Sct_extract_metric gm mask threshold


#1

Hello,

Is there a way to increase the threshold of the gm mask (to 0.8 instead of 0.5) when you are using the bin method in sct_extract_metric?

Thanks!

Best,
Romina


#2

Hi Romina,

Not directly, however you can first binarize the GM mask (using threshold of 0.8), and then use sct_extract_metric with the generated mask. Something like this:

sct_maths -i MASK_GM -bin 0.8 -o MASK_GM_BIN08
sct_extract_metric -i METRIC -f MASK_GM_BIN08 -method bin

Cheers,
Julien


#3

Great, thanks Julien. In your experience, is 0.5 a well established threshold or is this data specific i.e. do you think it would be best overall to stick to 0.5 even if there is a bit of overlap?

And one final question (apologies) - related to the above; if there is overlap between the wm and gm - is it necessary to also threshold the wm mask (I wasn’t sure how this would affect the map method)?

Best wishes,
Romina


#4

Hi Romina,

0.5 is the threshold that makes more sense when addressing partial volume effect. For example, if two structures overlap at 50% each within a voxel, setting the threshold to 0.5 will appropriately weight the contribution of each region mask (whereas if you set it to 0.8, you will introduce a bias).

If you are using the map (or the ml) method, you should absolutely not use thresholded masks, because this method is based on Gaussian mixture theory, and hence assumes that the input mask matrix includes partial volume information for each ROI.

Best,
Julien