Anatomical Processing Script: fsl_anat (BETA version)
This tool provides a general pipeline for processing anatomical images (e.g. T1-weighted scans).
Most of the pipeline involves standard use of FSL tools, but the bias-field correction has been substantially improved, especially for strong bias-fields typical of multi-coil arrays and high-field scanners.
The stages in the pipeline (in order) are:
reorient the images to the standard (MNI) orientation [fslreorient2std]
automatically crop the image [robustfov]
bias-field correction (RF/B1-inhomogeneity-correction) [FAST]
registration to standard space (linear and non-linear) [FLIRT and FNIRT]
brain-extraction [FNIRT-based or BET]
tissue-type segmentation [FAST]
subcortical structure segmentation [FIRST]
The overall run-time is heavily dependent on the resolution of the image but anything between 30 and 90 minutes would be typical.
Various stages in the pipeline can be turned off. Currently not all combinations have been tested and so users are advised to only turn off the first few or last couple of stages currently.
Usage: fsl_anat [options] -i <structural image> fsl_anat [options] -d <existing anat directory> Arguments (You may specify one or more of): -i <strucural image> filename of input image (for one image only) -d <anat dir> directory name for existing .anat directory where this script will be run in place -o <output directory> basename of directory for output (default is input image basename followed by .anat) --clobber if .anat directory exist (as specified by -o or default from -i) then delete it and make a new one --weakbias used for images with little and/or smooth bias fields --noreorient turn off step that does reorientation 2 standard (fslreorient2std) --nocrop turn off step that does automated cropping (robustfov) --nobias turn off steps that do bias field correction (via FAST) --noreg turn off steps that do registration to standard (FLIRT and FNIRT) --nononlinreg turn off step that does non-linear registration (FNIRT) --noseg turn off step that does tissue-type segmentation (FAST) --nosubcortseg turn off step that does sub-cortical segmentation (FIRST) -s <value> specify the value for bias field smoothing (the -l option in FAST) -t <type> specify the type of image (choose one of T1 T2 PD - default is T1) --nosearch specify that linear registration uses the -nosearch option (FLIRT) --betfparam specify f parameter for BET (only used if not running non-linear reg and also wanting brain extraction done) --nocleanup do not remove intermediate files
the bias-field correction assumes that the field is "strong", typical of that arising from a multi-coil array or a high-field scanner. For images acquired using birdcage coils or on 1.5T scanners, the --weakbias option will be faster and may produce equally good results.
- the brain extraction is based on transforming a standard-space mask to the input image using the FNIRT (non-linear) registration
Using the -d option the script can be run again (with a subset of stages) to update an existing result.
This section describes the main output files - it is not a complete list, but highlights the most important outputs.
The output directory will end with .anat and by default will have the same basename as the input image (and be in the same directory). If the -o option is used the directory name will use the specified name, followed by .anat
The specified input image is copied into the output directory and named T1, T2 or PD, depending on the setting of the -t option (default is T1).
Reorientation and Cropping
If run, the original image (we shall call it T1 from here on in as an example), will be replaced by the reoriented and/or cropped version. The original versions are saved as files: T1_orig and T1_fullfov.
In addition, transformation files are provided to allow images to be moved between spaces: i.e. T1_orig2std.mat and T1_nonroi2roi.mat and their inverses and combinations.
The bias-corrected version of the image is called T1_biascorr.
Registration and Brain-Extraction
The registration (to standard space) produces the following images that are in MNI space with a 2mm resolution:
T1_to_MNI_lin (linear registration output)
T1_to_MNI (non-linear registration output)
T1_to_MNI_nonlin_field (non-linear warp field)
T1_to_MNI_nonlin_jac (Jacobian of the non-linear warp field)
T1_vols.txt - a file containing a scaling factor and brain volumes, based on skull-contrained registration, suitable for head-size normalisation (as the scaling is based on the skull size, not the brain size)
The brain-extraction produces:
Tissue-type segmentation (done with FAST) produces:
T1_biascorr - refined again in this stage
T1_fast_pve_0, T1_fast_pve_1, T1_fast_pve_2 - partial volume segmentations (CSF, GM, WM respectively)
T1_fast_pveseg - a summary image showing the tissue with the greatest partial volume fraction per voxel
Subcortical segmentation (done with FIRST) produces:
T1_subcort_seg - summary image of all subcortical segmentations
all other outputs in the first_results subdirectory
'T1_first_all_fast_firstseg - same as T1_subcort_seg`
- a host of other images relating to individual segmentations
T1_biascorr_to_std_sub.mat (in the main anat directory) - a transformation matrix of the subcortical optimised MNI registration