Journal of Biomechanics 94 (2018) 230-234 Contents lists available at ScienceDiroct Journal of Biomechanics journal homepage: www.elsavier.com/locate/jbiomech www JBiamech:com ELSEVIER, Short communication Development of a finite element musculoskeletal model with the ability @® to predict contractions of three-dimensional muscles ae Junyan Li**, Yongtao Lu®*, Stuart C. Miller‘, Zhongmin Jin“, Xijin Hua® * Schoo Scene and Teno), Mddeer Univers, Lond, UK ° Deparment of Pairing Macon, Dain University of Teeloy. Daan. chine “Sports and Exerc Meine, Queen Mary Univers Landon, Mie Ed Hospi london “Soot of Mecha Engnering, Sates Janong Une, Cheng China “site or Biomech, ET Zar Zar Sazetond ARTICLE INFO ApsTRAcT ‘Accepted 31 uly 2019 ‘Representation of realistic muscle geometries ie needed for systematic biomechanical simulation of mis caloskeetal systems. Most of the previous musculoskeletal models are based on multibody dynamics simulation with muscles simplified as one-dimensional (1D) line-segments without accountng for the large muscle attachment areas, spatial fibre alignment within muscles and contact and wrapping, between muscles and surrounding tissues. In previous musculoskeletal models with three-dimensional (GD) masces. contractions of muscles were among the inputs rather than calculated. which hampers Posen peer, {he predictive capability ofthese models. To address these issues, a finite element musculoskeletal model ——ie ‘with the ability to predict contractions of 3D muscles was developed, Muscles with fealistic 3D geometry, oa Spatial muscle fibre alignment and musele-musele and muscle-bone interactions wore accounted fo. Active contractile stresses of the 3D muscles were determined through an efficient optimization approach based on the measured kinematics ofthe lower extremity and ground force during gait. Tis ‘model also provided stresses and strains of muscles and contact mechanics ofthe muscle-mustle and :musce-bone interactions. The total contact force ofthe knce predicted by the model corresponded well, {o the in vivo measurement. Contact an wrapping between muscles and surrounding tissues were evi- sent, demonstrating the need to consider 3D contact models of muscles. This modelling framework serves asthe methodological basis for developing musculoskeletal modeling systems in finite element ethod incorporating 30 deformable contact models of muscles, joints, ligaments and bones, © 2019 Elsevier Ltd, Al rights reserved. Kepwors: 0 fine clement muscle model Musculoskeletal made Introduction determine a unique solution of muscle forces, they have limited ‘accuracy. This is because aspects including large muscle attach- Biomechanics plays an important role in the function of mus~ les, bones and joints. Computer modelling of the musculoskeletal system is by fa the only non-invasive approach to predict biome- chanics of joints and deep muscles. Most of the previous muscu- loskeletal models are based on multibody dynamics simulation, incorporating three-dimensional (3D) rigid bones attached by ‘one-dimensional (1D) ine-segment muscles. Additionally, 3D con- tact joints have been calculated within several recent models (Adouni et al, 2012; Chen et al, 2015). Although the models with AD muscles are efficient to solve the redundancy issue, ie. muscles ‘outnumber equations of equilibrium requiring optimization co sponding authors at: Schol of Science and Technology, Middlesex ty. London NW AH, UK mal adresses: jert@pmalcom UL yonstaolv@a eden (YW hnups:ioiort0.1016ijbiomech2019.07082 (002-5290) 2019 Eee Al ght reserve ‘ent areas, spatial fibre alignment within muscles, and contact and wrapping. between muscles and surrounding tissues play important roles in the mechanics of musculoskeletal models and ‘cannot be represented in 1D models (Webb et al, 2014). Impor- tantly, 1D models do not provide biomechanical parameters such ‘as stfesses and strains of muscles and contact mechanies between, ‘muscles and surrounding tissues that are important for investiga- tion of the function and degeneration mechanism of musculoskele- tal systems. As such, more detailed models are required that incorporate 3D muscle representation, By coupling boundary conditions between a rigid muscu- loskeletal model and a deformable knee model, predictions of deformation and contact mechanics in the knee were enabled (Halloran et aly 2010; Shu et al, 2018). However, this approach is not suitable for 3D muscles, because boundary