Hi!
We are processing images using
recon-all -all -3T -mprage -qcache -hippocampal-subfields-T1 -brainstem-structures
We've encountered an issue when we're trying to use multiple inputs (acquired at the same timepoint and with comparable quality). The issue is due to the two images having slightly different voxel sizes:
ERROR: MultiRegistration::
loadMovables: images have different voxel sizes. Currently not supported, maybe first make conform? Debug info: size(1) = 1, 1, 1.2 size(0) = 1.05469, 1.05469, 1.2 MultiRegistration::loadMovables: voxel size is different /tmp/adni/freesurfer/6.0.0/0a9cfea9-7668-4a2c-927c-378ff5109fe1/mri/orig/002.mgz.
which we can see confirm from the DICOMs in the PixelSpacing tag.
So, we have three questions:
1. In general, should using multiple images jointly for processing yield better results? 2. If so, what would you suggest for our case with images of two different protocols? Any flags that could help? Or perform coregistration pre-run? 3. Let's say we have a cohort where all subjects have two T1-images but where some subjects have two images acquired with slightly different protocols (e.g. different voxel sizes) and jointly processing can't be done. Do you think it is better to combine images in as many cases we can (and use only a single image for the remaining cases) or only use one image per case for all cases? In other words, we are wondering if it could bias our data set if the subset of that has been multi-image processed is "better" than those processed with only a single image.
We're running on FS6 (freesurfer-Linux-centos6_x86_ 64-stable-pub-v6.0.0-2beb96c) and I've attached a log file from one of the failing cases.
Thank you for your help!
Best regards,
-- Gustav Mårtensson | PhD student Division of Clinical Geriatrics Department of Neurobiology, Care Sciences and Society Karolinska Institutet ___________________________________ Karolinska Institutet – a medical university
Dear Gustav,
1. If they have been acquired in the same session with the same protocol, you will increase your SNR by the square root of the number of volumes you use. This can potentially also help in case of minor motions artifacts. However, if your SNR is already high, it's probably better to stick to the better single volume, than to use an averaged volume.
2. I think the general recommendation is to not mix protocols. If you want to use both scans you can conform them prior to using them as an input for recon-all, e.g. mri_convert -c <input volume> <output volume>.
3. If all images were acquired with the same protocol I would use the better volume and omit the other. Otherwise, I would stick to the volumes acquired with the same protocol.
The effect of intrasession averaging has for example been investigated here: https://www.ncbi.nlm.nih.gov/pubmed/23668971
"It can be seen how for several structures averaging does not change the relative power to the cross-sectional analysis (hippocampus, putamen, thalamus), for a few structures averaging increases errors (amygdala, right hemisphere entorhinal and pallidum) and for a few other structures averaging reduces errors (right hemisphere caudate, left hemisphere entorhinal)."
and
"In agreement with two multi-site 1.5 T reproducibility studies, one focused on cortical thickness reproducibility (Han et al., 2006) and one focused on subcortical, ventricular and intracranial volume reproducibility (Jovicich et al., 2009), we found that averaging two MPRAGE acquisitions acquired within a session made relatively minor contributions to improvement in the across-session reproducibility. The acquisition of two MPRAGE volumes is still recommended mainly for practical reasons: if one volume is bad (e.g. due to motion artifacts) then the other can still be used for segmentation without averaging."
Best, Falk
Von: freesurfer-bounces@nmr.mgh.harvard.edu [mailto:freesurfer-bounces@nmr.mgh.harvard.edu] Im Auftrag von Gustav Mårtensson Gesendet: Donnerstag, 17. August 2017 22:43 An: Freesurfer support list Betreff: [Freesurfer] recon-all with multiple images with different voxel sizes
Hi!
We are processing images using
recon-all -all -3T -mprage -qcache -hippocampal-subfields-T1 -brainstem-structures
We've encountered an issue when we're trying to use multiple inputs (acquired at the same timepoint and with comparable quality). The issue is due to the two images having slightly different voxel sizes:
ERROR: MultiRegistration::
loadMovables: images have different voxel sizes. Currently not supported, maybe first make conform? Debug info: size(1) = 1, 1, 1.2 size(0) = 1.05469, 1.05469, 1.2 MultiRegistration::loadMovables: voxel size is different /tmp/adni/freesurfer/6.0.0/0a9cfea9-7668-4a2c-927c-378ff5109fe1/mri/orig/002.mgz. which we can see confirm from the DICOMs in the PixelSpacing tag.
So, we have three questions:
1. In general, should using multiple images jointly for processing yield better results? 2. If so, what would you suggest for our case with images of two different protocols? Any flags that could help? Or perform coregistration pre-run? 3. Let's say we have a cohort where all subjects have two T1-images but where some subjects have two images acquired with slightly different protocols (e.g. different voxel sizes) and jointly processing can't be done. Do you think it is better to combine images in as many cases we can (and use only a single image for the remaining cases) or only use one image per case for all cases? In other words, we are wondering if it could bias our data set if the subset of that has been multi-image processed is "better" than those processed with only a single image.
We're running on FS6 (freesurfer-Linux-centos6_x86_ 64-stable-pub-v6.0.0-2beb96c) and I've attached a log file from one of the failing cases.
Thank you for your help!
Best regards,
-- Gustav Mårtensson | PhD student Division of Clinical Geriatrics Department of Neurobiology, Care Sciences and Society Karolinska Institutet ___________________________________ Karolinska Institutet - a medical university
freesurfer@nmr.mgh.harvard.edu