Summary
The Theia3D Apollo update is coming soon! Theia3D kinematics have been evaluated against marker-based data many different times, including against varying model definitions. One of the most apparent differences that has come up during these evaluations are those in the pelvis and hip kinematics due to Theia3D’s use of a neutral pelvis. In this blog post, we continue our lead-up to Theia3D Apollo later this month by discussing the differences and changes that have been made to the Theia3D model which improve on the pelvis and hip kinematics, and go further to improve hip joint center positioning.
If you’ve followed the development and validation of Theia3D over time, you may be familiar with where some of the differences lie when it is compared against marker-based measurements, based on previously published results. Theia3D has been shown to perform exceptionally well in measuring the largest range of motion segment poses and joint angles, especially sagittal and frontal plane lower limb joint angles. Two areas where it generally obtains slightly different results with respect to marker-based measurements (without getting into whether this is good or bad), are the smaller, more challenging to measure kinematics that incorporate segment axial rotations (e.g. knee internal/external rotation), and the baseline anterior tilt of the pelvis.
In some of the earliest validation studies, near-constant offsets in the hip flexion/extension joint angles were observed between Theia3D and marker-based data, the cause of which were not fully articulated initially. However, with very small differences in the thigh segment sagittal plane rotation (RMSD = 2 deg) but a clear offset of roughly 11 degrees in hip flexion/extension angles, it is quite apparent that these offsets were due to differences in the definition of the pelvis segment, in particular its anterior tilt (for more information on Theia3D Apollo’s updated validation, see our previous blog post!).
Based on these results, Theia3D’s pelvis definition was found to be more neutral, whereas the marker-based pelvis (defined using the Visual3D Composite Pelvis) had a moderate anterior tilt. This aspect of the V3D composite pelvis is actually noted in its documentation, which states that “if the markers have been placed accurately, the pelvis segment coordinate system will typically be tilted forward”. So this being a known phenomenon, it is unsurprising that a difference was found.
Figure 1: Thigh segment angle rotations in the sagittal plane (left) between marker (blue) and markerless (orange) show very similar measurements, while hip flexion/extension joint angles show a very clear offset between the systems. Figures reproduced with permission from the authors (Kanko et al., 2021. J Biomech 127: 110665).
In subsequent validation testing, this offset was shown to be effectively eliminated when using a more neutral, virtual pelvis segment for the marker-based kinematics, consistent with typical methods used in Visual3D. Results from a similar validation study on running kinematics with a slightly different marker-based model that used this virtual pelvis showed much more similar results between marker-based and Theia3D pelvis and hip kinematics.
Figure 2: Pelvis segment anterior tilt angles (left) between marker (red) and markerless (blue) show relatively similar results when using a neutral, virtual pelvis segment. The neutral marker-based pelvis has the expected impact of eliminating the previously noted offset in hip flexion/extension (right). Green lines show asynchronous markerless measurements under a different clothing condition; check out this blog post for more discussion of this validation study. Figures reproduced with permission from the authors (Kanko et al., 2024. J App Biomech 2:129-137).
And so, here we are again, looking at pelvis anterior tilt and hip flexion/extension angles as they relate to the 2024 release of Theia3D Apollo. For this major release, one of the more significant changes to Theia3D’s markerless model is in pelvic tilt, which now better reflects natural anterior pelvic tilt. To examine the changes to the kinematic comparison when using this updated pelvis definition, we compared the newest markerless model against a marker-based model using the neutral, virtual pelvis segment and the anteriorly tilted Visual3D Composite pelvis segment.
As expected, the new markerless pelvis definition results in a more physiological level of anterior pelvic tilt and shows improved agreement with the Visual3D Composite pelvis segment and the corresponding hip flexion/extension angles. Conversely, the new markerless pelvis shows an offset relative to the neutral virtual pelvis segment and its corresponding hip flexion/extension angles. The Visual3D Composite marker-based pelvis and the Theia3D Apollo pelvis both show an average anterior pelvic tilt of approximately 10 degrees, whereas the neutral virtual pelvis hovers right around 0 degrees, as expected based on its definition.
