Incorporating in vivo load variability in modelling

Introduction:

Abrasive wear and delamination are frequently observed in studies evaluating the wear behaviour of total knee arthroplasty (TKA). Up to now there is a mismatch in the outcome of retrieval analysis and laboratory tests with regards to the frequency of the observed wear mode. Highly demanding daily activities together with an aging protocol can be used to increase the probability of delamination for conventional Ultra High Molecular Weight Polyethylene (UHMWPE) in in vitro tests due to increased stress and degraded material properties [Schwiesau et al. 2013, Grupp et al. 2017].

New materials are developed to increase the delamination resistance of UHMWPE by reducing the susceptibility for oxidation. Irradiation processing parameters and additives like vitamin E can influence the material properties and consequently the wear behaviour [Oral et al. 2012, 2016].

In this study highly demanding daily activities are simulated to evaluate the influence of radiation condition and duration of artificial aging on the wear behaviour of an TKA bearing material blended with vitamin E.

 

Material:

UHMWPE with 0.1 wt% vitamin E was radiated with ~ 30 kGy at two different temperatures. Oxidation according to a current standard protocol was applied for two and six weeks. The tested TKA design was cruciate retaining with CoCr femoral components.

 

Methods:

High flexion activities [Kutzner et al. 2010, Bergmann et al. 2014] predominate in the simulation (40% stairs ascending, 40% stairs descent, 10% level walking, 8% chair raising, 2% squatting). Five million test cycles are simulated with standard environmental conditions [ISO14243-1]. For each test group three specimens are tested. Artificially aging of the UHMWPE was generated according to ASTM F2003-2. The gravimetric wear was evaluated according to ISO 14243-2. The wear mode was analysed optically.

 

Results:

Abrasion is the predominating wear mode. No case of delamination was observed. Significant differences are observed between the samples irradiated at elevated temperature compared to conventional radiation conditions. No influence on the wear rate was observed for the duration of the oxidation process.

 

Discussion:

Based on the simulation of daily patient activities it was possible to differentiate the wear behaviour between the radiation conditions of the bearing materials. The tested materials show a high delamination resistance compared to conventional UHMWPE. Crosslink density is influenced by the radiation temperature and associated with the wear behaviour.

 

References:

Schwiesau et al. Med Eng Phys. 2013 Aug;35(8):1204-11

Grupp et al. Acta Biomaterialia 48 (2017) 415–422

Oral et al. J Biomed Mater Res Part B 2013:101 B:436-440

Oral et al. J Biomed Mater Res Part B 2016:104 B:316–322

Kutzner et al. Journal of Biomechanics 43 (2010) 2164–2173

Bergmann et al. (2014) PLoS ONE 9(1): e86035

Introduction

Musculoskeletal force prediction often involves the application of complex computational models using cost functions that implicitly convey the modeller’s prior assumptions about the mapping from task space to force space. It has been proposed that the human central nervous system uses supervised, error-based learning to solve this mapping (1), with advantages in speed and flexibility. A similar approach may bring analogous benefits for musculoskeletal force prediction. To test this hypothesis, deep neural networks were trained and pitted against recent performance benchmarks of the state of the art in musculoskeletal modelling for the prediction of in-vivo force data.

Methods

Data from the six Grand Challenge competitions (2) to predict in-vivo knee forces (2010-2015) were used to train and evaluate deep neural networks. For each competition year, a separate model was trained using trials taken from all other years, strictly excluding any competition data from the training set. To formulate the input to the network, three-dimensional ground reaction force (GRF) vectors from each time step of each training trial were concatenated to produce a single training matrix of dimensions [time, 3]. A corresponding target vector was produced by concatenation of the scalar values for the medial knee joint reaction force (JRF) measured from an instrumented knee prosthesis at each time step. These tensors were used to train a four-layer feedforward neural network by minimisation of root mean square error (RMSE) between its predictions of the JRF and true target values. Network hyperparameter optimisation was performed by evaluation using a subset of the training data. Final evaluation of network accuracy was performed against the two unseen trials that were used as the test trials in the true competitions.

Results

Neural networks beat the winning submissions in four of the six competitions. The average time required for prediction was 71milliseconds.

