[wiki:Download top]

FreeSurfer Release Notes

These Release Notes cover what's new in a release, and known issues.

See also DownloadAndInstall, and BugReporting.

Stable v4.0.0

21 August 2007BR

Stable release version 4.0.0 is a major release with new features and bug fixes.

What's New

New features:
- The corpus callosum is now segmented into five separate labels in the subcortical segmentation volume 'aseg.mgz'
- White matter is now parcellated for gyral regions
- mris_wm_volume - calculates the total white matter volume
- Improved talairach alignment, with automatic failure detection
- Improved atlas-based skull-strip
- Longitudinal analysis (recon-all -long, see help section in recon-all --help)
- Qdec - [wiki:FsTutorial/QdecGroupAnalysis group analysis GUI tool]
- DTI recon stream: see dt_recon --help, also new: dmri_poistats (see --help) Linux Centos only
- FSFast:
  - Full NIFTI compliance
  - mkanalysis GUI -- set up analysis AND contrasts
  - selxavg3-sess -- now replaces stxgrinder, implicit intensity normalization, better whitening
  - tksurfer-sess -- replaces paint-sess and surf-sess
  - tkmedit-sess -- accepts multiple contrasts
  - isxconcat-sess + mri_glmfit -- intersubject averaging (RFx, FFx, weighted RFx), replaces func2sph-sess, func2tal-sess, and isxavg-??-sess, GRF
  - No longer needed: stxgrinder-sess, paint-sess, func2sph, fun2tal, isxavg-??, mkcontrast, inorm-sess
- Average subject subcortical segmentation
- Updated numerics libraries (NRC and GSL replaced by VXL), paving the way for an open-source release.
Bug fixes:
- Corrected midline parcellation
- Corrected aseg: removal of grey matter around callosum
- Corrected average subject surface area
- Corrected average subject thickness

Known issues

In tkmedit, the displayed volume after loading the talairach.xfm file is not correct. The work-around is to either load the talairach.lta file, or use tkregister2 instead. This will be fixed in the next release.
mris_annot_to_segmentation may segfault. A fixed version will appear in the next release.
In tksurfer, if you have both curvature and an overlay loaded, and you turn both off, the overlay stays on. This will be fixed in the next release.
When making flatmaps with mris_flatten, sometimes the map never unfolds properly (where there can be a section where it folds over itself). The fix for this bug will appear in a future release.
In qdec, the Analyze button can be hidden on small screens. This will be fixed in the next release.
The qdec app produces a 'Bad Match' error when started on some Linux platforms. There is no work-around at this time.
Input data having a FOV greater than 256 will fail in the -autorecon2 stage during mri_fill, because the wm.mgz file is 256³ while the norm.mgz and aseg.mgz files are FOV³ (where FOV > 256). The work-around is to use the --cropsize256 flag with recon-all with -autorecon1.
The surface files ?h.volume are not produced by recon-all yet, but are mention in mris_preproc. Production of the ?h.volume files by recon-all (calculating ?h.area * ?h.thickness) will be available in a future release.

Stable v3.0.5

7 Feb 2007BR

Stable release version 3.0.5 is a bug-fix release.

What's New

Bug fixes:
- Fixed bug in NIFTI reader in FreeSurfer (see details below in v3.0.4 Known Issues).
- Fixed bug in 'unpacksdcmdir' and 'mri_convert' commands on the Intel Mac distribution.
- Fixed bug in recon-all such that now xopts is correctly processed by mri_nu_correct.mni.

