You should not use either T1.mgz as that has been highly processed for other purposes. The brainmask has also been processed, skull stripped and bias correction. I have not looked at what has to happen to get the most correlation between the methods; they are different methods and are supposed to give different results (we hope that samseg is an improvement on the aseg). I recommend using the raw input when specifying recon.

On 3/14/2023 3:26 PM, Giulio Siracusano wrote:

        External Email - Use Caution        

Dear Experts,
I'm trying to understand what is the best input choice for running SAMSEG --recon in order to have the closest results as possible when compared against ASEG after executing (separately) an instance of recon-all all (and the raw input files in NIFTI format).
I'm very well aware that ASEG and SAMSEG are different methods.
I'm using the aseg.stats file as a results of the recon-all all command (here renamed as aseg.recon-all.stats for an easier identification). These stats should be my reference.
 
Currently I've run the samseg command using the following input variants:
1) raw input file (brain MP-RAGE sequence) converted into a mgz archive (Raw.mgz).
samseg --t1w ./Raw.mgz --s 002_S_0685_DIRECT --recon --refmode t1w
The corresponding results are attached into (samseg.direct.stats)

2) This time I've used the file T1.mgz which is one of the intermediate results generated after running recon-all -autorecon 1. Specifically, this is the output of (mri_normalize -g 1 -mprage nu.mgz T1.mgz).
samseg --t1w ./T1W/T1.mgz --s 002_S_0685_DIRECT_T1W --recon --refmode t1w
The corresponding samseg.stats file is attached and called (samseg.direct.t1.stats).

3) this time I've used the file brainmask.mgz which is the last intermediate result generated after running recon-all -autorecon 1. Specifically, this is the output of mri_watershed.
samseg --t1w ./brainmask.mgz --s 002_S_0685_DIRECT_BRAIN --recon --refmode t1w
The corresponding samseg.stats file is attached and called (samseg.brain.stats).

I've noticed that the SAMSEG stats file (samseg.brain.stats) as obtained when using the brainmask.mgz as input file is the closest with the ASEG statistics (aseg.recon-all.stats).
I also observe that the unknown volume is the lowest (# Measure Unknown, 5329.102436, mm^3) when using brainmask.mgz as input for SAMSEG, whereas is higher (Measure Unknown, 325302.894793, mm^3) when using T1.mgz as input file and is the highest (Measure Unknown, 946924.354060, mm^3) for the raw input.

I know that for running recon-all we use the raw input files (for example, in NIFTI format),
but I wanted to know what is the best possible configuration for SAMSEG in terms of input file in order to have the closest match between SAMSEG and ASEG statistics.
Could it be possible to use the intermediate result T1.mgz or brainmask.mgz as input file for running samseg command?
Is it a technically correct procedure?
Do we need to use only the raw data as input for running samseg --recon command?

Hope this is clear
Regards
GS 

_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer