Dear Doug,

I have uploaded 3 subjects with 0.7mm3 data so that you can take a look: None of the subjects has macroscopical pathological changes, the juxtacortical hypointensities should be physiological finding.

1. example_subject_0.7mm.tar.gz

This is subject processed by recon-all -all -T2pial -hires processed by dev version from 07/02/2017.

In this subject you can see the cutting of the pial surface in the midline (due to the wrong aseg labeling to contralateral gray matter) at the RAS coords

2,-70,17

and juxtacortical hypointensity affecting pial/white surface in

7,29,43

2. juxtacortical_hypointensities_0.7mm_HCP_pipeline.tar.gz

This is subject processed by HCP pipeline with quite large amount of juxtacortical hypointensities largely affecting pial/white surfaces, located for example at the RAS coords:

11,-64,54 15,-48,48

White surface produced by standard recon-all -hires -white is the file ?h.white.prehires.

3. test_first_wm_peak_HCP_pipeline_base.tar.gz

This is subject processed by HCP pipeline with combination with 2nd step of longitudinal stream (-base). In this subject I tested -first_wm_peak behavior (option for mris_make_surfaces used in FreeSurferHiResWhite.sh in HCP pipeline).  There has been change of implementation of -first_wm_peak in the December 2016. It seems that the recent implementation of -first_wm_peak positions the white surface too internally as I commented on here:

http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg51414.html

White surface produced by standard recon-all -hires -white is the file ?h.white.prehires.

Additionally, there are files  lh.white.deformed_ver_201611  - file produced by using -first_wm_peak (original version before code change)  lh.white.deformed_ver_201701 - file produced by using -first_wm_peak (version after code change with white surfaces positioned very internally, I consider it not anatomically relevant)  lh.white.deformed_ver_201701_wo_first_wm_peak - file produced without using -first_wm_peak.

Thank you for your time and looking forward to your comments!

Regards,

Antonin

Hi Antonin, can you upload that subject? In one case using 700um, the white surface looked fine, and I did not see much of a difference with the 1mm scan. On 3/14/17 5:39 AM, Antonin Skoch wrote: Dear experts,

can also be the difference between v5.3 and v6.0 behavior in HCP pipeline and possible sub-optimal behavior of v6.0 with HCP pipeline given by different in image input to mris_make_surfaces in HCP pipeline (FreeSurferHiResWhite.sh) ? For white surface estimation the recon-all input is brain.finalsurfs, which is very aggressively filtered so that the white matter voxels have typically values 110 and the SNR in deep white matter areas approach infinity. In contrast, for HCP pipeline the input T1w_hires_masked_norm image is filtered no to such an extent. Therefore there is much lower contrast-to-noise ratio in the region of GM-WM interface for mris_make_surfaces in FreSurferHiResWhite.sh. This maybe renders mris_make_surface more vulnerable to local deviations in intensity in GM-WM interface. My observation (without any thorougful tests) is that with v6.0 in HCP pipeline I had very often to edit wm.mgz in the region of GM-WM interface in the areas of (physiological) perivascular subcortical T1 hypointensities due to the ?h.white "leak" to these areas. And also, as I wrote in some of my previous (unanswered) posts:

http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg51414.html

the change of -first_wm_peak behaviour in v6.0 in fact worsened white surface estimation in my data I tested with HCP pipelines. I could upload some example subjects if you are interested and have time to dig in. In any case, it is probably hard to compare performance between standard recon-all and HCP pipeline and different versions of FreeSurfer due to different input to mris_make_surfaces and different flags used. The HCP pipeline, as it currently stands, would probably need different optimization of code or different command line options due to the different input used. Regards,

Antonin Skoch

Hi Bruce,

I don¹t think the hires T1w and T2w stuff are as accurate as 5.3 in the HCP 0.7mm data that I have looked at (and as I recall at least one other user had some similar issues, as did one of my collaborators), but that was all I meant to refer to. Sorry if my reply came across more general than that, as of course we want to move to using FreeSurfer V6+ in the future. It is the highres T1w and T2w stuff that my pipelines are designed to exploit, however. Also there are some changes I have to make to the HCP Pipelines for FreeSurfer V6+ because of differing file names, etc. Best,

Matt.

On 3/13/17, 9:40 PM, "Bruce Fischl" <freesurfer-boun...@nmr.mgh.harvard.edu on behalf of fis...@nmr.mgh.harvard.edu> wrote:

Hi Matt

To state without qualification that "6.0 has some regressions as far as surface placement" is incorrect. We quantified its test-retest reliability, accuracy and power to detect disease effects and all were improved relative to 5.3. We tested V6 on hundreds of datasets across an array of pathologies and different MR sequences, field strengths and scanners. We visually inspected dozens of brains multiple times in the process of improving accuracy and robustness. I can believe that on Wash U HCP data there could be some specific issues, but to imply that V6 is generally less accurate is simply incorrect.

cheers Bruce

On Mon, 13 Mar 2017, Matt Glasser wrote:

...

Not yet. We are hoping FreeSurfer 6.1 will work nicely with the HCP Pipelines. For now version 6.0 has some regressions as far as surface placement goes and there are also some adaptations we need to make to the pipelines.

Peace,

Matt.

From: freesurfer-boun...@nmr.mgh.harvard.edu on behalf of CAGNA Bastien bastien.ca...@univ-amu.fr Reply-To: Freesurfer support list freesurfer@nmr.mgh.harvard.edu Date: Monday, March 13, 2017 at 7:21 AM To: "freesurfer@nmr.mgh.harvard.edu" freesurfer@nmr.mgh.harvard.edu Subject: [Freesurfer] Can we use fressurfer 6 with the Human connectome project's pipeline ?

Dear freesurfer experts,

I'm wondering if it's possible to run the human connetome project's minimal processing pipeline using freesurfer 6 ?

Does it require some update of the pipeline or is there anybody that have already enjoyed the new freesurfer version with this pipeline ?

Thank you for your attention and your help,

Bastien Cagna