Known Issues

Input data having a FOV greater than 256 will fail in the -autorecon2 stage during mri_fill, because the wm.mgz file is 256³ while the norm.mgz and aseg.mgz files are FOV³ (where FOV > 256). The work-around is to use the --cropsize option of mri_convert (see its --help for details) to convert the orig.mgz file to 256³.
During cortical parcellation, areas of the anterior cingulate cortex (rostral anterior cingulate in the parcellation atlas), are mislabelled as corpus callosum or "unlabelled subcortical". A fix for this is in the works, possibly for the next release.
When making flatmaps with mris_flatten, sometimes the map never unfolds properly (where there can be a section where it folds over itself). The fix for this bug will appear in a future release.
On the Mac, make_average_subject (and make_average_volume and make_average_surface) will fail if more than 252 subjects are specified. The work-around is to type 'limit descriptors unlimited' before running these scripts. A fix will appear in the next release.
mris_anatomical_stats, as executed by recon-all, reports incorrect results for the measures 'average integrated rectified mean curvature', 'average integrated rectified Gaussian curvature', 'folding index', and 'intrinsic curvature index'. The other reported measures ('total white matter volume', 'total surface area', 'total gray matter volume' and 'average cortical thickness') are correct as reported by recon-all's execution of mris_anatomical_stats. There is no work-around. A fix will appear in the next release.

Stable v3.0.4

25 Nov 2006BR

Stable release version 3.0.4 is a bug-fix release.

What's New

Bug fixes:
- Fixed problem with left hemi of fsaverage subject (see v3.0.3 Known Issues, below, for details).
- Fixed bug whereby the pial surface can cross the midline in the cingulate/callosum area to the other hemisphere. The change [attachment:bert-medial-wall-diffs.tiff affects this area].
- Fixed 'glutInit' bug occuring on Fedora Core and Debian Linux installations using tksurfer.
- Fixed bug in mri_normalize which core dumped for low SNR input volumes.
- Fixed bug in tksurfer with "Don't cross functional threshold" parameter.
- Fixed problem with public declaration of Tcl/Tk/Tix/BLT environment variables which caused problems with FSL and 3D Slicer.
New features:
- Label file named <subjid>/label/?h.cortex.label is a single label for all of cortex.

Known Issues

Input data having a FOV greater than 256 will fail in the -autorecon2 stage during mri_fill, because the wm.mgz file is 256³ while the norm.mgz and aseg.mgz files are FOV³ (where FOV > 256). The work-around is to use the --cropsize option of mri_convert (see its --help for details) to convert the orig.mgz file to 256³.
The 'unpacksdcmdir' command, and 'mri_convert' of some DICOM formatted data, will fail on the Intel Mac distribution v3.0.4, due to a byte-endian issue. The v3.0.4b distribution has a fix for these problems.
recon-all does not process xopts for mri_nu_correct.mni. The fix will appear in the next release.
There is a small bug in the NIFTI reader in FreeSurfer, which causes the sform/qfrom matrix to be slightly off for oblique volumes that are non-isotropic (the more oblique and the more non-isotropic, the worse). Part of the error involves the matrix being slightly non-orthogonal. The fix for this bug will appear in the next release. There is NO effect of this bug if any of the following apply:
1. You use mri_convert to convert from DICOM (or any other non-NIFTI format) to NIFTI
2. Your volumes are not oblique
3. Your volumes are isotropic
During cortical parcellation, areas of the anterior cingulate cortex (rostral anterior cingulate in the parcellation atlas), are mislabelled as corpus callosum or "unlabelled subcortical". A fix for this is in the works, possibly for the next release.
When making flatmaps with mris_flatten, sometimes the map never unfolds properly (where there can be a section where it folds over itself). The fix for this bug will appear in a future release.

Stable v3.0.3

24 May 2006BR

Stable release version 3.0.3 is a bug-fix and feature-addition release.

What's New

Bug fixes:
- Fixed a bug in tksurfer in which if an overlay and the curvature was loaded, the curvature would draw under the overlay even if the curvature display was turned off.
- Modified the way NIFTI1 files are stored (these are the single .nii and .nii.gz files). The change is somewhat esoteric: the vox_offset is now 352 instead of 348. 352 is the NIFTI spec, so we were technically out of compliance before, although AFNI, FSL, and SPM did not seem to have a problem with 348.
- Fixed bug in make_average_surface and make_average_volume scripts which limited the number of subjects that could be specified on the command line.
New features:
- asegstats2table and aparcstats2table - scripts to generate text/ascii tables of freesurfer stats data (eg, aseg.stats). This can then be easily imported into a stats program.

