Hi Simon
I'm also puzzled about your "quantitative T1 maps". Are these T1 values in ms? If so, the image won't have "T1-weighted contrast" as WM will be darker (shorter T1) than GM, in which case everything would fail. I assume that what you are going to FreeSurfer (and Avi's code) is instead the larger flip angle FLASH image.
cheers Bruce
On Thu, 12 Jul 2012, Snyder, Avi wrote:
I don't see that it is possible to avoid double resampling if you use FreeSurfer to segment. As far as I can tell, the problem with your dataset is that a substantial fraction of cases results a clearly poor Talarrach atlas transform, e.g., too "fat" or too "thin" or distorted. There is no provision, as far as I am aware, to, initialize the Talairach transform module based on a prior computations. There are two main possibilities that underly bad Talairach transforms (1) input images very displaced or rotated from standard acquisitions (2) mismatch between contrast mechanisms in the sample T1W and the atlas representative target. It is possible to correct error mechanism 2. A dialog regarding this possibility is ongoing with the FreeSurfer developers.
Avi
On 7/12/12 5:09 AM, "Baudrexel, Simon" Simon.Baudrexel@kgu.de wrote:
Dear Avi,
thank you for your reply, probably I didn't make my point clear enough. Our high resolution quantitative T1-maps (derived from multispectral FLASH) have excellent GM-WM contrast and once autorecon1 (talairach-avi) gets over this tricky coregistration into standard space all other following steps look very fine (including the subcortical segmentation which I'm interested in). Since registration is often more robust when images are approximately realigned from the beginning, I thought to use SPM (or fsl) affine coregistration parameters (to MNI space) as a guess in talairach_avi would be worth a try (I think the problem lies in a somewhat different contrast of the neck and skull which appear brighter in our qT1 maps compared to MPRAGE). I could use SPM-coregistered T1 maps instead an put them into the freesurfer pipeline which, however, would mean an additional interpolation (which is not what I want). Is there another workaround? ( I read that in tkregister2 there is the option to apply fsl transformation matrices - something like that would be needed - but I'm no too familiar with the way freesurfer handels and stores those transformations). As my doctorate fellow sees himself manually segmenting all the subcortical structurs any help would be greatly appreciated -:)
Thanks,
Simon Baudrexel Brain Imaging Center Goethe University Frankfurt ________________________________
Von: Snyder, Avi [avi@npg.wustl.edu] Gesendet: Donnerstag, 12. Juli 2012 01:58 Bis: Baudrexel, Simon; Bruce Fischl Cc: freesurfer@nmr.mgh.harvard.edu Betreff: Re: AW: [Freesurfer] Using SPM transformation matrices as initial guess in talairach_avi
Simon,
The standard Talairach atlas transform module in FreeSurfer was designed for T1W high-resolution structural images (e.g., MP-RAGE or SPGR). This is a design choice partly driven by the fact that the FreeSurfer segmentation algorithm assumes high-resolution T1 weighting. Reliable atlas transformation with any other contrast mechanism would be highly uncertain. Atlas transformation modules theoretically could be produced for other contrast mechanisms, e.g., R1 or FA. However, it is not clear that, even if this were done, that the tissue classification part of FreeSurfer would work. If you have R1 and/or FA images that need to be analyzed using FreeSurfer segmentations, I would think that you should submit a minimally processed high-resolution T1W image to FreeSurfer and then use the generated results as ROIs in subsequent analyses. Of course, this strategy requires that the images containing the contrast to be evaluated are co-registered to the segmented MP-RAGE.
I hope I have correctly understood your scientific objectives.
Avi Snyder
On 7/10/12 4:20 AM, "Baudrexel, Simon" Simon.Baudrexel@kgu.de wrote:
Hi Bruce,
I'm using freesurfer version 5.1.0. I currently try to process quantitative T1 (R1) maps derived from 4 and 18 deg. FLASH images and FA maps. Interestingly, transformation appears to fail less often when using single (16 deg.) FLASH images, only. SPM segmentation routine on the other hand works pretty well on both T1 and FLASH images (probably because of using header origin information - which was set to the anterior commissure- as first guess in the registration).