Discussion

Supervised models can implicitly learn a mapping from task space to force space in a way that mirrors proposed mechanisms of human motor learning, using data, and contrasts with the heavily engineered solutions favoured by musculoskeletal models. Deep neural networks are often assumed to require large amounts of data to be effective but they were able to outperform the state of the art in musculoskeletal modelling here with a very small dataset, and using only the GRF as input. By allowing for accurate, instantaneous prediction of internal forces without the need for costly kinematic capture, and as more sensor data become available, this method may enable more and wider applications of clinical force estimation.   

References

1. Marr D, J Physiol, 202(2):437-70, 1969.
2. Kinney, A, et al, J Biomech Eng, 135(2), p.021012, 2013.

The lightweight construction of the human musculoskeletal system results from a complex interplay of different mechanisms of action. The mechanisms of action for themselves and in combination are not yet fully understood and topics of current research.

A deeper understanding of the mechanisms of action in the musculoskeletal system can be used for the development of prostheses, in rehabilitation technology and sports biomechanics. Furthermore, a detailed knowledge of the functional adaptation and remodeling of bones can contribute important insights and predictions for bone fracture healing and integration of implants into bone. Since muscle forces cannot be measured and the mechanisms of action of functional remodeling are unknown, biomechanical simulations are used in biomechanical development and examination processes.

This paper presents a systematic overview of the state-of-the-art of science on mechanisms of action in the musculoskeletal system to summarize the current state of research and to provide a basis for new simulation models.

It is well known, that the lightweight construction of the musculoskeletal system is achieved by the interaction of force and motion elements (i.e. muscles, bones, tendons, joints and ligaments) and the regulation of these elements by the sensorimotor system.

It is shown, that lightweight construction, in addition to the control functions, is achieved by means of the following effects acting on a structural level: hierarchical bone formation, functional adaptation of bone and muscle to strain as well as tension chording via muscles and tendons. This interaction is not arbitrary, it is achieved by a coordinated interplay of the mentioned mechanisms with the aim of minimizing bending moments and thus mass and resources demands of the organism. The lightweight mechanisms are in a dynamic equilibrium (homeostasis). It is assumed that the musculoskeletal system is thus in an optimal state of energy consumption and structural use. Due to the complex control behavior of the musculoskeletal system, it is not possible to observe from the outside how the system manages to regulate optimally. Because of that, the influences of the different effects and their interactions are not yet determined or understood in detail.

By the systematic overview, the restructured state-of-the-art of science is transferred into relationship diagrams in order to illustrate the mathematical connections between the mechanisms of action in the musculoskeletal system. With the help of this diagrams, a new modelling approach based on mathematical optimization methods is being developed, which can be used to draw conclusions about the interplay of the mechanisms. This contributes to a better understanding of the individual effects from which future musculoskeletal simulation models can benefit.

Introduction: The simulation of daily living activities using musculoskeletal models can provide insights into the articular loading of healthy and diseased joints. Loading conditions estimated for young, healthy subjects can be used to inform the design of implants intended for mobile, active patients, e.g. hip resurfacing. In this context, we aim to quantify the magnitude and directionality of the hip contact forces (HCFs) in a small cohort of young, healthy individuals while walking at self-selected speed (WN) and stair climbing (SU).

Methods: Eight subjects (males: 4, females: 4; age: 26.6±2.2 years; mass: 66.5±11.8 kg; height: 172.8±11.6 cm) without previous history of articular pain or diseases, were recruited for this study. The anthropometry, kinematics and ground reaction forces of each participant were recorded in the Human Performance Laboratory of Charing Cross Hospital. Kinematics was measured using a full body marker set (57 markers) recorded by a 10 cameras Vicon system.  A walkway and a staircase (three steps, step height 15 cm, step depth 25 cm, inclination: 37°), instrumented with three force plates (Kistler, Type 9286BA), were used for recording WN and SU. A generic musculoskeletal model of the lower limb [1], was used to simulate these tasks in OpenSim [2] and compute joint angles, joint moments, muscle forces (by minimizing the sum of muscle activation squared) and HCFs.

Results: HCFs for a representative participant are presented in Table 1, in terms of magnitude and orientation with respect to the femoral reference system. The maximum HCFs were on average 3.4 BW (second peak) for WN and 2.95 BW for SU (first peak), with direction of contact forces varying the most during stair climbing. Mean HCF peak magnitude was 3.5% larger for SU than for WN.

