Thanks for the confirmation, Doug, hope all is well with you. A few notes as follow-up below:

We haven’t used N4 for bias correction ourselves much nor explored it carefully, as we have been fairly happy with the N3/’zheng’ approach (using the recommendations from Zheng et al.) - we will compare them some time in the near future. For now, given the experience with FS7.1.1 default N4 corrections, we are going to stick with the MNI N3 zheng (stronger bias) call options we used before (e.g., an expert option file with the line: mri_nu_correct.mni --no-ants-n4). We did rerun several of the recent 7.1.1 cases with our n3/zheng approach and it has improved the correction in our data, including no longer tlrc failures, improved signal falloff correction for better temporal lobe, improved hippocampal segmentations. Here are a couple of examples:

Above - aseg 7.1.1 default missed hpc on left, with n3/zheng captured hpc

Above - aseg 7.1.1 default poorly separated hpc and amyg on left, better with n3/zheng on right

One naïve question, with respect to the n4 performance/parameters, my read from our mri_nu_correct.mni.log, it looked like there might be some parameters that were passed along to nu_correct for ANTS (in blue bold below) but perhaps were not actually implemented (in red bold below), but I am not experienced with the tool so wouldn’t know if my interpretation is correct. In the mri_nu_correct.mni.log, these parameters looked passed to nu_correct:

· /usr/local/freesurfer/bin/mri_nu_correct.mni

--no-rescale --i orig.mgz --o orig_nu.mgz --ants-n4 --n 1 --proto-iters 1000 --distance 50

nIters 1

mri_nu_correct.mni 7.1.1 …

· AntsN4BiasFieldCorrectionFs -i orig.mgz -o /dev/shm/tmp.mri_nu_correct.mni.2116356/nu0.mgz …

Regardless, given these changes for v7.x with ANTSN4 that no longer calls any -3T option, it might be best if references to this flag are modified on the web, unless other options become available, to avoid confusion e.g., recon-all - - help still says:

USING IMAGES FROM A 3T SCANNER

The -3T flag enables two specific options in recon-all for images acquired with

a 3T scanner: 3T-specific NU intensity correction parameters are used in the

Non-Uniform normalization stage, and the Schwartz 3T atlas is used for

Talairach alignment.

And the -3T option is still listed on MailScanner has detected a possible fraud attempt from "secure-web.cisco.com" claiming to be https://surfer.nmr.mgh.harvard.edu/fswiki/OtherUsefulFlags

Many thanks for your time,

Peace,

Chris

Christine Fennema-Notestine, Ph.D.

From: freesurfer-bounces@nmr.mgh.harvard.edu <freesurfer-bounces@nmr.mgh.harvard.edu> On Behalf Of Douglas N. Greve
Sent: Friday, November 19, 2021 7:58 AM
To: freesurfer@nmr.mgh.harvard.edu
Subject: Re: [Freesurfer] Freesurfer v7.x and bias correction -3T option not working

Hi Christine, yes, your interpretation is correct. The -3T flag only applied when using the mni tools, so it has no effect with ANTS. What was the nature of the failure? We might be able to tweak the ANTS command line to improve it.
doug

On 11/15/2021 12:52 PM, Notestine, Christine wrote:

        External Email - Use Caution

       We recently moved to FreeSurfer 7.1.1 running it on ~ 100 different types of cases using the -3T options, but after a case failure, we looked deeper into the v7.x changes related to bias correction, and it seems that using the -3T option did not produce a different bias corrected outcome than using the default for any cases (see description below). IF this is truly the case, our workaround would be to continue using the MNI N3 approach options we had before (e.g., an expert option file with the line: mri_nu_correct.mni --no-ants-n4), since that improved the correction in our data previously, but would like to make sure we are interpreting the v7 approach correctly. Thoughts on this appreciated, many thanks:

1.       In earlier FreeSurfer versions, we used the options based on the Zheng paper with N3 as recommended for 3T data. So with our switch to v7.1.1 which is now using ANTs N4, we looked into the 3T options with the flags described on the wiki, and used those flags in our runs accordingly.

2.       However, once we had this case failure (which had worked in a prior FS version run with the N3 3T flags), we dug into things and did some test runs. We found that in Freesurfer v7.1.1 when using ANTs N4 (i.e., AntsN4BiasFieldCorrectionFs) [instead of MNI N3 (i.e., nu_correct)], the effect of specifying the recon-all -3T (or -nuintensitycor-3T) option did not change the outcome relative to the default bias correction:

1.       In v7.1.1, it seems that when ANTs N4 is used for bias field correction, the same bias field correction is performed whether the -3T (or -nuintensitycor-3T) option is specified or not. We assessed this by creating nu.mgz difference images (between runs with and without the -3T option) and confirmed by reviewing logs and the mri_nu_correct.mni script.

2.       When the ANTs N4 bias field correction is used with the -3T options, the parameters for stronger bias field correction (i.e., "--n 1 --proto-iters 1000 --distance 50") are passed to mri_nu_correct.mni, but then seem to be dropped by mri_nu_correct.mni and not included in the subsequent call to AntsN4BiasFieldCorrectionFs.

3.       Below are the relevant lines extracted from a mri_nu_correct.mni.log when stronger bias field correction was specified via the -3T (or -nuintensitycor-3T) option to recon-all but the parameters do not seem to be used by ANTs:

mri_nu_correct.mni.log:

...

/usr/local/freesurfer/bin/mri_nu_correct.mni

--no-rescale --i orig.mgz --o orig_nu.mgz --ants-n4 --n 1 --proto-iters 1000 --distance 50

nIters 1

mri_nu_correct.mni 7.1.1

...

AntsN4BiasFieldCorrectionFs -i orig.mgz -o /dev/shm/tmp.mri_nu_correct.mni.2116356/nu0.mgz

Using shrink factor: 4

...

--

Christine Fennema-Notestine, Ph.D.
_______________________________________________
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