atlasBREX: Automated averaged template-derived brain extraction in animal MRI

alt text

Due to optimization for the human brain, most common skullstripping/brain-extraction methods, such as AFNI's 3dSkullStrip or FSL's BET, achieve insufficient results for non-human brains, which then require further manual intervention. Making use of the available brain-extraction from a template/atlas, this approach implements brain-extraction through reversal (rigid- and non-rigid) deformation of a template-derived mask.

• time-saving and straightforward (with various optional parameters for further optimization)
• multi-step registration (2- or 3-step) for improved registration to low resolution datasets
• robust FSL (FLIRT, FNIRT) or ANTs (SyN) registration frameworks
• compatible with T1-/T2-weighted datasets

• FSL (FMRIB Software Library)
• Bash >= 4 (echo $SHELL)

Optional:

• AFNI (Analysis of Functional NeuroImages): 3dAutomask, 3dUnifize
• ANTs (Advanced Normalization Tools): antsRegistration, antsApplyTransforms, N4BiasFieldCorrection, ImageMath and antsRegistrationSyN

Last tested with ANTs v2.3.4, AFNI_20.0.18 ('Galba') and FSL v6.0.3.

v1.5 - non-linear registration between skullstripped template and target volume for low-resolution volumes.

v1.2 - SyN makes use of an initial transform.

v1.1 - Fixed an error regarding the NIFTI header.

bash atlasBREX.sh -b <input> -nb <input> -h <input> -f <input>
(no multi-line commands)

• b_template_brain.nii.gz (brain-extracted template)
• nb_template.nii.gz (whole-head template)
• sj_170308_1.nii.gz (subject #1)
• sj_170308_2.nii.gz (subject #2)
• sj_170308_3.nii.gz (subject #3)

Copy all gzipped (.nii.gz) NIFTI volumes and atlasBREX.sh into a common folder. There should be no directory or file containing 'orig', 'temp' or '_.nii.gz' in this folder.

1. Step: Brief interactive pilot run (approx. 2-3 minutes) using the -f n flag to determine a reasonable fractional intensity value. AtlasBREX will propose 3 images for selection (choose the option with least extracranial tissue) during this pilot-run.

• use the -nrm 1 flag for T1w intensity normalization, where AFNI is available (improves provisional brain-extraction and registration accuracy)
• if brain regions appear clipped using FLIRT/FNIRT, try to adjust the fractional intensity or reduce the FOV using robustfov (FSL) to improve preliminary brain-extraction.
• if available, try -reg 2 for the ANTs registration framework

bash atlasBREX.sh -b b_template_brain.nii.gz -nb nb_template.nii.gz -h sj_170308_1.nii.gz -f n

2. Step: Run atlasBREX (with non-linear deformation) on all subjects in an automated and unattended manner:

for file in *'sj_'*'.nii.gz'*
 do
 bash atlasBREX.sh -b b_template_brain.nii.gz -nb nb_template.nii.gz -h $file -reg 1 -w 10,10,10 -msk a,0,0 -f 0.2
 wait
done

• Sample dataset: https://github.com/jlohmeier/atlasBREX/tree/master/demonstration
• Example helper script: https://github.com/jlohmeier/atlasBREX/blob/master/example.sh
• Visual (hands-on) walkthrough w/ rhesus macaque (public sample data from Allen Institute for Brain Science)

Compulsory arguments:

-b brain-extracted atlas or template
-nb whole-head (non-brain) atlas or template
-h target 3D volume
-f fractional intensity threshold [1 > n > 0] for provisional (BET) brain-extraction (e.g. -f 0.2).
 interactive pilot: [-f n] proposes 3 default thresholds for user selection. 
 for multi-step registration, different values for high-res, 
 low-res and native volumes can be entered (e.g. -f 0.2,0.5,0.8)

Optional arguments:

-l low-resolution 3D volume (3-step registration)
-n (functional) 3D/4D volume (2-/3-step registration)
-t/-tmp disable removal [1] of temporary files (default: 0) 
-w/-wrp define FNIRT (FSL) warp-resolution (e.g. -wrp 10,10,10),
 for SyN (ANTs) enter warp [-wrp 1] flag (e.g. -wrp 1 -reg 2)
-r/-reg FNIRT w/ bending- [-reg 0] or membrane-energy regularization [-reg 1] 
 ANTs/SyN w/ [-reg 2] or w/o [-reg 3] additional N4BiasFieldCorrection (def: 1)
-nrm provisional intensity normalization w/ T1 [-nrm 1] or T2 [-nrm 2] (req: AFNI)
 (recommended for low-resolution volumes)
-lr optimized parameter settings for low-resolution volumes:
 [1] non-linear registration between whole-head template and target volume with mask
 [2] non-linear registration based on skullstripped template and target volume (def: 0)
-msk mask binarization threshold (in %) for fslmaths 
 w/ optional erosion and dilation (e.g. -msk b,10,0,0) (def: b,0.5,0,0)
 [-msk b,[100 < n > 0] for threshold, [0-9] for n-times erosion,
 [0-9] for n-times dilation] or 3dAutomask (e.g. -msk a,0,0, req: AFNI) 
 [-msk a,[0-9] for n-times erosion,[0-9] for n-times dilation]
-vox provisional voxel-size adjustment [-vox 1] (def: 0)
-dil n-times dilation of the brain-extraction from linear registration 
 prior to non-linear registration (e.g. -dil 4)

• use -help/--help for further details.
• see log.txt for a summary after running the script.
• use -f n during the test-run to determine a suitable fractional intensity threshold. atlasBREX will propose 3 brain-extractions and lets you choose at the beginning of the procedure.
• use -nrm flag for intensity normalization (AFNI required) for low-resolution volumes (both T2w and T1w) and (useful for) T1w scans.
• use -dil flag in case of registration failures during non-linear registration. (n >= 4 as recommended starting point)
• each (non-template) input volume will be brain-extracted, appended with a "_brain.nii.gz" or "_brain_lin.nii.gz" suffix.
• adjust warp-resolution if you encounter issues with memory (e.g. std::bad_alloc).
• non-linear transformation matrix and warp file are preserved (to reduce total computation time for brain-extraction of multiple downstream input volumes).
• if result appears miswarped, review your volumes for inconsistencies regarding q- and s-form within NIFTI headers.
• number of threads can be set within the script for parallel computing (ANTS).
• make sure you're using Git LFS when downloading the sample files: https://developer.lsst.io/git/git-lfs.html

• FSL: http://fsl.fmrib.ox.ac.uk
• AFNI: https://afni.nimh.nih.gov/afni/
• ANTs: https://github.com/stnava/ANTs
• ABSORB: https://www.nitrc.org/projects/absorb

If you use this application or part of its source code, please cite atlasBREX with a reference to this Github page.

atlasBREX: Automated template-derived brain extraction in animal MRI
Scientific Reports, volume 9, Article number: 12219 (2019)
DOI: 10.1038/s41598-019-48489-3

Integrated in Macapype: https://macatools.github.io/macapype/generated/macapype.nodes.extract_brain.AtlasBREX.html

Selected preclinical studies that have used atlasBREX:

Lohmeier J, Silva R, Tietze A, et al. Fibrin-targeting molecular MRI in inflammatory CNS disorders. Eur J Nucl Med Mol Imaging. 2022. DOI: 10.1007/s00259-022-05807-8.

Messinger A, Sirmpilatze N, Heuer K, et al. A collaborative resource platform for non-human primate neuroimaging. NeuroImage. Elsevier. 2021;226:117519. DOI: 10.1016/j.neuroimage.2020.117519.hal-03167240.

Garcia-Saldivar P, Garimella A, Garza-Villarreal EA, et al. PREEMACS: Pipeline for preprocessing and extraction of the macaque brain surface. NeuroImage. 2021;227:117671. ISSN 1053-8119. DOI: 10.1016/j.neuroimage.2020.117671.

Puthillathu N, Moffett JR, Korotcov A, et al. A Brief Isoflurane Administration as an Adjunct Treatment to Control Organophosphate-Induced Convulsions and Neuropathology. Front Pharmacol. 2023. DOI: 10.3389/fphar.2023.1293280/full.

Ren BX, Huen I, Wu ZJ, et al. Early postnatal irradiation‐induced age‐dependent changes in adult mouse brain: MRI based characterization. BMC Neurosci. 2021;22:28. DOI: 10.1186/s12868-021-00635-2.