conditions of|. ea Journal of Bomechanc 8 (2019) 230-234 m 1D and 3D muscle models cannot be realistically coupled. Muscu- Joskeletal models incorporating 3D muscles have been developed in Bnite element method, aecounting for spatial fibre orientation nd interactions between muscles and bones (Webb et al, 2014; Zoliner et al, 2015; Mo et al., 2018), However, the redundancy issue of muscle contractions has not been addressed in the pres ‘ous 3D muscle models, which hampers the predictive capability, of these models Inorder to address the above issues, the aim of this study was to develop a finite element musculoskeletal model with the ability to predict contractions of 3D muscles. Muscles with realistic 3D geometry, spatial muscle fibre alignment and muscle-muscle and ‘uscle-bone interactions were accounted for. The redundancy issue of muscle contractions was solved through an efficient opti ‘mization approach. Additionally, the total contact force ofthe knee predicted by the model was compared to in vivo measurement data for validation. 2. Methods The 3D geometric models of muscles and bones used in this, study are based on the TLEM 2.0 database (Carbone et a, 2015), ‘The model includes bones of the right lower extremity Le. pelvis, femur, patella, tibia/ibula, talus and foot, connected by the hip, ‘knee and ankle (Fg. 1). Only the principal muscles involved in flex- ion and extension of the knee including quadriceps (ie. rectus femoris, vastus lateralis, vastus intermedius and vastus medialis) and hamstrings (Le. semitendinosus, semimembranosus, biceps femoris caput breve and biceps femoris caput longum) were incor- pporated in the model to enhance computational efficiency. The regions of tendon and muscle tissues were identified and some missing tendons connecting the muscles and bones were recreated in Solidworks (V2017, Dassault Systémes, France) using a combina- tion of MRI images, muscle geometry, bone landmarks and knowI- edge of anatomy (Drake et al, 2008) ‘The mesh was created in Hypetmesh (V2017. Altair, USA), with the bones and muscles represented using 36,162 four-noded tetra- hhedral elements and 18,028 eight-noded hexahedral elements, respectively. The mesh density was selected such that a change of less than 5% difference in muscle forces occurred if the number of elements were doubled. The insertionjorigin sites of the mus- cle/tendon tissues were rigidly attached onto the bones. Friction- less contact was defined for the muscle-musele and muscle-bone -_ / / / (eter om interactions. Computational efficiency was enhanced by assuming. the knee as a one degree-of-freedom hinge joint ({lexion/exten- sion) with the patella immobilized onto the tibia (Li et al, 2015), ankle and hip as three degrees-of-freedom ball-and-socket joints land bones as rigid ‘Muscles and tendons were defined as incompressible trans- versely isotropic Mooney-Rivlin material incorporating muscle fibres that align along the geometry of muscles (Fig. 1) (Weiss et al, 1996). This constitutive model was described in detail in the FEBio Theory Manual (Maas and Weiss, 2007). Active contrac- tion was incorporated into the muscle material model, with the total stress in the solid mixture («) as the sum of the solid stress ‘due to strain (a*) and the active contractile stress (a) on-o+e a ‘The active contractile stress was uniform across all the ele- iments in each muscle, Parameters of the constitutive model are shown in Table 1. Boundary conditions of the model are based on the fifth Grand Challenge Competition in which the gait data and in vivo measured knee contact forces of a subject were provided (Fregly et al, 2012), The femur was constrained in all degrees of freedom, Based on the trajectories of markers in the gait data, angles of the hip, knee and ankle of the subject during normal Speed walking were derived in Visual 3D (V6; C-Motion, USA) and then used as inputs for the rotation of joints in the muscu- loskeletal model. The ground reaction force was applied to the foot. Inorder to solve the muscle redundancy issue, active contractile stresses of the muscles in the finite element musculoskeletal ‘model were optimized until (1) the cost function (Le. sum of mus- cle active contractile stresses cubed: 5-1 (*)’ (Adouni et al., 2012)) was minimized and (2) the ‘resultant knee moment approached zero (<0:3N x m). The optimization was based on the muscle active contractile stresses and the corresponding joint ‘moments in the FE model. Analyses were conducted at 15 evenly distributed time instances of the stance phase of a gait, starting from hee!