Low scaling and IK error but deformed model

[Radhakrishnan Vignesh]

Hello,

I have scaled a model using data from CT scans converted to lab space and also performed IK analysis using similar data. The scaling error is less than 1 cm (RMS) and 2cm (MAX) and residual error during IK is less than 2-4cm (RMS). Also, the calculated knee angles correlates very well with gold standard data.

But the model is deformed. I am using a single leg model and since the leg6dof8 model deformed, created a single leg model from gait2354. I have attached an image of the deformed model. How does such a model provide such accurate results or is it that the model is not accurately representing the markers?

Thanks
Vignesh

deformedModel1.JPG (16.88 KiB) Viewed 505 times

[Radhakrishnan Vignesh]

Hi, does anyone have an answer to this issue?

Thanks,
Vignesh

[Thomas Uchida]

In what way is the model "deformed"? If the issue is that the ankle is plantarflexed when you do not expect it to be plantarflexed, perhaps there are no markers on the calcaneus (it's unclear from the image in the post)? If there are no markers on the calcaneus, then it would be possible to have low marker error during scaling and IK while still having an unreasonable model. For another perspective on the matter, you could preview the experimental data, sync it with the IK solution, and hide the bones: the IK Tool is merely aiming to reduce the error between marker pairs.

[Radhakrishnan Vignesh]

Hello Thomas Uchida,

Thanks for your reply. By deformation I mean that the pelvis is bent and the knees are also bent despite the three markers supposed to be at the knee.

By hiding the bones and matching the ik solution, does it mean even if there is a physical deformation but the error is low, the solution is acceptable ?

[Thomas Uchida]

Please see the "Troubleshooting Tips" section of the "Getting Started with Scaling" page in the User's Guide (https://simtk-confluence.stanford.edu:8 ... otingTips:) and the "Best Practices and Troubleshooting" section of the "Getting Started with Inverse Kinematics" page (https://simtk-confluence.stanford.edu:8 ... leshooting). The "Tips and Tricks" video may be helpful as well: https://simtk-confluence.stanford.edu:8 ... oTutorials.

[Radhakrishnan Vignesh]

Thank you . I had a look at some of them and have ensured that the guidelines for the error ranges like RMS error and max error are met.

To confirm , if the error is low but the animation shows the bones moving weirdly , for example for a single leg model, does it mean the results is still good based on the low error ?

[Thomas Uchida]

if the error is low but the animation shows the bones moving weirdly , for example for a single leg model, does it mean the results is still good based on the low error ?

This is a very good question. Low marker error is (generally) a necessary but insufficient condition for the model to be useful. For example, if you have only 1 marker on a person's torso and track the marker trajectory perfectly, you will have zero marker error but the body pose will be random. This is an extreme example but a similar situation could arise depending on where markers were placed and what marker pairs are being used for scaling and tracking. Perhaps another place to look for examples would be in your local Resources directory or in Tutorial 3 (https://simtk-confluence.stanford.edu:8 ... e+Dynamics). Unfortunately, beyond this, I would need more information about the markers and workflow to diagnose the issue.

[Radhakrishnan Vignesh]

Thank you. I am working on inverse dynamics right now but my goal was to compare the residual error and the accuracy of inverse kinematic results between skin-mounted markers and markers I projected on to the bone using CT scans.

The angles calculated using both was very similar to ones obtained using fluroscopy but the animation of the model does not look very smooth. So I was not sure how the calculated angles can be so close to the gold standard but the animation look abnormal in a way