Figure 3: This figure shows updated lower limb segment angles during walking for Theia3D Apollo (‘markerless_v2024-1-0-4409-H3-K3-A3’, dark blue) compared to the marker-based dataset using a virtual, neutral pelvis segment definition (‘marker-neutral’, black) and the anteriorly-tilted Visual3D Composite pelvis segment definition (‘marker-V3Dcomposite’, teal).
Furthermore, the hip flexion/extension joint angles from Theia3D Apollo and the Visual3D Composite pelvis model are nearly indistinguishable, whereas the neutral virtual pelvis shows an offset towards extension that is expected based on the pelvis segment angles shown above.
Figure 4: This figure shows updated lower limb joint angles during walking for Theia3D Apollo (‘markerless-v2024-1-0-4409-H3-K3-A3’, dark blue) compared to the marker-based dataset using a virtual, neutral pelvis segment definition (‘marker-neutral’, black) and the anteriorly-tilted Visual3D Composite pelvis segment definition (‘marker-V3Dcomposite’, teal).
Keen-eyed readers may notice one other difference shown in these results, related to the feet: the marker-based models also had slightly different foot segment definitions, and differences are not related to the pelvis definition. This perhaps deserves a brief post of its own, so stay tuned!
One other aspect of the Theia3D pelvis that you may not have been previously aware of is the hip joint center width, which is the vector distance between the hip joint centers. The definition of the hip joint centers is, in general, a challenging aspect of biomechanical modeling. The main approaches for obtaining the estimated hip joint centers has been through regression-based models, or using functional movements to find the center of rotation of the thigh. When using regression approaches with marker-based data, the hip joint center positions are directly influenced by the position of the pelvis markers, which are themselves easily influenced by the volume of soft tissue around the pelvis. This leads to hip joint center widths that vary widely from person to person, anywhere from roughly 14 to 22 cm for this particular marker-based dataset. However, based on measurements from imaging studies we know that in reality there is very little variability in the distance between hip joint centers across individuals, which is also true for men and women! Measurements of hip joint center positions from one study of 120 adults of varying ages resulted in an average hip width of 17.0 +/- 0.9 cm (Hara et al., 2016. Sci Rep, 6: 37707). This is another key area where Theia3D Apollo offers improvements over previous versions and marker-based models, as the modeled hip widths across all trials show very similar measurement attributes to those from imaging studies, with an average of approximately 16.5 cm and a similar level of variability. However, it should be clarified that because the hip width is a vector distance between joint centers, this does not necessarily imply that the overall hip joint center positioning is more accurate in any of the three global planes. So, while the hip joint center widths obtained from Theia3D 2023 were relatively closely matched to those from the marker-based dataset, the new Theia3D Apollo model’s hip joint center widths more closely match those from larger samples of a more direct measurement of hip joint center widths. This is a great example of one area where marker-based data is not an effective target for assessing the accuracy of markerless modeling methods.
Figure 5: Hip joint center width box plots from a previously-published imaging dataset (Hara et al., 2016. Sci Rep, 6: 37707), the marker-based reference dataset, Theia3D 2023, and the 2024 update, Theia3D Apollo.
Conclusion
So, to summarize this discussion of modeling differences - the Theia3D Apollo release features a new and more physiological representation of the pelvis, reflected in more realistic anterior tilt and more accurate hip joint center positions. These are changes that, for one, bring markerless biomechanical modeling closer in line with the measurements obtained from marker-based models, and for another, bring markerless measurements closer to ground truth imaging measurements than marker-based data. Overall, we see both of these changes to be further improvements of the Theia3D model and which continue to increase its biomechanical accuracy.
More content on Theia3D’s 2024 update, Apollo, is in the works, so keep your eyes on the blog! Check out our previous post on Theia3D Apollo's updated validation results here, or to learn more about Theia, click here to book a demo.
Comments