-strike (0, 08) to toe-off (0.7 s, 100%). To enhance computational efficiency, the finite element model at each quasi-static time instance (eg. at 0.5 s) in the optimization process was simulated based on the model at the previous time instance (ie. at 0.45 s) in which the optimization criteria were achieved, rather than starting from the original state (ie. at 0). Finite element modelling was performed in FEBIo (V2.6.4; hitp:// {ebio.org/febio), MATLAB (R2017a, Mathworks, MA) was adopted x Mosh Fro oriontaton (grey aon), Fix. 1. the fine clement musculoskeletal model with 3D muscles, bre entation in the muscles was ilstated by the grey arows. The pricipl muscles iol in ‘lexionjextesion ofthe knee were incorporated inthe model, incudng rectus femoris ght green), vast lateralis (dark red). vasts intermedi (ight Bue). vast edi ar i semitendinos (orange) emimembranoses (pink) beeps femoris caput reve (Ask bie) and biceps femoris caput Tong ight re). (oe references t olor in hs igure legend, the ear i welrred othe web version a hs ate)2 |. et a /ournal of mechan 94 (2019) 220-234 Constitutive rode arreters. The vals were based on previous ties (ener fetal 2005: eran et al, 2005; Rave and Pallan, 2007; My et al 2018). These Mania! (V26) (Milas and Wess 2007) yy ad af the mis ad Iasmum values of the active contact Sess (oH) tespecively used a the optimzaton procure Mascle constants “Tendon constants 015 Mra G 15 G for the optimization ({mincon function) and automation proce- ‘dures, Active contractile stresses and tensile stresses of the mus- ‘les and contact pressure of the muscle-musele and muscle-bone interactions were analyzed. To assess the validity of the model. the total contact force ofthe knee predicted by the musculoskeletal ‘model was compared to int vivo measurement data of the same subject, 3. Results ‘The time instants at heel-strike, weight-acceptance, mid-stance, push-off and toe-off between the model predictions and the exper- imental measurement were closely comparable (Fig. 2). Compared to the experimental measurement, the predicted joint force was 1% higher at weight-acceptance, 21% lower at mid-stance and 12% higher at push-off, with a mean absolute percentage error of 11%. ‘As shown in Fig 3, the hamstrings were primarily activated dur- ing heel-strike, while the quadriceps played the major role after- wards until toe-off. The maximum active contiactile stress ‘occurred in the vastus lateralis at push-off. The tensile stresses ‘were concentrated in the tendon regions in which the tissue was thinner than the muscle regions (Fig. 4). The maximum tensile stress occurred in the hamstrings at heel-strike, while the mini- ‘mum tensile stress of the muscles occurred at mid-stance. The 2000 1600, ‘200 200 400 Total knee contact oc ( —— Boerimental O% KAO OB T00K Stence Heel Weight Ma Puh OT site acceptance since off of Fig. 2. Comparison ofthe pected total contac forces of the ke andthe in vvo easement data Te time nan thet ste, weight acceptance, mista, Push-olf apd tocalf Beeween the model predictions andthe experiment Ineasuement were closely comparable. Compared othe experimental measure rent the predicted Joint force was Thier at weight acepance 21% Tower st Inutance and 12% higher a pest, witha mean alte perentage err of recta —Seubcen See eae TVetenteress Seater cate eee, Sooamsese os Ea bos Boe 8 go 0 oe wom Stance ig. 3, active contractile stesss ofthe muscles determined theaugh optimisation, The hamstrings were primany activated daring helt, we the qusicns Played the major rele afterwards ntl taco. Anases were conducted 3t 15 ‘evenly strbuted time instances ofthe stance cle ‘maximum tensile stress of the quadriceps occurred at toe-ofl. As illustrated in Fig. 5, the muscle-muscle and muscle-bone contacts were evident. The level of contact pressure was minimal compared to the tensile stresses of the muscles. 