auto_acpc_reorient : Cross-platform automatic AC-PC realignment/reorientation and coregistration robust to brain damage using SPM12
Cross-platform automatic AC-PC realignment/reorientation and coregistration for both healthy volunteers and brain damaged patients using template matching in SPM 12.
This is a set of routines to perform automatic reorient and coregistration with the toolbox Statistical Parametric Mapping 12 (SPM12).
Setting up the AC-PC and reorienting images is a recurrent issue in between-subjects group analyses, since they rely on coregistration methods that, like the "unified segmentation" of SPM12, are for most sensitive to initial conditions (the starting orientation of the image).
The main function, auto_acpc_reorient.m, automatically (but approximately) calculates a reorientation transform onto a target template in MNI space, in two steps:
- a non-linear coregistration of the input image onto a target template in MNI space is calculated using
spm_affreg, - then another transform is calculated using Mutual Information on a joint histogram (spm_coreg), and then applies only the rigid-body transform part of both coregistrations to reorient the input image. This allows to mainly set the origin on the AC and correct for head rotation, in order to further proceed with the segmentation/normalisation of the image.
This whole reorientation scheme relies on the "template matching" principle (as in the old normalize function), you therefore need to specify the appropriate template/reference image (we provide one, t1group, by default).
In any case, it is advised to check the automatically reoriented images afterwards, and fix the orientation manually, using SPM -> Display if necessary.
Another function, auto_acpc_coreg.m, expands on the same ideas to allow coregistration between modalities (eg, functional BOLD on structural MPRAGE). It is advised that auto_acpc_reorient() to be first applied on the structural before applying auto_acpc_coreg() on the other modality (even if you do manually fix the reorientation, as this ensures that the T1 is somewhat in the MNI space, making it easier for auto_acpc_coreg() to find the correct translation matrix).
These scripts are built on top of the original auto reorientation routines by John Ashburner & Carlton Chu.
To install this tool :
- Add
spm12andspm12\toolbox\OldNormto the path in MATLAB. - Simply add this (auto_acpc_reorient) folder to the path in MATLAB too (or
cdinside before typing the commands).
This will allow the commands auto_acpc_reorient() and auto_acpc_coreg() to be called from command-line (if no argument is given, a file selector dialog will open).
Type help auto_acpc_reorient, for all the details and various options for reorientation of a T1.
Type help auto_acpc_coreg for coregistering options.
Both scripts allows to use SPM filedialogs GUI, by simply typing auto_acpc_reorient or auto_acpc_coreg in the MATLAB prompt.
Note: by default, the 't1group' template will be used, which will use T1_template_CAT12_rm_withskull.nii. This is a template generated on 10 subjects using CAT12 that were manually reoriented to AC-PC and averaged, this provides better performance for reorientation than the more blurry MNI template. Note that this template is slightly better aligned to the AC-PC plane than the original MNI template, so that there may be a slight rotation bias compared to MNI if you use this custom template (usually it's mostly unnoticeable and this should have no influence if afterwards you do the SPM normalization on MNI on your data).
General guideline:
- If you want to reorient isotropic T1, use
auto_acpc_reorient. - If you want to reorient another modality (usually with less resolution), or an anisotropic T1, or an isotropic T1 but on your own custom T1 template, use
auto_acpc_coreg.
Note that the scripts can't be used from SPM12 GUI nor the BATCH system.
There is no guarantee that this will work 100% of the times, although it was observed to produce good results with our own data (young and old healthy subjects, AD/PD patients, most of brain damaged patients even with significant movement or metal artefacts).
The best results we got were by doing the following steps:
- Call
auto_acpc_reorient()on the structural in MNI space, so that it is also matching other SPM templates - Manually review and fix the misoriented structural images
- Coregister the functional to SPM12 EPI template (this allows a correct translation matrix and a good basis for rotation)
- Coregister the functional onto the structural (this fine-tunes rotation to precisely match the subject's structural)
The last 2 steps can be done by calling auto_acpc_coreg(), which has optimized default parameters for this task.
For indication, on a dataset of ~400 subjects with some heavily brain damaged or artifacted, the coregistration success rate was more than 95%. We have plans to conduct a more formal analysis but (as usual with open source) there is no guarantee it will happen in the near future :-)
For a comparison of various methods for AC-PC reorientation, the following article is a very good read:
Liu, Yuan, and Benoit M. Dawant. "Automatic detection of the anterior and posterior commissures on MRI scans using regression forests." 2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE, 2014.
Please cite this work as following (a paper is in preparation but not available yet):
auto_acpc_reorient, RRID:SCR_018308. https://github.com/lrq3000/auto_acpc_reorient
This software was developed by Stephen Karl Larroque (Coma Science Group, GIGA-Consciousness, University of Liege, Belgium).
The code of auto_acpc_reorient.m is a fork from auto_reorient.m written by John Ashburner (FIL, UCL, London, UK) and Carlton Chu (FIL, UCL, London, UK). Unfortunately, the original code was not released under a version control system, so the history in this repository does not include the history of the original commits. The history is hence a bit muddy, so if there is any inaccuracies, please let us know, we will happily update it.
We later found that K. Nemoto made a similar enhanced algorithm back in 2017, based on centering to the origin (instead of translating to match a template for us - what we call a pre-coregistration step) before applying the original auto_reorient.m script but tweaking it to apply a coregistration on a DARTEL template (instead of SPM12 ones in the original script, or a custom made template using CAT12 for this one). The resulting script, acpc_coreg.m, can be found on nemotos.net.
Another earlier fork of auto_reorient.m named spm_auto_reorient was done by Christophe Phillips (Cyclotron Research Centre, University of Liege, Belgium), which adds over the original supports for more templates, option to apply the reorientation matrix to other files, sanitization of input variables, integration in the SPM12 main GUI using cfg files and usage in the BATCH system. An earlier version of auto_acpc_reorient, named spm_auto_reorient_coregister, was forked from spm_auto_reorient, before a major rewrite to rebase on the original codebase.
The code is licensed under GPL (General Public License) v3 or later, except otherwise noted in comments around the code the other license pertains to (in particular the core of the affine reorientation code, which is licensed under CC-BY-SA v3.0 (notice the licensing at the bottom of the page) and authored by John Ashburner and Carlton Chu).
Adaptations can upgrade to later CC-BY-SA licenses as is permitted by the Creative Commons Foundation, as long as the original contributions are maintained under the original license.
This adaptation hence was upgraded to CC-BY-SA 4.0, which is one-way compatible to GPL v3. This upgrade of licensing is preferable as the Creative Commons Foundation stated that CC-BY-SA licenses are unfit to license software sourcecodes.
The templates are licensed under either GPL (General Public License) v3 or later, or CC0 (Public Domain) at your convenience.
If you have any feedback (questions, issues, suggestions), please contact Stephen Karl Larroque at: stephen dot larroque at uliege dot be.