Known Issues

In tksurfer, if using the -fthresh option, specify the number as a float (-fthresh 2.0) not an integer (don't use -fthresh 2).
There is a problem with the left hemi of the fsaverage subject, specifically lh.sphere.reg. This will cause some inaccuracies in the visualization of maps on the left hemi of fsaverage and other average subjects (but the statistics themselves should be ok). The corrected lh.sphere.reg for fsaverage will be included in the next release. Or, if you have an average subject that you want to correct, you do not have to rebuild it. Rather, you can simply:

cd youraveragesubject/surf
cp lh.sphere.reg lh.sphere.reg.old # backup
cp rh.sphere.reg lh.sphere.reg # copy rh.sphere to lh.sphere

There is a bug whereby the pial surface can cross the midline in the cingulate/callosum area to the other hemisphere. There is no workaround. The bug will be fixed in the next release. This change will [attachment:bert-medial-wall-diffs.tiff affect this area].
tksurfer will fail on Fedora Core and Debian Linux installations which use freeglut, as 'glutInit' is not executed. This bug will be fixed in the next release.
mri_normalize may core dump for low SNR input volumes. This bug will be fixed in the next release.
In tksurfer, fills that use the "Don't cross functional threshold" parameter do not correctly fill negative functional areas. This bug will be fixed in the next release (v3.0.4).
Tcl/Tk/Tix/BLT environment variables declared in freesurfer setup file causes problems with other apps, like FSL and 3D Slicer. This problem will be fixed in the next release. The workaround is to not call the Freesurfer setup script when using the other apps.

Stable v3.0.2

18 April 2006BR

Stable release version 3.0.2 is a bug-fix and feature-addition release.

What's New

Bug fixes:
- The --all-info or --version command-line corrupting bug has been fixed.
- The mris_fix_topology mrisDefectAnalysis bug has been fixed.
- The bug where control points were being truncated instead of rounded by mri_normalize, which could offset them by 1 voxel, has been fixed.
- The bug where curvature was being computed incorrectly for RH9 binaries has been fixed (affected spherical morph, and curvature overlay).
- eTIV calculations are now correct (they were 25% too large).
- mri_label2vol - fixed bug for segmentation IDs over 999. Runs much faster now.
New features:
- bem directory is created in subjects dir (for the MEG people)
- Better integration with FSL (compatible with FSL-nifti)
- Uses robust estimates of class statistics in mri_normalize, mri_segment and mris_make_surfaces. Helps in low contrast (such as SPGR-based) images in which the gray matter is oversegmented.

Known Issues

None at this time.

Stable v3.0.1

13 March 2006BR

Stable release version 3.0.1 is a bug-fix release.

What's New

FIX_VERTEX_AREA no longer need be set in the environment outside of recon-all.
In tkmedit, while loading a functional overlay or time course, the registration file name is now read properly.
The FSFast utility mkanalysis-sess.new now accepts only sequence names matching "TR " (instead of all occurrences of TR).

Known Issues

When running mri_surfcluster with average7 inside your csd file, make sure to specify --subject average7 --really-use-average7 or else it will underestimate the clustersize and cause the significances to be too low.
mris_fix_topology may exit recon-all with the following message in the log: mrisDefectAnalysis : defect 123 overlap defect 119, where your defect numbers will be different. If this happens for one of your subjects, there is no work-around. A fix will appear in the next release.
If the flags --all-info or --version are included at the beginning of a command-line, it will corrupt the subsequent command-line arguments. The work-around is to not include --all-info or --version with other command-line arguments. A fix will appear in the next release.
On the rh9 distribution, sometimes the ?h.curv files produced are incorrect, displaying as nearly all red (instead of green and red, indicating gyri and sulci). This produces incorrect cortical parcellation data. There is no known work-around at this time. This problem does not appear when using the Centos or Mac distributions.
tkmedit saves control points for intensity normalization in floating point RAS coordinates, but mri_normalize truncated these when they are read in with the -f command (the default behavior when the tmp/control.dat file exists). This can result in the control points being displaced by a voxel from the location that tkmedit displays them.
mri_label_volume estimated TIV calculation are about 25% too large. This will be fixed in the next release.
Be sure to read about the changes to recon-all, detailed below, and also read about the new [wiki:FsAverage fsaverage] subject.