Thanks Simon ________________________________________ Von: Bruce Fischl [fischl@nmr.mgh.harvard.edu] Gesendet: Montag, 9. Juli 2012 20:03 Bis: Baudrexel, Simon Cc: freesurfer@nmr.mgh.harvard.edu; Avi Snyder Betreff: Re: [Freesurfer] Using SPM transformation matrices as initial guess in talairach_avi
Hi Simon
what version are you using? Avi Snyder (ccd) has been tracking down some interactions between his talairach code and our intensity normalization that seem to have made it more fragilie, particularly in 5.1. He has some suggested fixes that may improve things for you.
cheers Bruce
On Mon, 9 Jul 2012, Baudrexel, Simon wrote:
Hello,
since I experienced a high rate of registration errors using talairach_avi in the conventional recon-all processing stream (more than 50% misregistration), I would like to incorporate SPM affine registration parameters (as provided in the seg_inv_sn.mat file obtained from the segmentation routine) as initial guess "to help" talairach-avi. Is there a possibility to do this? How do I have to adjust the transformation matrix that it can be handled by freesurfer?
I also tried to circumvent the registration-problem using skull-stripped images, which unfortunately also did not improve the results (I think a template made from skull-stripped images has to be used in that case)
Any help would be much appreciated
Regards, Simon
Hi Avi and Bruce,
thanks for your replies. Indeed we use the inverse of the T1 values i.e. 'R1' maps (units in 1/ms although I scaled WM to 110; colloquially we use T1/R1 interchangebly, sorry). The contrast is comparable to freesurfer output T1.mgz and of course different from the larger flip angle FLASH. I will follow your both suggestions, i.e. try to process the FLASH and also the pre-realinged R1 maps in case freesurfer registration step fails.
Cheers Simon ________________________________________ Von: Bruce Fischl [fischl@nmr.mgh.harvard.edu] Gesendet: Donnerstag, 12. Juli 2012 22:16 Bis: Snyder, Avi Cc: Baudrexel, Simon; freesurfer@nmr.mgh.harvard.edu Betreff: Re: AW: AW: [Freesurfer] Using SPM transformation matrices as initial guess in talairach_avi
Hi Simon
I'm also puzzled about your "quantitative T1 maps". Are these T1 values in ms? If so, the image won't have "T1-weighted contrast" as WM will be darker (shorter T1) than GM, in which case everything would fail. I assume that what you are going to FreeSurfer (and Avi's code) is instead the larger flip angle FLASH image.
cheers Bruce
On Thu, 12 Jul 2012, Snyder, Avi wrote:
I don't see that it is possible to avoid double resampling if you use FreeSurfer to segment. As far as I can tell, the problem with your dataset is that a substantial fraction of cases results a clearly poor Talarrach atlas transform, e.g., too "fat" or too "thin" or distorted. There is no provision, as far as I am aware, to, initialize the Talairach transform module based on a prior computations. There are two main possibilities that underly bad Talairach transforms (1) input images very displaced or rotated from standard acquisitions (2) mismatch between contrast mechanisms in the sample T1W and the atlas representative target. It is possible to correct error mechanism 2. A dialog regarding this possibility is ongoing with the FreeSurfer developers.
