WHAT'S NEW IN THIS RELEASE

Release 3.2β, September 2004

Note: the commands for setting up your environment to use FSL have changed very slightly - see the Running section of downloading and installing.

• FDT (FMRIB's Diffusion Toolbox) - first release of this complete toolkit for analysis of diffusion data, including probabilistic tractography.
• FLOBS (FMRIB's Linear Optimal Basis Functions) - a new method for finding an optimal set of HRF convolution basis functions, plus a Bayesian estimation method for improving signal/noise separation using these basis functions.
• SMM (Spatial Mixture Modelling) - a new tool for alternative hypothesis testing on statistic images using histogram mixture modelling with spatial regularisation of the voxel classification into activation and non-activation.
• Randomise - a randomisation-based inference tool for nonparametric statistical thresholding.
• BET v2 - can now estimate mesh representations of the inner and outer skull surfaces, and outer scalp surface, for example for use in EEG/MEG source modelling.
• NIFTI - FSL now uses the NIFTI-1 data format by default (though can still read and write old Analyze files). Single-file NIFTI-1 images can be read and written already compressed, saving much disk space.
• FEAT v5.4 new features:
  • Use of FLOBS (optimised HRF basis sets).
  • Fixed-effects higher-level analysis (e.g. for within-subject cross-session analysis).
  • Option to prevent cleanup of standard-space first-level stats.
  • User preferences file (and network-wide preferences file) which can set FEAT preferences for loading on startup.
  • Inclusion of MELODIC data exploration at the end of Pre-stats.
  • Time-series (data vs model) plotting inside Featquery.
• SIENA - can now do voxelwise cross-subject statistical atrophy analysis. Also, there is a small change to the partial volume estimation in FAST, causing slight differences in volume estimates; any multi-subject analysis using FAST/SIENAX estimated volumes must only use the previous version or only use this new version.
• fslerrorreport is a new script which should always be run before emailing the FSL email list with a problem concerning running programs - the output from this will give us useful information about your system.

Release 3.1, July 2003

• FLAME (Bayesian group stats modelling and estimation in FEAT5): updated to give increased estimation accuracy at the MCMC stage, and with added option for "FLAME stage 1 only" (nearly as fast as OLS and nearly as accurate as full FLAME).
• Beta-release of FSL for WindowsXP using Cygwin.
• Beta-release of FSLView interactive 3D/4D viewer (currently compiled for Linux/MacosX/Windows i.e. not yet IRIX/SunOS).
• Various minor additions (such as progress viewer for FEAT) and bug-fixes.

Release 3.0, October 2002

• FEAT v5:
  • Bayesian multi-subject FMRI analysis: General GLM for higher-level analysis (e.g. analysis across sessions or across subjects; paired t-tests, three-level groupings etc): FLAME (FMRIB's Local Analysis of Mixed Effects). FLAME uses very sophisticated methods for modelling and estimating the random-effects component of the measured inter-session mixed-effects variance, using MCMC to get an accurate estimation of the true random-effects variance and degrees of freedom at each voxel.
  • Slice timing correction
  • Basis function HRF convolutions
  • Pre-threshold masking
  • Contrast masking after thresholding
  • Optionally registration can use two structural images as well as standard space image.
  • Peristimulus timing plots in the web report.
  • Featquery - a program which allows you to interrogate FEAT results by defining a mask or co-ordinates (in standard-space, highres-space or lowres-space) and get mean stats values and time-series.
  • Although IRVA (semi-model-free analysis) is still bundled with FSL, it is no longer an option within FEAT. This is because this kind of analysis is effectively covered as a subset of the functionality of FIR basis function analysis in FEAT.
• MELODIC v2:
  • Probabilistic ICA (PICA) for FMRI. By adding a Gaussian noise model to the ICA model, and automatically estimating the dimensionality (the number of non-Gaussian components), it is now possible to create meaningful Z-statistic maps for each spatial component, allowing inference ("thresholding").
  • Inference on the Z-statistic images is carried out with a Gaussian/Gamma mixture model (fitted using EM), fitting central noise and activation classes; thus alternative hypothesis testing is used.
• Interoperation with FreeSurfer - it is now very easy to display FEAT results onto FreeSurfer-generated inflated or flattened surfaces, and also easy to carry out group FEAT analysis in standard-spherical-surface space instead of 3D standard space.
• Interoperation with AFNI - AFNI can now easily read FSL (particularly FEAT) output for easy visualisation.
• SIENA v2: increased accuracy by forcing all steps to be symmetric; added masking functionality including standard-space-based Z limits.
• FLIRT&MCFLIRT; now run on single-slice (or few-slice) data; the flirt -applyxfm option now allows the processing of 4D data; new GUIs - ApplyXFM, ConcatXFM.
• BET: new command-line options for surface initialisation; "bet" now runs on binarised and on 2D images; the "betpremask" script uses FLIRT registration to do standard-space masking instead of running "bet"; bet now works on 4D images in two possible ways (the "betfunc" script is suited to motion-corrected FMRI data, applying over-inclusive masking, whilst just running "bet" on 4D data estimates the brain mask from the first image and applies this to the rest);

Releases 2.0 - e, internal development versions

Release 1.3 for Mac OS X, November 2001

Release 1.3, June 2001:

• New: MELODIC - ICA-based model-free analysis of FMRI (and other 4D) data.
• New: FAST - Brain segmentation (into different tissue types) and bias field correction.
• New: FUGUE - Unwarps geometric distortion in EPI images using B₀ field maps.
• New: SIENA - Structural brain change analysis, for estimating brain atrophy.
• FEAT 4.03: new features include F-tests and improved organisation of GLM-setup GUI.
• FLIRT: new features include image masking/weighting, different output interpolations, and allowing 2D registration.

Release 1.2a, May 2001 (not an FSL release but the version bundled with MEDx 3.40):

• FEAT 4.02: new features include F-tests and improved organisation of GLM-setup GUI.

Release 1.2, December 2000:

• All GUIs can now run standalone - i.e., nothing in FSL needs MEDx in order to run.
• FSL now includes MCFLIRT motion correction, which is more accurate, and faster, than the other tested MC methods. FEAT uses MCFLIRT only.
• FILM (the GLM method inside FEAT) now uses general least-squares, prewhitening the data before the final fitting. This gives more efficient activation estimation than the other methods tested, particularly for dense single-event experiments. It includes autocorrelation correction, using smoothing of autocorrelation coefficients and nonlinear spatial filtering to achieve robust estimation.
• A new GUI "fsl" which calls the other main FSL tools.
• The "model-free ANOVA" method has been renamed "IRVA" (Inter-Repetition Variance Analysis) to avoid later confusion with ICA truly model-free analysis (to be released shortly).

Release 1.1, August 2000:

• FEAT (FMRI analysis) will now use FLIRT (registration) to automatically register your FMRI data to a high-resolution structural and/or standard image for you, without having to use the FLIRT GUI separately.
• FEAT can now run group statistics (fixed-effects and random-effects) across groups of FEAT directories, either in a single long FEAT run (consisting of multiple first-level analyses followed by group-stats) or on previously created FEAT directories.
• You can now explicitly demand that certain EVs in FEAT be orthogonalised with respect to other EVs in the design matrix.
• FSL now works with both the latest MEDx release (3.30) and the previous version (3.20).

Release 1.0 (beta), June 2000. First ever release, containing:

• BET
• SUSAN
• FLIRT
• FEAT
• IRVA (then referred to as "Model-free ANOVA")