Discussion: Compared to in vivo measurements from hip instrumented prostheses implanted in elderly population [3], HCFs at first peak increased of 19.8% and 17.5% for WN and SU respectively. Differently from [3], however, the largest HCF for WN occurred at the second peak, due to larger hip flexion moments than in the elderly patients. The observed increase of maximum HCFs from WN to SU was comparable to in vivo (5.5%) [3], despite a 16° more inclined staircase. Finally, the computed HCF vector pointed considerably more laterally than measured in vivo, where the inclination in the frontal plane is generally within 12-16° [3], due to the action of hip abductors.

Conclusion: HCFs estimated in a young healthy population lead to consistent magnitudes, but not directionality, with in vivo measurements.

References

1. Modenese L., et al., (2011). J Biomech, 44:2185-2193
2. Delp S.L., et al., (2007). IEEE Trans Biomed Eng, 54:1940-1950
3. Bergmann G., et al., (2001). J Biomech, 34:859-871.

 

Introduction: Biomechanical joint contact pressure distribution measurements have proven to be a very valuable tool in orthopaedic research to validate musculoskeletal models. The K‑scan pressure sensor from Tekscan (South Boston USA) is a commonly used device for this purpose. Despite the large interest in the sensor, the effective measurement accuracy for in vitro biomechanical joint contact measurements still remains uncertain. Drift behaviour of the sensor, varying contact areas and contact shape are reported as major suspects to affect the measurement accuracy. The goal of this research is to quantify the impact of those factors on the measurement accuracy. A generic approach is used to evaluate the effect of drift behaviour while for the contact area and shape, the research is focussed on tibiofemoral contact with implants.

 

Methods: Four sensing pads Tekscan 4000 are used. To evaluate the drift behaviour, the sensor is loaded in a hydraulic pressure device (Figure 1) providing a homogeneous pressure up to 30 MPa.

The sensor is cyclically loaded and unloaded with a static pressure of 15 MPa and a duration of 15 min. Each cycle, the drift behaviour is quantified and the average output is used as a measure for the sensitivity. After optimal preconditioning, the sensor is equilibrated in the same pressure cell and calibrated between a flat HDPE block and a grinded metal sphere to obtain optimal calibration conditions: similar material, flat surface and large contact area (Case 1 on figure 2).

To evaluate the effect of contact area, curvature and surface motion, four cases are considered: 1. sensor loaded in calibration conditions, 2. identical to case 1 but with reduced contact area, 3. with curved surfaces by using tibial and femoral implants and 4. case 3 with dynamic flexion motion of the femoral component. In every case the total measured force by the Tekscan sensor is compared to the applied force.

Results: Figure 3.a shows stabilization of the sensitivity after 10 (SD=2) cycles. An increased sensitivity of 21.1% (SD=6.2%) is observed after preconditioning. The short term sensor drift halves after 1 cycle and has a stable minimum after 8 (SD=3) cycles (Figure 3.b). After optimal sensor preconditioning, equilibration and calibration, a measurement accuracy of 1.2% (SD=1%) was achieved. Reducing the contact area in case 2 lowered the measurement accuracy till 5.5% (SD=4.5%). Introducing curvature in the contact (case 3) did not affect the measurement accuracy significantly. Adding relative motion to the implants increased the measurement error up to 17.9% (SD=10.8%).

Conclusion: By omitting preconditioning at a representative pressure level, the sensor will precondition locally, leading to unpredictable errors up to 21.1% (SD=6.2%). In combination with realistic dynamic loading with knee implants, errors up to 39.0% (SD=12.5%) can occur.

This invited talk will discuss how in vivo knee contact force measurements, made available by instrumented knee prostheses, have improved the field of neuromusculoskeletal modeling. The talk will be organized around the following three questions:

Question 1: Why develop neuromusculoskeletal models? This part of the talk will discuss how neuromusculoskeletal models that can predict internal body forces accurately could be useful to improve the design of clinical treatments for movement impairments.

Question 2: How have knee contact force measurements benefitted neuromusculoskeletal models? This part of the talk will discuss how knee contact force measurements have revealed deficiencies, encouraged improvements, and elicited participation in neuromusculoskeletal models.

Question 3: Where can knee contact force measurements improve neuromusculoskeletal models further? This part of the talk will discuss how knee contact force measurements have revealed the need for improved methods for model calibration and more realistic methods for model control.

The final portion of the talk will demonstrate how prediction of walking motion improves as the model control method becomes more realistic and subject-specific, which would likely lead to improved prediction of internal forces in the body as well.

