Hi Krystal,

Apologies for the delay in replying (grant deadline).

The lesion class in samseg is a bit special, as the internal lesion probabilities are thresholded (using the value passed on via the --threshold flag, or a default value 0.3 if you don't use that flag) to obtain the voxels labeled as 99 in the segmentation output. In contrast, the value in the .stats file is simply the sum (over all voxels) of those internal lesion probabilities.

If you want to inspect the impact of the exact threshold value on the number of voxels with label 99, you could run (as you did) samseg with the --save-posteriors flag to save the internal lesion probabilities (in yourSeg/posteriors/Lesions.mgz). Then you can play with mri_binarize and mri_segstats to obtain lesion volumes for different thresholds without having to re-run sameg anew.

Hope this helps,

Koen

On Sun, Jan 7, 2024 at 11:05 PM Krystal Xiwing Yau <krystaly@hku.hk> wrote:

External Email - Use Caution
Dear FS experts,

I am testing out which threshold works the best for the image. As I tested 0.3 and 0.9, the lesion masks do show the difference but the measurement is the same. Please see below for your information:

Fig 1. Lesion.mgz with a threshold of 0.3 applied to FLAIR image

Fig 2. Lesions.mgz with a threshold of 0.7 applied to FLAIR image

Do the volume of lesions supposedly different with threshold adjustments?

The following code is what I used for running samseg:
run_samseg --input $FREESURFER_HOME/${i}.nii $FREESURFER_HOME/${i}_flair_reg.nii --pallidum-separate --lesion --lesion-mask-pattern 0 1 --threshold 0.7 --output $SUBJECTS_DIR/${i}_samsegOutput_thold0.7/ --save-posteriors --threads 8
Thank you for your attention.

Regards,
Krystal

Tel.: +852 2255 5024

Fax: +852 2872 5828

Address: Room 304, 3/F, New Clinical Building, 102 Pokfulam Road, Hong Kong.
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
MailScanner has detected a possible fraud attempt from "secure-web.cisco.com" claiming to be https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer