Freesurfer community,
For those who wish to install a 'development' release of Freesurfer on Linux RedHat 9, CentOS 4 or CentOS 4 x86_64 platforms, distribution files are now posted at:
http://surfer.nmr.mgh.harvard.edu/pub/dist/?M=D
On this distribution site, a README is available:
freesurfer-Linux-platform-dev20050905-full.README.txt
The new distributions include:
freesurfer-Linux-rh9-dev20050905-full.tar.gz
freesurfer-Linux-centos4-dev20050905-full.tar.gz
freesurfer-Linux-centos4_x86_64-dev20050905-full.tar.gz
The 'dev' release contains all the functionality of the upcoming 'stable' release, but has not been fully tested, particularly in regards to installation issues.
This 'dev' release is intended for those who wish to attempt an installation on their Linux platform, and are willing to give feedback on possible installation problems.
This 'dev' release contains most of the libraries that Freesurfer requires: the MINC toolkit, Tcl/Tk/Tix and BLT libraries, GSL and Qt, as well as sample MRI data.
The release may be installed in any directory.
A Mac OS X 'dev' build is not available just yet.
Please direct comments or questions to this mailing list.
Hi Nick,
I downloaded freesurfer-Linux-rh9-dev20050905-full.tar.gz and installed it on the following system: Fedora Core 4; AMD Sempron 2600+ on an ECS KT600-A board; Nvidia GeForce 2; 1 GB DDR 400 RAM.
'recon-all -subjid XXX -autorecon-all' would not run at first because it could not find some of the mni tools. The ones that it was having difficulty with were the perl scripts (not the binary MNI files). I noticed the first line in each of the scripts points to what looks like the MGH specific perl installation (i.e., /usr/pubsw/bin/perl), which does not exist on my system. I tested two different solutions 1) change the first line of each file and 2) mkdir -p /usr/pubsw/bin and then add a symbolic link to my perl binary. Both methods appear to work, and 'recon-all -subjid XXX -autorecon-all' has been running successfully for about 12 hours. I assume it is working!
I also noticed that recon-all expected the mgz files for each separate anatomical to both be located in $FREESURFER_HOME/subjects/XXX/mri/orig. This is opposed to what I recall for the old COR method of placing separate anats in 001 002 etc (for obvious overwriting reasons). I believe this is documented in the recon-all help file, but could be confusing for some users at first.
Csurf, tkmedit, and tksurfer gui's loaded without a problem on my FC4 machine. Csurf loads surfaces fine, but I cannot get it to load mgz files. It seems to expect COR files. Out of habit, I did all of my volume edits using 'Edit Segmentation' in csurf, but may have to switch tkmedit/tksurfer. No problem there.
In sum, the current dev version for Redhat 9 appears to be running fine on a basic Fedora Core 4 machine.
Thanks, Brian
Brian,
I'm glad to hear that the installation is now working for you.
The recon-all -autorecon-all command can take 24 or more hours to complete, depending on the quality of the scan, and the difficulty of auto-fixing any defects, and of course the machine upon which it is running. Your machine appears to be more than adequate.
Concerning the perl scripts which needed editing, which files (with paths) required editing? I'd like to remove any hard-coded paths in our installation.
Concerning the mgz files, recon-all does expect to find each structural scan in the mri/orig directory, with filenames 001.mgz, 002.mgz, etc.. Given a subject having three sets of DICOMS (three runs of structural scans), I do the following to setup a subject:
cd $SUBJECTS_DIR mksubjdirs <subjid> cd <subjid> cd mri/orig mkdir 001 mkdir 002 mkdir 003 cp -r <path_to_dicoms>/95555-2-*.dcm 001/ cp -r <path_to_dicoms>/95555-3-*.dcm 002/ cp -r <path_to_dicoms>/95555-7-*.dcm 003/ mri_convert -it dicom -ot mgz 001/ 001.mgz mri_convert -it dicom -ot mgz 002/ 002.mgz mri_convert -it dicom -ot mgz 003/ 003.mgz
Now you will have three .mgz files in mri/orig, and recon-all will automatically take these three structural scans and average/motion- correct them.
The reason that we now encourage .mgz files over the COR format is because the COR format is limited to storage of volume data in unsigned char format only, and does not allow int, float, double or long formats. Also, the COR format historically assumes a slice size of 256 by 256 voxels. For these reasons, the COR format is deprecated in favor of the newer mgh (and mgz) format.
Concerning csurf, we no longer actively maintain csurf, in favor of recon-all. csurf is essentially an older copy of the recon-all script, and is not kept in-line with recon-all. We encourage using the recon- all script because it is easier to see what is going on, and to maintain. But we'll look into the possibility of at least csurf supporting the opening of .mgz files.
Nick
On Mon, 2005-09-05 at 21:58 -0700, Brian Schweinsburg, Ph.D. wrote:
Hi Nick,
I downloaded freesurfer-Linux-rh9-dev20050905-full.tar.gz and installed it on the following system: Fedora Core 4; AMD Sempron 2600+ on an ECS KT600-A board; Nvidia GeForce 2; 1 GB DDR 400 RAM.
'recon-all -subjid XXX -autorecon-all' would not run at first because it could not find some of the mni tools. The ones that it was having difficulty with were the perl scripts (not the binary MNI files). I noticed the first line in each of the scripts points to what looks like the MGH specific perl installation (i.e., /usr/pubsw/bin/perl), which does not exist on my system. I tested two different solutions 1) change the first line of each file and 2) mkdir -p /usr/pubsw/bin and then add a symbolic link to my perl binary. Both methods appear to work, and 'recon-all -subjid XXX -autorecon-all' has been running successfully for about 12 hours. I assume it is working!
I also noticed that recon-all expected the mgz files for each separate anatomical to both be located in $FREESURFER_HOME/subjects/XXX/mri/orig. This is opposed to what I recall for the old COR method of placing separate anats in 001 002 etc (for obvious overwriting reasons). I believe this is documented in the recon-all help file, but could be confusing for some users at first.
Csurf, tkmedit, and tksurfer gui's loaded without a problem on my FC4 machine. Csurf loads surfaces fine, but I cannot get it to load mgz files. It seems to expect COR files. Out of habit, I did all of my volume edits using 'Edit Segmentation' in csurf, but may have to switch tkmedit/tksurfer. No problem there.
In sum, the current dev version for Redhat 9 appears to be running fine on a basic Fedora Core 4 machine.
Thanks, Brian
Hi Nick,
grep -H -R "/usr/pubsw/bin/perl -w" $FREESURFER_HOME/* Produced the results below, which appear to be the only files with hard links to the "/usr/pubsw/bin/perl."
grep -H -R "/usr/pubsw/bin/perl -w" $FREESURFER_HOME/* $FREESURFER_HOME/lib/mni/bin/make_template:#!/usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/nu_estimate_np_and_em:#!/usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/sharpen_volume:#!/usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/imp2field:#!/usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/nu_estimate:#!/usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/mritotal:#! /usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/autocrop:#! /usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/nu_correct:#!/usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/xfmtool:#! /usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/mritoself:#! /usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/nu_evaluate:#!/usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/field2imp:#!/usr/pubsw/bin/perl -w
My fix was simply: 1) mkdir -p /usr/pubsw/bin 2) ln -s /usr/bin/perl
The recon-all -subjid XXX -autorecon-all completed successfully in 27 hours on my Fedora Core 4 (AMD sempron 2600+) with surfaces having 0 holes. No manual edits were done.
The aseg.stats file is wonderful. Is there a way to overlay an individual's subcortical/cortical segmentation results on their t1.mgz volume (i.e, colorize voxels belonging to a specific structures)?
Thanks for such a great program! Brian
Brian,
Thanks for the info. We are currently updating the installation to fix- up the hard-coded links.
To view the subcortical segmentation results, for bert (the fully processed sample data), type:
tkmedit -segmentation mri/aseg.mgz \ $FREESURFER_HOME/FreeSurferColorLUT.txt bert norm.mgz
The label info is found in the 'Sgmtn label' box, as you move the mouse.
To view cortical parcellation labels:
tksurfer bert rh pial
then:
File->Label->Import Annotation Browse to file rh.aparc.annot and select OK Redraw View icon is the one with the up/down arrows
To increase brightness on that surface, in tksurfer:
View->Configure->Lighting->Brightness
FYI, we will soon be making available a 'wiki' site containing extensive and continuously updated documentation. In the meantime, the 'articles' link at the current website (https://surfer.nmr.mgh.harvard.edu) contains links to articles describing the techniques underlying freesurfer.
Nick
On Tue, 2005-09-06 at 23:03 -0700, Brian Schweinsburg, Ph.D. wrote:
Hi Nick,
grep -H -R "/usr/pubsw/bin/perl -w" $FREESURFER_HOME/* Produced the results below, which appear to be the only files with hard links to the "/usr/pubsw/bin/perl."
grep -H -R "/usr/pubsw/bin/perl -w" $FREESURFER_HOME/* $FREESURFER_HOME/lib/mni/bin/make_template:#!/usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/nu_estimate_np_and_em:#!/usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/sharpen_volume:#!/usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/imp2field:#!/usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/nu_estimate:#!/usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/mritotal:#! /usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/autocrop:#! /usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/nu_correct:#!/usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/xfmtool:#! /usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/mritoself:#! /usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/nu_evaluate:#!/usr/pubsw/bin/perl -w $FREESURFER_HOME/lib/mni/bin/field2imp:#!/usr/pubsw/bin/perl -w
My fix was simply: 1) mkdir -p /usr/pubsw/bin 2) ln -s /usr/bin/perl
The recon-all -subjid XXX -autorecon-all completed successfully in 27 hours on my Fedora Core 4 (AMD sempron 2600+) with surfaces having 0 holes. No manual edits were done.