Subject-specific Finite Element (FE) models of bones from CT data are now able to predict strain levels in human bones with a good accuracy, as demonstrated by numerous validation studies. Their applications have been manifold, characterised by different approaches and different level of automation. In all cases, a key aspect is the identification of the boundary conditions, i.e. the loads that act on bones during human movement.

Differently from what happens in experimental set-ups where loads are accurately controlled or determinable, in physiological, and even more in pathological, conditions loads are highly variable.

This is true in populations studies, where inter subject variability is a key issue by definition, but it is also true in the other extreme case of a complete subject-specific simulation of a particular subject where intra subject variability cannot be usually neglected.

Independently from the specific application of FE models of bones, the possibility of incorporating variable loading conditions in the model is of paramount importance.

Aim of the talk is to provide evidence for the importance of the inclusion of load variability in the simulation with subject-specific FE models of bone biomechanical behaviour; to illustrate, through a number of applications, different strategies for incorporating such variability into the models, depending on the scope of the study; to identify some possible future developments to further improve the loading assumptions in the FE modelling of bones to target still unresolved challenges.

Introduction

The long term success of uncemented humeral stemless implants relies on adequate primary stability, which can, among other parameters, be influenced by implant design.

The objective was to compare the primary stability of two different humeral stemless implant geometries, in the anatomical configuration, in response to loading conditions representative of upper limb daily life activities, and assess if a design may be better suited for a given set of activities.

Methods

This finite element study was based on an established method [1]. Two commercially available stemless implants (subject to regulatory clearance per region) were virtually implanted in a foam bone block. One implant anchor consists of 4 open and thin wings with a superior collar while the other exhibits 6, thicker wings (Figure 1). In order to isolate the effect of geometry, all parameters but the geometry were kept the same for the two designs. Generic linear-elastic material properties were applied to the cancellous bone (E = 213MPa) and to the titanium implant anchors (E = 114 GPa). A 0.6 friction coefficient was selected for the bone/implant interface [2]. Physiologic glenohumeral joint resultant forces and moments representative of 29 different upper limb activities measured with instrumented shoulder implants [3] were applied to the anchor directly (humeral head not modelled). Micromotion for each design during each activity was assessed, and a linear regression analysis was performed to identify the loading components most affecting micromotion.

 

Results

Both designs behave in the same manner with respect to the different activities. Both designs were most sensitive to the anterior-posterior shear force component (Figure 1) and to the axial rotation moment component (along the humeral axis). The same 7 activities (combing, lifting 10kg, high and loaded elevation or abduction, painting, hitting a nail) were found in the 10 upper limb activities with the highest AP shear component and in the 10 activities with the highest axial rotation moment arm.

 

Discussion

Although both implants have very different geometries, the features they do have in common (e.g. the wings perpendicular to the resection plane) may make them respond to the different activities in a similar manner.

The activities that most affected primary stability by inducing a high shear force and a high axial rotation moment involve a loaded hand and/or a hand away from the trunk.

This study focussed on the effect of design alone, neglecting the effect of different surface treatments, press-fit or surgical procedure. These can be accounted for in the next steps to provide input for the next designs and/or guidelines for the rehabilitation period.

 

References

[1] Favre and Henderson, Clin Biomech 2016

[2] Grant et al., J Biomech 2007

[3] https://orthoload.com/database/ > Shoulder Joint (last visited on Nov 2017)

Introduction

Patellofemoral mechanics is often evaluated using a small number of subject-specific finite element models in the literature [1]. While these offer great insight they are expensive to create requiring MRI and or CT for each subject and are not possible for large cohorts often required in clinical studies. A recent alternative is to adopt existing databases of imaging and functional anatomy data and build population-based models that represent specific cohorts.

In this study we adopt population-based modelling to evaluate the influence that patellar morphology and quadriceps forces have on joint contact. We hypothesize that patellar shape and quadriceps force may explain why a large percentage of the population have increased joint loading on the lateral patella facet while a small percentage have medial or central facet loading.

Methods

An existing dataset of 50 segmented patellofemoral joints complimented by patient motion capture data was adopted from a previous MRI study [2]. A generic mesh was morphed to each knee for consistency in statistical analysis using free-form deformation [3]. A principal component analysis (PCA) was conducted using the open-source software, the Musculoskeletal Atlas Project (MAP) [4]. The MAP client produced finite element models of each PCA mode in order for joint contact to be simulated in ABAQUS. Von Mises cartilage stress was predicted.