Stable v3.0

5 March 2006BR

Stable release version 3.0 is a major release containing many new features over the prior stable release (version 2.2).

Important! The [wiki:recon_2dall recon-all] script has been updated. Please read the ReconAllChanges page which describes these changes! Users of prior 'dev' releases should also take note of the changes.

COR file format is no longer accepted by recon-all. mgz file's have replaced this old format.
brainmask.mgz has replaced what was the brain volume. You should visualize and make all edits to brainmask.mgz and not to brain.mgz
recon-all will now save all your changes by default. Any manual changes/edits to the talairach, setting of control points, edits to volumes etc will be saved and used by default and you must specify when you wish to ignore these changes.
Data processed with an older release should make use of the -legacy flag in [wiki:recon_2dall recon-all].

What's New

Sub-cortical segmentation - Each MRI volume voxel is assigned one of 37 possible neuroanatomical labels, including left and and right caudate, putamen, pallidum, thalamus, lateral ventricles, hippocampus, and amygdala. The labeling allows for volumetric measurement of these structures. The segmentation information also improves the ability to create a filled white-matter volume.
Automated filling of ventricles and basal ganglia - Improves the ability to create a filled white-matter volume (which is used in the surface creation process). Note: this requires the use of the new average subject [wiki:FsAverage fsaverage] because the surface registration changes.
Improved automation - The need to conduct manual edits to account for skull-stripping and cutting-plane problems is reduced. The topology fixer is much better at handling temporal lobe strands. Overall, greater than 90% of hemispheres processed by recon-all do not require manual intervention.
Improved GUI tools - The GUI tools tkmedit and tksurfer incorporate many usability improvements requested by the user-base.
New cortical parcellation - A neuroanatomical label is assigned to each location on the cortical surface model based on probabilistic information estimated from a manually labeled training set.
Surface-based GLM analysis - Utilities for conducting group analysis studies with GLM.
Multiple comparison correction for surface analysis - False discovery rate (FDR), permutation and Monte Carlo simulations.
New average subject. [wiki:FsAverage $FREESURFER_HOME/subjects/fsaverage]. Keeps info about average surface area.
Average subject customization. Make your own average subject from your cohort.
Extensive testing - The recon-all processing stream has been tested against several hundred subjects, different pulse-sequence types, and different MRI system platforms.
Online tutorial of the Freesurfer work-flow - This release is compatible with the [wiki:FsTutorial on-line tutorial]. Tutorial data must be downloaded separately.
Simplified distribution download and installation - The critical library and toolkit dependencies (MNI, FSL, GSL, Tcl/Tk/Tix/BLT, and Qt) are included to ease installation. These third-party open-sourced packages [wiki:SetupConfiguration can be removed if they are already installed on the target system] (although full Freesurfer functionality is no longer assured). For Mac OS X, a GUI installation package is provided.
Compatible with all mouse pads
Now Y2K compliant!

Known Issues

Make sure to run "setenv FIX_VERTEX_AREA" if using any commands outside of recon-all, particularly the mri_glmfit and mri_surfcluster programs. To be safe, you can just add the line "setenv FIX_VERTEX_AREA" into FreeSurferEnv.csh or your .cshrc file. This will only affect those downloading the new version. This will automatically be done for those in the Martinos Center.
In tkmedit, while loading a functional overlay or time course, the registration file name may not be read properly. A workaround is to load the overlay or time course from the command line (see ["tkmedit"] for details).
The surface registration curvature target has changed with the new automatic filling of ventricles and basal ganglia. Do not mix newly reconstructed subjects with those that used the old target. For new data recons, display overlay maps on the new average subject [wiki:FsAverage fsaverage]. For older data, use average7.
There are some changes in the way that mri_surfcluster and mri_glmfit computes cluster sizes and how these interact with the average subject (and average7) see the notes [wiki:FsAverage fsaverage] for more detail.