gait Posture 20, 1–13.

Praagman, M., Chadwick, E.K., van der Helm, F.C., Veeger, H.E., 2006. The relation-ship between two different mechanical cost functions and muscle oxygenconsumption. J. Biomech. 39, 758–765.

Redl, C., Gfoehler, M., Pandy, M.G., 2007. Sensitivity of muscle force estimates tovariations in muscle-tendon properties. Hum. Movement Sci. 26, 306–319.

Scholz, J.P., Schoner, G., 1999. The uncontrolled manifold concept: identifyingcontrol variables for a functional task. Exp. Brain Res. 126, 289–306.

Shiavi, R., 1985. Electromyographic patterns in adult locomotion: a comprehensivereview. J. Rehabil. Res. Dev. 22, 85–98.

Skelton, D.A., Todd, C.J., 2007. Prevention of Falls Network Europe: a thematicnetwork aimed at introducing good practice in effective falls prevention acrossEurope: four years on. J. Musculoskeleton Neuronal Interact. 7, 273–278.

Stansfield, B.W., Nicol, A.C., Paul, J.P., Kelly, I.G., Graichen, F., Bergmann, G., 2003.Direct comparison of calculated hip joint contact forces with those measuredusing instrumented implants. An evaluation of a three-dimensional mathe-matical model of the lower limb. J. Biomech. 36, 929–936.

Sutherland, D.H., 2005. The evolution of clinical gait analysis, Part III—kinetics andenergy assessment. Gait Posture 21, 447–461.

Taddei, F., Ansaloni, M., Testi, D., Viceconti, M., 2007. Virtual palpation of skeletallandmarks with multimodal display interfaces. Inf. Health Social Care 32,191–198.

Talkowski, J.B., Brach, J.S., Studenski, S., Newman, A.B., 2008. Impact of healthperception, balance perception, fall history, balance performance, and gaitspeed on walking activity in older adults. Phys. Ther. 88, 1474–1481.

Tanavalee, A., Yuktanandana, P., Ngarmukos, C., 2001. Surgical epicondylar axis vsanatomical epicondylar axis for rotational alignment of the femoral compo-nent in total knee arthroplasty. J. Med. Assoc. Thai 84 (Suppl. 1), S401–408.

Testi, D., Quadrani, P., Viceconti, M., 2010. PhysiomeSpace: digital library servicefor biomedical data. Philos. Trans. A—Math. Phys. Eng. Sci. 368, 2853–2861.

Thelen, D.G., Anderson, F.C., Delp, S.L., 2003. Generating dynamic simulations ofmovement using computed muscle control. J. Biomech. 36, 321–328.

Wu, G., Siegler, S., Allard, P., Kirtley, C., Leardini, A., Rosenbaum, D., Whittle, M.,D’Lima, D.D., Cristofolini, L., Witte, H., Schmid, O., Stokes, I., 2002. ISBrecommendation on definitions of joint coordinate system of various jointsfor the reporting of human joint motion—Part I: ankle, hip, and spine.J. Biomech. (International Society of Biomechanics) 35, 543–548.

Xiao, M., Higginson, J., 2010. Sensitivity of estimated muscle force in forwardsimulation of normal walking. J. Appl. Biomech. 26, 142–149.

Yang, K.H., Shen, K.L., Demetropoulos, C.K., King, A.I., Kolodziej, P., Levine, R.S.,Fitzgerald Jr., R.H., 1996. The relationship between loading conditions andfracture patterns of the proximal femur. J. Biomech. Eng. 118, 575–578.

Yeadon, M.R., King, M.A., Forrester, S.E., Caldwell, G.E., Pain, M.T., 2010. The needfor muscle co-contraction prior to a landing. J. Biomech. 43, 364–369.