In this study, we implemented two widely used MS modeling software, AnyBody and OpenSim, in a knee joint MS-FE modeling workflow [15]. We assessed how the two MS modeling software affect the FE-model derived articular cartilage tissue-level mechanical responses. Complementing body- and joint-level MS findings, and supporting our hypothesis, our tissue-level FE comparisons revealed a great number of similarities between the estimates obtained from different models, but some discrepancies were also present. The AnyBody-driven FE model estimated slightly higher maximum principal stresses in the medial tibial cartilage compared to the OpenSim-driven FE model. In addition, the estimated CoP showed some differences between the models.
The estimated vertical JCFs were consistent between AnyBody and OpenSim MS models’ estimates for both subjects, agreeing with results about estimated tibiofemoral JCFs between the software from an earlier study [23]. On the contrary, during the late stance phase, OpenSim estimated another posterior force peak which was not the case with AnyBody. Acknowledging this difference is important, since the anteroposterior forces highly affect the stabilization of the knee by causing changes in the ligament tensions [48], which could lead to changes in the cartilage mechanical responses. The most notable differences in the MS modeling results were in the knee rotation moments, especially in the internal and external rotation moments, where AnyBody estimated large external rotation moments for both subjects during the late stance phase, whereas OpenSim estimated the moment to be close to zero. These results suggest that there is a considerable difference in the estimated muscle activations and forces during the second ground reaction force peak (75% of the stance phase). For example, the peak force of the semitendinosus muscle, a muscle that highly affects the knee rotations, was around 161 N in AnyBody for subject A, whereas OpenSim estimated the force to be 95 N.
As already pointed out above, a crucial explaining factor for the discrepancies in the MS outputs is the differences arising from the estimated muscle activations in the two MS models, leading to differences in muscle force estimates as shown in the study by Trinler et al. [22]. Demonstrating this, here OpenSim estimated the peak force used by the gastrocnemius medialis muscle to be around 2500 N for subject A, whereas AnyBody estimated it to be over three times smaller, only 722 N. Since the forces of gastrocnemius muscle have a large effect on knee JCFs, the differences in the estimated forces led also to the differences seen in the FE results, such as the maximum principal stress. An important note to consider when comparing the muscle properties is that the MS models used in this study are based on different cadaver studies [49, 50], which can lead to model-specific differences. For example, muscle properties obtained from different cadavers could affect the muscles’ strength, insertion points, and moment arms [23].
Since the MS outputs were used as inputs in the FE analysis, the resulting cartilage mechanics estimates were naturally affected by the differences in the MS results. The slightly higher estimates for the maximum principal stresses in the lateral tibial cartilage in the OpenSim-driven FE model, especially during the early stance phase, were in part a result of the lower abduction and higher internal rotation moments compared to the AnyBody-driven model. In a similar fashion, the higher maximum principal stresses on the medial side in the AnyBody-driven model were partly a result of the high external knee rotation moments during the latter half of the stance phase (Fig. 3c). The changes in the moments are caused by the differences in the muscle activation and force estimations (e.g., Eq. 1 and Eq. 2) and the model-specific differences mentioned before.
The estimated maximum shear strains and collagen fibril strains showed similar results in both models (Fig. 4). The OpenSim-driven FE model estimated slightly higher strains in the lateral tibial cartilage for subject A, whereas the AnyBody-driven model estimated the strains to be slightly higher in the medial tibial cartilage. These discrepancies are caused by the same factors as in the case of the maximum principal stress.
The most visible differences occurred in the estimates for the CoP (Fig. 5). For instance, on the lateral side the OpenSim-driven model estimated the CoP to be more anterior and lateral compared to the AnyBody-driven model, especially with subject B. The CoP has been suggested to play a crucial role in analyzing the knee loads and to strongly correlate with the generated moments in the knee joint [39]. Showcasing this, in a study Haim et al. [54], the modulation of the CoP with different insoles was shown to affect the moments in the knee joint. Due to this association, the differences in the CoP estimates can be explained by the same factors affecting the knee moments, i.e., muscle activations and moment arms.
Although the differences in the maximum principal stresses and strain estimates between the models were small, one could still argue for their importance. For example, in a study by Mononen et al. [8], maximum principal stresses of 7 MPa were used as a threshold for collagen fibril damage, whereas in other studies a level of 30% for maximum shear strain was suggested to trigger cartilage degradation [42, 51–54]. If these thresholds could be assumed to be valid for these specific subjects (e.g., Subject A), the relatively small differences could still lead to an estimation that KOA would develop in the lateral tibial cartilage based on the results from another model, while the other model would predict a healthy future.
It is important to consider that different MS modeling software and models were used for the comparisons in this study. Thus, the results might have been affected not only by the differences in software-specific calculation methods, for example the muscle requirement problem (Eq. 1 and Eq. 2) and activation estimations, but also by the model-specific differences, such as muscle insertion points, muscle bundles, joint center locations and segment definitions [22, 24]. Although this may be thought of as a limitation, our aim was to evaluate the similarity of the outputs from two widely used MS modeling software and their standard models without major modifications. In other words, if one wanted to use both AnyBody and OpenSim by applying the same exact models, it would be possible to obtain identical MS, and subsequently FE, model estimates.
Limitations of this study were also the low number of subjects, and that normal gait was the only activity analyzed. In future studies, the number of subjects should be increased to determine the statistical significance of the estimated similarities and differences. Also, other activities, such as stair climbing or squatting, should be taken into consideration. In addition, implementing subject-specific muscle properties (e.g., strength, insertion points) could provide more accurate estimates for the muscle activations and forces.
This study highlights the importance of understanding and acknowledging the effects of using different computational modeling tools, thus helping researchers to analyze and interpret the numerical predictions if different MS modeling approaches are utilized. Acknowledging these differences helps researchers to understand uncertainties related to personalized multiscale MS-FE models for simulations of, e.g., tissue failure and degeneration.