Template Evaluation

[EBirnie]

Hi all!

I successfully generated a population specific brain template based on T1 and T2 images with antsMultivariateTemplateConstruction2.
Now I want to evaluate the template based on 4 points:

1. Convergence and Data saturation
2. Image Quality
3. Population Average
4. Clinical Use

I have ideas (and already some results) for the 1st, 3th and 4th point, but run into problems with the 2nd.
Because the template is essentially an average of all the input images, commonly used image quality measures as the SNR, CNR, gradient magnitude and entropy say little.
In literature I see that many templates are assessed based on the registration-based segmentations they produce in patient space. I see that this important for clinical use. However, I want to take more into account than that alone.

Since I am not the only one generating a brain template here, I am curious which measures you use the assess the templates you generate! Or if you have some ideas that I can try!

I am looking forward to you reactions.
Thanks in advance!

[cookpa]

It's an interesting question. One idea for point 2 would be spatial smoothness, I think AFNI has a command for this. Low-quality registration will tend to blur the output and reduce tissue contrast.

[ntustison]

There's a TID-based network that @stnava put together for image quality assessment. Example here.