Guide brain extraction in whatever runtime is available: local shell, Conda
environment, container, module-based cluster, or scheduler-backed HPC system.
Do not assume Neurodesk, SLURM, Lmod modules, or any specific install path.
-
Identify the input image
Read the request for a NIfTI path. If none is given, ask for one.
Verify it exists:
ls "<input_path>"
Check that it ends in .nii or .nii.gz. If not, warn that it may not be a
NIfTI and ask for confirmation.
Check that the input is suitable for brain extraction:
python3 -c "import nibabel as nib; img = nib.load('<input_path>'); print(img.shape)"
If it is a 4D timeseries, ask the user whether to extract a single reference
volume or use a tool appropriate for functional data.
-
Identify outputs
If an output prefix was provided, use it. Otherwise default to the input
directory with _brain appended before the extension. Set OUT_DIR,
OUTPUT, and MASK or tool-specific equivalents.
-
Present tool options
Load references/tool-comparison.md and present the comparison table. Ask
which tool the user wants. If they ask for a recommendation:
- Default anatomy T1w, no special constraints -> FSL BET
- Difficult anatomy, low contrast, or high accuracy needed -> mri_synthstrip or HD-BET
- Multi-modal T2, FLAIR, DWI, or unusual contrast -> mri_synthstrip
- Maximum accuracy and a suitable template is available -> ANTs
- Functional/EPI reference data -> AFNI 3dSkullStrip or 3dAutomask
-
Discover the runtime
After tool selection, verify availability without assuming a platform:
command -v <executable>
<executable> --help 2>&1 | head -40
If the executable is unavailable, check likely environment mechanisms:
command -v module >/dev/null && module avail <tool_or_module_name>
command -v conda >/dev/null && conda env list
command -v apptainer >/dev/null || command -v singularity >/dev/null || command -v docker >/dev/null
Use the reference file for tool-specific executable names and install hints.
For HD-BET, the executable is usually hd-bet; the module/package name may
be hdbet.
-
Write a reproducible script when appropriate
Load the selected reference:
- FSL BET ->
references/fsl-bet.md
- ANTs ->
references/ants-brain-extraction.md
- mri_synthstrip / SynthStrip ->
references/synthstrip.md
- HD-BET ->
references/hd-bet.md
- AFNI 3dSkullStrip ->
references/afni-3dskullstrip.md
Determine the next script name:
LAST=$(ls analysis_[0-9][0-9]_*.sh 2>/dev/null | sort | tail -1)
N=$([ -z "$LAST" ] && echo "01" || printf "%02d" $(( ${LAST:9:2} + 1 )))
Write analysis_<N>_brain_extraction.sh with:
#!/usr/bin/env bash and set -euo pipefail- input/output variables and
mkdir -p for output/log directories
- only the environment setup needed on the current system
- the selected extraction command
- output existence checks
If a scheduler is available and useful for the job, include or create the
appropriate submission wrapper. Otherwise make the script runnable directly.
-
Run or submit
For short/local jobs, run the script directly after confirming with the user
when the command is potentially long-running. For scheduler-backed systems,
submit with the local scheduler command (sbatch, qsub, bsub, etc.) and
report how to monitor logs/status.
-
Validate and hand off to QC
Verify the expected brain image and mask exist and are non-empty:
test -s "<expected_output>"
test -s "<expected_mask>"
Then invoke the brain-extraction-qc workflow with:
brain-extraction-qc <original_input> <brain_mask> <output_prefix>.
