@measure_bb_thick in macaques

AFNI Message Board
1

Hello AFNI team,

Is it necessary to change default parameters in @measure_bb_thick in macaques as in :
https://afni.nimh.nih.gov/pub/dist/HBM2018/OHBM_2018_Thickness.pdf

I am thinking of:
-resample
-maxthick
and
-depthsearch

The program is particularly long in my case
I am using masks of 0.5mm isotropic

Thank you all!
Clément

2

Computation time does vary a lot with the spatial resolution. Defaults might work better for human data. Are you finding that you can’t change options to match your data? The @measure_erosion_thick is faster and also works well with a somewhat different definition of thickness.

3

Hello Daniel,
No, it doesn’t create any problem (so far)
I was just wondering what parameters did you put in your example in order to avoid to re-invent the wheel especially since I am really not sure what they do exactly and what will be the impact of changing them.
Any proposition to try at first for

-resample
-maxthick
and
-depthsearch

I will definitely try @measure_erosion_thick if you think that is the best!
Thank you!
Clémernt

4

I’m not saying the erosion method is best, but it is pretty good. It is definitely faster than the BB method. The ball and box method measures thickness in a very different way, so nodules or bumps would have different measures. Both are somewhat sensitive to the sampling resolution. The BB method will work about twice as fast if “boxes” aren’t included with “-balls_only”. For macaque resolutions at 0.5 or 0.25 mm resolution data, upsampling isn’t really required, but the thickness precision is more limited with lower thickness values. The best way is to try and see. Thickness precision is limited by voxel size for all methods really.

We are working on a distance transform method that is just about ready for release too. The distance transform part was started in an AFNI hackathon (Code Convergence) with Chris Rorden. He wrote up a very nice version as Depth3D, that is both fast and accurate. The method computes the shortest distance to an edge of the mask at every voxel in a very efficient way. Transforming depth to thickness can be done as in the other methods by projecting 2*maximum depth either volumetrically or by volume to surface projection along normal vectors. Then smoothing the data in the volume or surface gives the final map.

https://github.com/neurolabusc/DistanceFields
https://afni.nimh.nih.gov/pub/dist/HBM2020/GlenEtal_FindingYourCenter_OHBM2020.pdf

5

For info:
I tested both methods without changing any parameters.
The result seems good and no significant variations were found between the two methods.
Thank you!
Clément