SSM model validation

[Pieter Vanslambrouck]

I have a general question regaring the validation of an SSM model. Which methods can be used to verify that an SSM model is representative for the entire population?

I found a method in this paper:

I attached a screenshot with the relevant paragraphs from the paper. In summary, leave-one-out validation steps are performed to verify that the SSM is able to represent unseen samples.

Are there any other methods to ensure the generalization of an SSM?

Best regards,

Pieter Vanslambrouck

KU Leuven

[Behzad Vafaeian]

Hi Pieter,

I think you forgot to write the name of the paper. Or, I cannot see it in the email.

Can you add the name of the paper please.

Thanks very much

--
You received this message because you are subscribed to the Google Groups "scalismo" group.
To unsubscribe from this group and stop receiving emails from it, send an email to scalismo+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/scalismo/9a31858f-6c4e-497e-9f4f-93238d00a912n%40googlegroups.com.

[Gregory Lahman]

If I am not mistaken you can also use a training and testing set, where you set aside a certain percentage of your meshes for evlauation purposes, then calculate the model and use that testing set to calculate a generalizability score.

[Pieter Vanslambrouck]

Apologies, something went wrong to include the paper in the last email. The paper is:

Statistical Shape Modeling of Skeletal Anatomy for Sex Discrimination: Their Training Size, Sexual Dimorphism, and Asymmetry
Audenaert, E. A. et al.

[Marcel Luethi]

Dear all,

Thanks everybody for the interesting discussion and sharing the paper.

Personally, I am not a big fan of just reporting single scores, like generalisation, specificity, etc. I think they should be complemented by other means of verification. One simple possibility would for example be to measure some standard anatomical distances on the model, calculate how these distances vary across random samples of the model.and then to compare that with reported studies in the medical literature. I also think that we should always show visualizations of the samples, as they might convey more information than a specificity score. To some degree, these things were done in the paper you cited paper about sexual dimorphism. However, I think one could/should go even further.

In this context I like the paper "Bayesian workflow", by Andrew Gelman et al. quite a lot. (https://arxiv.org/abs/2011.01808). It speaks more generally about how to build and validate statistical models, but I think we can apply a  lot of the thinking in shape modelling.

Best regards,

Marcel

--

You received this message because you are subscribed to the Google Groups "scalismo" group.
To unsubscribe from this group and stop receiving emails from it, send an email to scalismo+u...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/scalismo/73d1a5ad-026b-459e-b5b6-39ded7e5d07fn%40googlegroups.com.