FSL_anat.sh — FSL Anatomical Preprocessing¶

Location: BRC_structural_pipeline/scripts/FSL_anat.sh
Authors: Ali-Reza Mohammadi-Nejad, Stamatios N Sotiropoulos
Last updated: 02 Oct 2018

Purpose¶

A comprehensive standalone FSL anatomical preprocessing script closely modelled on FSL's fsl_anat tool, customised for the BRC pipeline. It accepts a single or list of structural MRI images (T1, T2, or PD) and performs the full FSL anat pipeline ending in an <output>/temp.anat/ directory, whose outputs are subsequently organised by move_rename.sh.

Arguments¶

Argument	Default	Description
`-i <image>`	—	Input structural image
`-o <name>`	—	Output name stem; creates `<name>/temp.anat/`
`-d <dir>`	—	Use an existing `.anat` directory
`-s <smooth>`	`20`	Bias field smoothness in mm
`-m <mask>`	—	Lesion mask (inverted for bias field and registration)
`-t <T1\\|T2\\|PD>`	`T1`	Image type for FSL FAST tissue model
`--list <csv\\|file>`	—	Comma-separated list or file of images to average
`--clobber`	off	Overwrite existing `.anat` output directory
`--noreorient`	off	Skip reorientation
`--nocrop`	off	Skip auto-cropping
`--nobet`	off	Skip brain extraction
`--noreg`	off	Skip registration to standard space
`--nononlinreg`	off	Skip non-linear registration
`--noseg`	off	Skip tissue segmentation
`--nosubcortseg`	off	Skip subcortical segmentation
`--nobias`	off	Skip bias field correction
`--strongbias`	off	Use strong bias correction (niter=5, smooth=10)
`--weakbias`	off	Use weak bias correction (default: niter=10, smooth=20)
`--betfparam <f>`	`0.1`	BET fractional intensity threshold
`--nocleanup`	off	Retain all intermediate files
`--anatbasedFS`	off	Use FreeSurfer brain mask in place of FNIRT-derived mask
`-logfile <path>`	—	Log file path
`-mridir <path>`	—	FreeSurfer `mri/` directory (required with `--anatbasedFS`)
`-orig <image>`	—	Original (pre-crop) T1 for FOV extraction reference

Pipeline Stages¶

Averaging Multiple Scans (optional)¶

If --list is provided, images are averaged using AnatomicalAverage.

Negative Range Fix¶

fslstats T1 -p 0   # min value
fslstats T1 -p 100 # max value

If negative values are present among positive ones, the minimum is subtracted. If all values are negative, they are made positive while preserving zeros.

Reorientation¶

fslreorient2std T1 > T1_orig2std.mat

Forward and inverse reorientation matrices are saved for downstream transforms.

Auto-Cropping¶

robustfov -i T1_fullfov -r T1 -m T1_roi2nonroi.mat

Crops the FOV and records the crop-to-full-FOV transform.

Lesion Masking¶

If a lesion mask is provided, it is reoriented and cropped to match the processed T1. A binary inverse mask (lesionmaskinv) is used to exclude lesion voxels from bias estimation and registration.

Bias Field Correction¶

Weak bias (default)Strong bias (--strongbias)

A single BET pass gets a rough brain mask, followed by two rounds of FSL FAST bias estimation:

bet T1 T1_initfast2_brain -m -f 0.1
fast -o T1_fast -l 20 -b -B -t 1 --iter=10 --nopve --fixed=0 T1_initfast2_maskedrestore

An extra large-scale correction step using quick_smooth precedes the FAST-based correction:

# Stage 1: large-scale (subsampling-based smoothing)
fslmaths T1 -div T1_s20 T1_hpf
# Stage 2: detailed FAST-based correction
fast -o T1_initfast -l 10 -b -B -t 1 --iter=5 --nopve --fixed=0 T1_hpf2_maskedbrain

Registration and Brain Extraction¶

flirt -interp spline -dof 12 -in T1_biascorr -ref MNI152_T1_2mm \
      -omat T1_to_MNI_lin.mat
fnirt --in=T1_biascorr --ref=MNI152_T1_2mm --fout=T1_to_MNI_nonlin_field \
      --cout=T1_to_MNI_nonlin_coeff --aff=T1_to_MNI_lin.mat

The MNI brain mask is warped back to native space via the inverse FNIRT warp for brain extraction.

Tissue-Type Segmentation¶

fast -o T1_fast -l 20 -b -B -t 1 --iter=10 T1_biascorr_maskedbrain

The bias field is refined using FAST, and the corrected image is re-warped to MNI.

Skull-Constrained Volume Estimation¶

pairreg MNI152_T1_2mm_brain T1_biascorr_bet \
        MNI152_T1_2mm_skull T1_biascorr_bet_skull \
        T12std_skullcon.mat
avscale T12std_skullcon.mat  # → vscale

Grey and white matter volumes are scaled by vscale and written to T1_vols.txt.

Subcortical Segmentation¶

first_flirt T1_biascorr T1_biascorr_to_std_sub
run_first_all -i T1_biascorr -o first_results/T1_first \
              -a T1_biascorr_to_std_sub.mat

Internal Helper Functions¶

Function	Description
`quick_smooth(in, out)`	Fast Gaussian smoothing approximation via 4× subsampling + FLIRT resampling. Used for large-scale bias estimation.
`get_opt1(arg)`	Extracts option name (before `=`) from a `--key=value` argument
`get_arg1(arg)`	Extracts value (after `=`) from `--key=value`; exits if no value
`get_imarg1(arg)`	Like `get_arg1` but also strips the image extension using `remove_ext`
`get_arg2(opt, val)`	Extracts value from `-key value` two-token pair
`run(cmd...)`	Logs the command at level 2 then executes it