Results and Discussion

PCA analysis revealed that the population of 50 knees was well represented by 4 modes giving 9 finite element models (mean + 2 standard deviations (SDs) for each mode). This represented 87% of the population variation. Mode 1 was most dominant and accounted for size and patella tilt representing 38% of the population (Figure 1a & b). Cartilage contact Von Mises stress patterns resulting from patellofemoral joint loading was exhibited primarily on the lateral side of the patella and femoral groove. Moreover, 77.8% of patellofemoral joints presented loading on the lateral side of the patella, while 22.2% showed central-medial loading due to morphology (Figure 1c). Subject quadriceps force increased this trend to 82% on the lateral side and 18% on the central-medial side. We showed that patellofemoral morphology was the primary factor contributing to the dominant lateral facet loading of the patella, the most commonly loaded and damaged cartilage site. Quadriceps force slightly increased this trend.

 
Figure 1: (a) Joint size and (b) patella tilt variation in the population with blue (+2SD) and red (-2SD); (c) Variation of joint cartilage contact pressure from lateral to central-medial.

 
References

1. Akbarshahi M., et al. Med Eng Phys, 36(9):1122-33, 2014.
2. Besier T., et al. Med Sci Sports Exerc, 47(11):2416–2422, 2015.
3. Fernandez J., et al. Biomech Model Mechanobiol 2(3):139-55, 2004.
4. Zhang J., et al. Biomedical Simulation. Springer Series. 8789:182-192, 2014.

 

Introduction

Bone deformation plays a decisive role in bone adaptation. Insufficient mechanical loading on bone can lead to osteopenia. However, to date, the current understanding of the in vivo human bone loading regimes, strain amplitude and distribution are still limited. It remained unclear that how different locomotive activities affect the mechanical loading acting on the bone in vivo. Combining the in vivo bone strain measurements in humans and finite element analysis, the purpose of the present study is to comprehensively understand the human tibia loading regimes, bone strain amplitude and distribution during different locomotor activities.

Materials and Methods

Tibia deformation was recorded in five subjects utilizing an optical segment tracking (OST) approach established previously in our lab. In detail, two marker clusters with three non-collinear retro-reflective markers on each cluster were affixed into the proximal and distal anterior-medial aspect of tibial cortex by bone screws. The markers trajectories were captured with a motion capture system during gait, jogging, stair ascent with forefoot and rear foot strike, hopping and squats. Tibia loading regimes and deformation amplitude, namely peak-to-peak (p2p) bending and torsional deformation angles, were computed from the relative movement of the proximal cluster with respect to the distal cluster. Tibia strain amplitude and distribution on tibia surface were analyzed using individualized finite element models of human tibia.

Results

The proximal tibia primarily bends to the posterior aspect (bending angle: 0.86° - 1.85°) and twisted to the external aspect (torsion angle: 0.65° - 0.90°) for all test subjects during walking. Jogging with forefoot strike has the potential to induce larger torsion and antero-posterior bending deformation than rear foot strike. The antero-posterior bending angles were 1.67°± 0.15° and 1.80°± 0.18° during rear foot and forefoot running. Comparatively, the torsional angle ranged between 1.72°± 0.08° and 2.25°± 0.21°. The medio-lateral bending angle of tibia remained at a rather low level. The impact of hopping led to an averaged peak bone strain value of 13827 ± 1980 µɛ. This was 13.1 times higher than the strains during standing (1058 ± 973 µɛ).

 

Discussion

Bending and torsion predominated the tibia deformation regimes during the investigated activities. Tibia strains are substantially greater than previously thought, in particular in the distal tibia, and especially during hopping, where they are up to two thirds larger compared to running or walking. The presented results suggest that forefoot running might be one of the most important exercises which will build the most resistant bones. These findings therefore are relevant to the development the training protocol as the potential countermeasures against osteopenia.

Introduction

Preclinical testing of implants considers THR patients a homogenous group. In reality, patients are heterogeneous and patient characteristics impact revision rates. Patients with lower BMI have a reduced revision risk [1] and different kinematics to patients with higher BMI [2],  indicating possible association between revision rates and function influenced by BMI. No study has stratified THR patients to understand how BMI influences hip contact force (HCF) at the bearing surface. This study aimed to identify differences in HCF in THR patients stratified by BMI.