4. Discussion In this study, a finite element musculoskeletal model with 30) rmuscles andthe ability to predict muscle contraction was devel- ‘oped! for the first time. This new model provided stresses and strains of muscles and contact mechanics of the muscle-muscle and muscle-bone interactions, thus allowing for more systematic biomechanical evaluation of the musculoskeletal system compared ‘with previous musculoskeletal models with 1D line-segment muscles “The muscle redundancy issue that hampers pervious 3D muscle ‘models was addressed through the optimization approach. There- fore, prediction of active contractile stresses of 3 muscies was ‘enabled, In the optimization proces, the finite element model at ‘each quasi-static time instance was simulated based onthe model atthe previous time instance, rather than starting from the original time instance By doing so, computational efficiency was improved by approximately 105 (ie. $4) times for a quasi-static simula tion including 15 analysis instances In this study, optimization was performed directly to the finite ‘clement musculoskeletal model. Any region of the musculoskeletal ‘model including muscles, joints. ligaments and bones can be set as 3D deformable contact madels with ease Due tothe large number ‘of degrees of freedom involved, the simulation was completed in approximately 5 days using a Windows 10 desktop with 32 GB of RAM and 16 Intel E5-2699 cores at 2.2 GHz. Further development ‘of the model will focus on enhancing the computational efficiency though a hybrid model incorporating both 3D and 1D muscles once the muscles that can be simplified as 1D without compromising accuracy are targeted. Generally, the knee contact force predicted by the computer model corresponded reasonably well to the in vivo measurement. ‘The difference in comparison might be due to the model simplifi- ‘ation described in the paragraph below and the musculoskeletal ‘variations between the subjects ofthe two database, apart from ‘errors of the in vivo measurement. Although muscle contractions can be reflected by knee contact forces to some extent, active|. ea Journal of Bomechanc 8 (2019) 230-234 mm Wu A Max Max Mae eure 130 MPa) ATMPe Hetstite Wight scomptanse Mista owe Mae Max our na5MPa 147 Mea Pushott Toe of Fig. 4. Tense stresses (Mf inthe muscles athe characteristic gai phases Tense streses were concerted inthe tendon regions connecting the bones. he maximum tens stress occured i dhe hamstrings a else, while the muni tense stress ofthe musees occurred at mi-stance. The maximum tensile sesso the ‘quadriceps was found at toe a Max 015MPa oMPa | 018MPa Fig. 5. Contact pessre (MP) af the semimembransus in contact with he distal femur aneior vi) and the semiteninass (bs poster Ww) at Ret see, aa contractile stresses of individual muscles were not compared to the experimental data such as electromyography (EMG) signals ‘due to uncertainties in acquisition and conversion of EMG signals nto muscle contractions. Future imaging measurements might serve as alternative effective approaches to validate deformation ‘of 3D muscle models during motion. The presence of contact and ‘wrapping between muscles and surrounding tissues found in this study further demonstrated the need to consider 3D contact mod- els of muscles, which is in agreement with the study of Webb et al. (2014), ‘The primary aim of this study was to develop the necessary ‘modelling methodology for incorporating 3D muscles into muscu- loskeletal models with the redundancy issue of muscle contrac- tions addressed. Whilst this was achieved, there were some limitations First, strains in the muscles were assumed to be 2er0 in the original model configuration, These data would be provided by future experiments measuring intial stretches within muscle fibres (Webb et al, 2014), Future experiments are also needed to offer more realistic data to define the regions of tendon and muscle tissues and orientation of fibres and non-uniform activation within ‘muscles for these models. Constitutive madels involving the tela- tionship between muscle contraction and fibre stretch will be accounted for in future studies. Although the muscles considered in this model play the principal role in flexion and extension of the knee (Lieb and Perry, 1968; Fukunaga et al., 1992; Jenkins, 2008; Yeow, 2013; Lube et al, 2016), incorporation ofthe complete muscles in the lower extremity would further enhance the ‘modelling accuracy. Contact joint models (Chen et al, 2015: Shu ct al,, 2018) wil aso be incorporated in future studies to enhance the modelling accuracy. ‘The finite clement model was solved quasi-statically through an implicit solver ignoring the dynamic effect, because the implicit ‘method is more reliable than the explicit method for dynamic anal- yses involving relatively low-speed activities owning to its itera- tive approach (Naghibi Beidokhti et al, 2016). Additionally, the ‘quasi-static approach is more efficient, as simulations can be per- formed only at the time instances that are of interest rather than for the whole petiod. It should be noted that the optimization approach developed to solve the muscle redundancy issue can be also applied to explicit models, because both the simulation and ‘optimization at each time instance were performed continuously based on the optimized results at the previous time instance. This ‘modelling framework serves as the methodological basis for devel- ‘oping musculoskeletal modelling systems in finite element method incorporating 3D deformable contact models of muscles, joint, ligaments and bones. ‘Acknowledgements ‘his research was supported by Newton Fund 2017-RLWKS- 10075, The authors appreciate Zhenxian Chen for advising oncory {let four of mechan St (2018) 230-234 processing the Grand Challenge Competition data and Louis ‘Slabbert for his assistance on the computing facilities. Declaration of Competing Interest ‘The authors declare no contlic of personal or financial interests. References. ‘Adour. M. shiz A Sian R. 2012. Computational bodynamcs of human nc ent nga rom muscle ices tocatlage tresses | omech 13,2149 Bemker. SS Pinsky. PM. Delp. Si, 2005. 30 modelo muscle revels the causes of nonnifom stain inthe biceps rach J Bmech 38, 057-66, ‘carbone ut Pelikan, Pan der krog, MM. Janssen Damar M ‘Vigneron. Fits, Kooptan, HF}. Verdonscat.N_ 2018, TLEM 20 8 ‘YaGordr a compredeasive, musculoskeletal geometry dite for subject Specie modting of lower extremity.) Biomech 48.734-741. ‘che, 2 Wang. MY He, Jah. QU Dim 2 2015. et fcmponent mal Totton on Ke long (tal knee ahoplsty sing mult bedy yams ‘modeling under 2 simulated walking gl | Ortop. Res. 3, 1257—1206. brake. Re Vag A= Mitchel, AW, 2008 Cys Atom for Stents E-Book Hever Health Science Frew Bl Bese TF, Hoya, DG. Delp, SL. Banks SA. Pandy. MC, Dlima, DO. 2012. Grand challenge competion to predict i vivo re Tos Orthop. Rs. 3303-543, Fukunaga, T Roy, RA Shlloc, FG, Hodgson, A Day, MLK Lee, P, Kwong TH Bigeton, VAR, 1902 Payload cons sectional aes of human es muscles base on magnetic resonance imaging | Orthop. es. 10.826-934, Hatoran 1 Ackermann, M. Edema, A, van den Bogert A. 2010, Concurent Imoscueskletal dymamics and fie clement analyse pros aered at plteras to seduce foot sue loading | Buomech 3, 2610-2615. Jenkins DB. 2008 lnsheads Functional Antony the Lbs and Back LU. J, MeWillams, AB. Jn, 2, Per, Stone, ML, Redmond, AC, Stewart, TD. 3015: Unateral toa hip replacement pavers with symptomatic ey Lengt nguaty ave bora ip bones dng ming Cn ote 3 Lie ery J 1968. Quadriceps Futon: an anatomical and mechanical ty "sng amputated limbs. Js 90. 1535-1538 tbe, J Caton, S. Bechrane Mil, 1 Orkurl 0. Saal Steinke tidmmer,N~ 301 Reference data on musi vlume of healthy Baan et ane ower extremity mosees amin vivo magnet resonance imaging east) Sry Sng. nat 38, 97-108. Maas, SA. "Weis. JA. 2007. FEB Theory Manual uoiwaraebim Mo. Lc, Bt M Xba, 2, Zhang. G, Bu X, 2018 A lower Unb-plis fe ‘semen me with 3 ative muses. Ann Biomed. Eng 46 86-9. Naghii Redo Hanssen. D. Khoshgnar M. Sprenbes. A, Pedaheog ES, ‘Van Den Boogaard. Verdonschot N. 2016. A compantonberween dyRamic Impl and exc fine element simulatons of the native Kae ot. Med Eng Phys 38, 1725-1150. oe 0 Pian AJ 2007. Tree dimensional ite element modling of muscle Fores during macticatin.J Biomech 40, 3363-3372 ‘hus, Yaramete Kao, Saraswat PL. Ye Nitsuishi M. Sugita, N. 2018.4 ‘subjeer specie fine’ clement museuloseltal Tamawork for mechanics aul of tone replacement lomech 77, 145-156 ‘eran Jaks €,Blemker, SS. NavThow Hing. V, Lay, C. Fed, R, 2005. ‘eating and simulating skeletal muscle fom the visible human data Se EEE ‘Transactions on Visualzation and Computer Graphs 1.317-328. ‘Webs 1D, Blember, 88, Dep, SL, 2014 3D fie element models of shoulder ‘muscle for computing tins of actions and moment arms Comput: Methods Bec home Eng 17, 829-857, ‘Webs, JA. Maker, BN, Govind, 8, 196. Rite element implementation of Incompresbe transversely stropi hypersasciy, Comput. Methods AP. ech ng 135, 107-128 ‘You, CH. 2013. Hamstings and quadiceps muscle contributions to enecsy "fverton a sspation atthe kee nt dung stance, Sing ate ight Dass of evel runing Knee 20, 100-105. ‘alper, AM. Pk, JM. MeWalter. EJ. Gold GE. Kuhl E2015. On highheels 2nd Short mics!» mulsle made! for sarcomere lose in the getraeneis muscle Theor Be 385, 3010, pines. utah