Hi,

I was doing a quick comparison of the output from versions 1.4.1 and 1.7.5, and it looks like there are some gross discrepancies between the versions (see the attached excel sheet). Thus, I'm just wondering what could account for these differences (e.g., what changes were made between versions). The input was exactly the same for each and the raw CSF numbers come from running CMTK statistics on the MID02 output files.

Thanks,
Maddy

Hi Maddy, 
Please use the latest version 1.7.6, some significant changes were made in order to facilitate 'batch mode' processing and the discrepancies you are finding should now be resolved. 
Best,

Hi,

I just ran a quick comparison of Auto_EACSF versions 1.4.1_RHL6 and 1.7.7 on a dataset, and there are some interesting differences in the output. I was just wondering what changes in processing were made between versions (and if this might account for the outliers). I have attached an Excel sheet with the output to this post (the input was exactly the same for each and the raw CSF numbers come from running CMTK statistics on the MID02 output files).

Thanks,
Maddy

Hi, 
Thanks for the detailed differences. The changes were done to improve 'batch mode' processing as there were significant problems when trying to execute it this way. However, the differences you are getting are quite large and it shouldn't be the case. I'll take a look to find out where the changes have altered the results. 
Best,

Hi,

I hope you are doing well! I just wanted to check and see if you have been able to find anything that might account for the version differences in output.

Thanks,
Maddy

Hi Maddy, 

I did a run and I noticed there are difference between the two versions. Nevertheless, the output for the run I made looks acceptable. 

The problem with version 1.4.1 was that during the execution, some of the parameters were not passed correctly to other scripts and it was not possible to execute it sequentially for multiple cases. Most of the changes were done to allow consistent execution when running it this way. 

What atlases are you using for the skull stripping? Are you using the same ones for both runs? 
Are you using T1 and T2 images?

Hi Juan,

That is good to know. For both runs, I used the exact same inputs--a T1w and T2w image for each participant and the same skullstripping atlas (the brainmasking multi-atlas provided on this tool's site).

Thanks!
Maddy

Hi Maddy, 

Upon further revision of the differences, one source could also be the version of BRAINSFit being used. 
Could you check what version of this tool are you using? 

Thanks,

Hi Juan,

For each run, I used the BRAINSFIT tool inherent to each version of Auto_EACSF. For example, when running v1.4.1_RHL6, I used the BRAINSFIT found within that version's bin. Are the BRAINSFIT tools found within v1.4.1_RHL6 and 1.7.7 two different versions of BRAINSFIT?

Thanks,
Maddy

I verified this and they are the same version. it should be OK if you are using the versions distributed in the packages. 
However, do the outputs of the rigid registration look the same in your runs?

Great. The outputs for the rigid registration are quite different.

In the v1.4.1_RHL6 folder:

111312_T1_stx.nii.gz
111312_T1_stx_T2JointT1_tmp1_mask.nii.gz
111312_T1_stx.nrrd
111312_T1_stx_T2JointT1_tmp2.nii.gz
111312_T1_stx_255.nii.gz
111312_T1_stx_T2JointT1_tmp2_mask.nii.gz
111312_T1_stx_255_Skull.nii.gz
111312_T2_stx.nii.gz
111312_T1_stx_T2JointT1_mask.nii.gz
111312_T2_stx.nrrd
111312_T1_stx_T2JointT1_tmp1.nii.gz

In the v1.7.7 folder:

111312_T1_stx.nrrd
111312_T2_stx.nii.gz
111312_T2_stx.nrrd

They are organized a bit differently. You can find the same outputs in version 1.7.1 in the "SkullStripping" directory.