The aseg.stats file is wonderful. Is there a way to overlay an individual's subcortical/cortical segmentation results on their t1.mgz volume (i.e, colorize voxels belonging to a specific structures)?
Thanks for such a great program! Brian
-- Brian C. Schweinsburg, Ph.D. Assistant Adjunct Professor of Psychiatry University of California, San Diego Office: 858-552-8585 x6109 Cell: 858-354-3756
From: Nick Schmansky nicks@nmr.mgh.harvard.edu Date: Tue, 06 Sep 2005 11:16:22 -0400 To: "Brian Schweinsburg, Ph.D." bschweinsburg@ucsd.edu Cc: Freesurfer Mailing List freesurfer@nmr.mgh.harvard.edu Subject: Re: [Freesurfer] new 'dev' release for Linux rh9, centos4 and centos4_x86_64
Brian,
I'm glad to hear that the installation is now working for you.
The recon-all -autorecon-all command can take 24 or more hours to complete, depending on the quality of the scan, and the difficulty of auto-fixing any defects, and of course the machine upon which it is running. Your machine appears to be more than adequate.
Concerning the perl scripts which needed editing, which files (with paths) required editing? I'd like to remove any hard-coded paths in our installation.
Concerning the mgz files, recon-all does expect to find each structural scan in the mri/orig directory, with filenames 001.mgz, 002.mgz, etc.. Given a subject having three sets of DICOMS (three runs of structural scans), I do the following to setup a subject:
cd $SUBJECTS_DIR mksubjdirs <subjid> cd <subjid> cd mri/orig mkdir 001 mkdir 002 mkdir 003 cp -r <path_to_dicoms>/95555-2-*.dcm 001/ cp -r <path_to_dicoms>/95555-3-*.dcm 002/ cp -r <path_to_dicoms>/95555-7-*.dcm 003/ mri_convert -it dicom -ot mgz 001/ 001.mgz mri_convert -it dicom -ot mgz 002/ 002.mgz mri_convert -it dicom -ot mgz 003/ 003.mgz
Now you will have three .mgz files in mri/orig, and recon-all will automatically take these three structural scans and average/motion- correct them.
The reason that we now encourage .mgz files over the COR format is because the COR format is limited to storage of volume data in unsigned char format only, and does not allow int, float, double or long formats. Also, the COR format historically assumes a slice size of 256 by 256 voxels. For these reasons, the COR format is deprecated in favor of the newer mgh (and mgz) format.
Concerning csurf, we no longer actively maintain csurf, in favor of recon-all. csurf is essentially an older copy of the recon-all script, and is not kept in-line with recon-all. We encourage using the recon- all script because it is easier to see what is going on, and to maintain. But we'll look into the possibility of at least csurf supporting the opening of .mgz files.
Nick
On Mon, 2005-09-05 at 21:58 -0700, Brian Schweinsburg, Ph.D. wrote:
Hi Nick,
I downloaded freesurfer-Linux-rh9-dev20050905-full.tar.gz and installed it on the following system: Fedora Core 4; AMD Sempron 2600+ on an ECS KT600-A board; Nvidia GeForce 2; 1 GB DDR 400 RAM.
'recon-all -subjid XXX -autorecon-all' would not run at first because it could not find some of the mni tools. The ones that it was having difficulty with were the perl scripts (not the binary MNI files). I noticed the first line in each of the scripts points to what looks like the MGH specific perl installation (i.e., /usr/pubsw/bin/perl), which does not exist on my system. I tested two different solutions 1) change the first line of each file and 2) mkdir -p /usr/pubsw/bin and then add a symbolic link to my perl binary. Both methods appear to work, and 'recon-all -subjid XXX -autorecon-all' has been running successfully for about 12 hours. I assume it is working!
I also noticed that recon-all expected the mgz files for each separate anatomical to both be located in $FREESURFER_HOME/subjects/XXX/mri/orig. This is opposed to what I recall for the old COR method of placing separate anats in 001 002 etc (for obvious overwriting reasons). I believe this is documented in the recon-all help file, but could be confusing for some users at first.
Csurf, tkmedit, and tksurfer gui's loaded without a problem on my FC4 machine. Csurf loads surfaces fine, but I cannot get it to load mgz files. It seems to expect COR files. Out of habit, I did all of my volume edits using 'Edit Segmentation' in csurf, but may have to switch tkmedit/tksurfer. No problem there.
In sum, the current dev version for Redhat 9 appears to be running fine on a basic Fedora Core 4 machine.
Thanks, Brian
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
freesurfer@nmr.mgh.harvard.edu