Avi
On 7/12/12 5:09 AM, "Baudrexel, Simon" Simon.Baudrexel@kgu.de wrote:
Dear Avi,
thank you for your reply, probably I didn't make my point clear enough. Our high resolution quantitative T1-maps (derived from multispectral FLASH) have excellent GM-WM contrast and once autorecon1 (talairach-avi) gets over this tricky coregistration into standard space all other following steps look very fine (including the subcortical segmentation which I'm interested in). Since registration is often more robust when images are approximately realigned from the beginning, I thought to use SPM (or fsl) affine coregistration parameters (to MNI space) as a guess in talairach_avi would be worth a try (I think the problem lies in a somewhat different contrast of the neck and skull which appear brighter in our qT1 maps compared to MPRAGE). I could use SPM-coregistered T1 maps instead an put them into the freesurfer pipeline which, however, would mean an additional interpolation (which is not what I want). Is there another workaround? ( I read that in tkregister2 there is the option to apply fsl transformation matrices - something like that would be needed - but I'm no too familiar with the way freesurfer handels and stores those transformations). As my doctorate fellow sees himself manually segmenting all the subcortical structurs any help would be greatly appreciated -:)
Thanks,
Simon Baudrexel Brain Imaging Center Goethe University Frankfurt ________________________________
Von: Snyder, Avi [avi@npg.wustl.edu] Gesendet: Donnerstag, 12. Juli 2012 01:58 Bis: Baudrexel, Simon; Bruce Fischl Cc: freesurfer@nmr.mgh.harvard.edu Betreff: Re: AW: [Freesurfer] Using SPM transformation matrices as initial guess in talairach_avi
Simon,
The standard Talairach atlas transform module in FreeSurfer was designed for T1W high-resolution structural images (e.g., MP-RAGE or SPGR). This is a design choice partly driven by the fact that the FreeSurfer segmentation algorithm assumes high-resolution T1 weighting. Reliable atlas transformation with any other contrast mechanism would be highly uncertain. Atlas transformation modules theoretically could be produced for other contrast mechanisms, e.g., R1 or FA. However, it is not clear that, even if this were done, that the tissue classification part of FreeSurfer would work. If you have R1 and/or FA images that need to be analyzed using FreeSurfer segmentations, I would think that you should submit a minimally processed high-resolution T1W image to FreeSurfer and then use the generated results as ROIs in subsequent analyses. Of course, this strategy requires that the images containing the contrast to be evaluated are co-registered to the segmented MP-RAGE.
I hope I have correctly understood your scientific objectives.
Avi Snyder
On 7/10/12 4:20 AM, "Baudrexel, Simon" Simon.Baudrexel@kgu.de wrote:
Hi Bruce,
I'm using freesurfer version 5.1.0. I currently try to process quantitative T1 (R1) maps derived from 4 and 18 deg. FLASH images and FA maps. Interestingly, transformation appears to fail less often when using single (16 deg.) FLASH images, only. SPM segmentation routine on the other hand works pretty well on both T1 and FLASH images (probably because of using header origin information - which was set to the anterior commissure- as first guess in the registration).
Thanks Simon ________________________________________ Von: Bruce Fischl [fischl@nmr.mgh.harvard.edu] Gesendet: Montag, 9. Juli 2012 20:03 Bis: Baudrexel, Simon Cc: freesurfer@nmr.mgh.harvard.edu; Avi Snyder Betreff: Re: [Freesurfer] Using SPM transformation matrices as initial guess in talairach_avi
Hi Simon
what version are you using? Avi Snyder (ccd) has been tracking down some interactions between his talairach code and our intensity normalization that seem to have made it more fragilie, particularly in 5.1. He has some suggested fixes that may improve things for you.
cheers Bruce
On Mon, 9 Jul 2012, Baudrexel, Simon wrote:
Hello,
since I experienced a high rate of registration errors using talairach_avi in the conventional recon-all processing stream (more than 50% misregistration), I would like to incorporate SPM affine registration parameters (as provided in the seg_inv_sn.mat file obtained from the segmentation routine) as initial guess "to help" talairach-avi. Is there a possibility to do this? How do I have to adjust the transformation matrix that it can be handled by freesurfer?
I also tried to circumvent the registration-problem using skull-stripped images, which unfortunately also did not improve the results (I think a template made from skull-stripped images has to be used in that case)
Any help would be much appreciated
Regards, Simon
The information in this e-mail is intended only for the person to whom it is addressed. If you believe this e-mail was sent to you in error and the e-mail contains patient information, please contact the Partners Compliance HelpLine at http://www.partners.org/complianceline . If the e-mail was sent to you in error but does not contain patient information, please contact the sender and properly dispose of the e-mail.
freesurfer@nmr.mgh.harvard.edu
-
Baudrexel, Simon -
Bruce Fischl