Methods

144 THR patients, >12 months post-surgery, underwent 3D kinematic (Vicon, UK) and kinetic (AMTI, USA) analysis whilst walking at self-selected speed. HCF’s, normalized by body weight, were computed through multibody modeling (AnyBody Technology, Denmark) over one gait cycle. Patients were stratified into three BMI groups (<25, 26-30 and >30) and a mean for each patient was calculated from three to five walking trials. Differences between BMI groups were analyzed using one-dimensional statistical parametric mapping [3]. ANOVA was used to test for significant differences across groups. The test statistic SPM{F} was evaluated at each point in the normalized time series, and a critical threshold corresponding to an error rate of α= 0.05 was calculated based on random field theory. Supra-threshold clusters with their associated p-values were then identified.

Results
The mean HCF demonstrated the characteristic double hump-pattern for all BMI groups (Figure 1). Three different supra-threshold clusters reached or exceeded the critical threshold of F= 6.820 indicating systematic between-group differences, with the chances of observing similar clusters in repeated random samplings being p= 0.05, 0.015 and 0.044, respectively.

Discussion

The higher BMI group demonstrated a reduced normalized HCF at the push off phase of gait.  A reduced GRF at this phase in the gait cycle can indicate a compromised or abnormal gait [4]. This suggests that the patients with a higher BMI are functionally compromised compared to the lower BMI groups. Absolute HCFs in real-world situations are not normalized to bodyweight and the higher BMI group would be expected to have increased HCF compared to the lower BMI groups, with a likely consequence of this group experiencing both functional compromise and increased loads. Further analysis of absolute HCFs are to follow but the current data indicate at least some functional compromise in THR patients with higher BMI, which is not reflected in current preclinical testing.

Acknowledgments

Funding from the European Union’s Seventh Framework Program (FP7/2007-2013), grant agreement GA-310477 (Life-Long Joints)

References

1. Culliford DJ et al (2012) Osteoarthr Cartil 20 p519
2. Foucher KC et al (2015) Osteoarthr Cartil 23 p1685
3. Pataky TC et al (2012) Comput Methods Biomech Biomed Engin 15 p295
4. Muniz AMS et al (2009) Gait Posture 29 p31

Combined knowledge of the functional kinematics and kinetics of the human body is critical for understanding a wide range of biomechanical processes including musculoskeletal adaptation, injury mechanics, and orthopaedic treatment outcome, but also for validation of musculoskeletal models. Until now, however, no datasets that include internal loading conditions (kinetics), synchronized with advanced kinematic analyses in multiple subjects have been available. Our goal was to provide such datasets and thereby foster a new understanding of how in vivo knee joint movement and contact forces are interlinked – thus also providing an underlying foundation for guiding biomechanical interpretation of new knee replacement designs.
In this collaborative study, we have created comprehensive kinematic and kinetic datasets of the lower limb musculoskeletal system for worldwide dissemination by assessing a unique cohort of 6 subjects with instrumented knee implants (Charité – Universitätsmedizin Berlin) measured using a moving fluoroscope (ETH Zürich) and other measurement techniques (including whole body kinematics, ground reaction forces, video data, and electromyography data) for multiple complete cycles of 5 activities of daily living.
The cohort of subjects presented mean peak tibio-femoral joint contact forces during walking of 2.74 BW, 2.73 BW during sit-to-stand, 2.57 BW during stand-to-sit, 2.64 BW during squats, 3.38 BW during stair descent, and 3.39 BW during ramp descent. Internal rotation of the tibia ranged from 3° external to 9.3° internal. The greatest range of anterior-posterior translation was measured during stair descent (medial 9.3 ± 1.0 mm, lateral 7.5 ± 1.6 mm), and the lowest during stand-to-sit (medial 4.5 ± 1.1 mm, lateral 3.7 ± 1.4 mm).
The first sample dataset is now available online for public use in biomechanical and orthopaedic research and development at https://cams-knee.orthoload.com/. After a proprietary period for data analysis, the comprehensive CAMS-Knee datasets will become freely available for non-commercial usage. In order to download the full datasets, recipients will be required to sign a licence agreement, provide full name, position, and contact details, but also specify their intended usage of the data. With this information, we anticipate building a community of users, who will be able to interact, support each other, and even provide e.g. open source models based on the datasets. As a result, we expect the CAMS-Knee data to positively impact on current scientific and clinical approaches for the assessment and management of joint disease and injury, with tremendous potential for becoming reference datasets in medical innovation world-wide.