Hi FreeSurfer community:
I wanted to ask for general thoughts on spatial smoothing of data.
In particular, some searching of the archives reveals that a bit of smoothing can help the quality of the data, as well as the inter-subject registration of significance volumes.
FreeSurfer can perform both volume-based smoothing and surface-based smoothing. Theoretical concerns and some testing seem to indicate that smoothing along the surface yields more robust results and more consistent results on the data in practice-is this correct?
Secondly, I have read some of the general thoughts about smoothing (more localized region, less smoothing; more subjects, less smoothing). Does anybody have some more practical insight into appropriate smoothing kernels for anatomical data (e.g. thickness), and functional data (typically BOLD, typically recorded at 3 x 3 x 4mm)? I'm interested in subject pool sizes between 10 and 50.
Thanks very much in advance for all of the help and thoughts-
Joakim
Hi Joakim
yes, I believe your first point is true as you are implictly only smoothing within gray matter on the surface, as opposed to volume smoothing that includes CSF and even skull at larger kernel sizes.
W.r.t the optimal kernel size, that's a harder question. In general you smooth to (1) reduce noise, and (2) account for residual misregistration. The matched filter theorem says that your kernel should match the hypothesized size of the effect you are looking for, so there is no general right answer. The best kernel size is smaller if you are expecting focal effects, and bigger if they are more diffuse.
sorry, I wish there was a better answer than that.
cheers Bruce
On Thu, 9 Dec 2010, Joakim Vinberg wrote:
Hi FreeSurfer community:
I wanted to ask for general thoughts on spatial smoothing of data.
In particular, some searching of the archives reveals that a bit of smoothing can help the quality of the data, as well as the inter-subject registration of significance volumes.
FreeSurfer can perform both volume-based smoothing and surface-based smoothing. Theoretical concerns and some testing seem to indicate that smoothing along the surface yields more robust results and more consistent results on the data in practice-is this correct?
Secondly, I have read some of the general thoughts about smoothing (more localized region, less smoothing; more subjects, less smoothing). Does anybody have some more practical insight into appropriate smoothing kernels for anatomical data (e.g. thickness), and functional data (typically BOLD, typically recorded at 3 x 3 x 4mm)? I'm interested in subject pool sizes between 10 and 50.
Thanks very much in advance for all of the help and thoughts-
Joakim
Bruce,
Thanks for the excellent advice!
Two immediate follow-up questions come to mind:
I've read the analyses behind the matched filter theorem, and my follow-up question is if there would be any deleterious effects by using a fwhm below the size of a theorized ROI (as is often the case). I've actually done many previous analyses with no smoothing at all, so I am curious about using small smoothing kernels ("just enough" smoothing to eliminate some noise and allow for exploration of ROIs of various sizes). I have not found particular analyses to suggest this approach, and am curious if anyone else has any suggestions.
Secondly, it seems that both mris_preproc, and mri_surf2surf use surface smoothing by default, and I just wanted to make sure I have understood the Wiki correctly.
Thanks!
Joakim
-----Original Message----- From: Bruce Fischl [mailto:fischl@nmr.mgh.harvard.edu] Sent: Thursday, December 09, 2010 4:51 PM To: Joakim Vinberg Cc: freesurfer@nmr.mgh.harvard.edu Subject: Re: [Freesurfer] guidelines for spatial smoothing
Hi Joakim
yes, I believe your first point is true as you are implictly only smoothing within gray matter on the surface, as opposed to volume smoothing that includes CSF and even skull at larger kernel sizes.
W.r.t the optimal kernel size, that's a harder question. In general you smooth to (1) reduce noise, and (2) account for residual misregistration. The matched filter theorem says that your kernel should match the hypothesized size of the effect you are looking for, so there is no general right answer. The best kernel size is smaller if you are expecting focal
effects, and bigger if they are more diffuse.
sorry, I wish there was a better answer than that.
cheers Bruce
On Thu, 9 Dec 2010, Joakim Vinberg wrote:
Hi FreeSurfer community:
I wanted to ask for general thoughts on spatial smoothing of data.
In particular, some searching of the archives reveals that a bit of smoothing can help the quality of the data, as well as the
inter-subject
registration of significance volumes.
FreeSurfer can perform both volume-based smoothing and surface-based smoothing. Theoretical concerns and some testing seem to indicate that smoothing along the surface yields more robust results and more consistent results on the data in practice-is this correct?
Secondly, I have read some of the general thoughts about smoothing
(more
localized region, less smoothing; more subjects, less smoothing). Does anybody have some more practical insight into appropriate smoothing kernels for anatomical data (e.g. thickness), and functional data (typically BOLD, typically recorded at 3 x 3 x 4mm)? I'm interested in subject pool sizes between 10 and 50.
Thanks very much in advance for all of the help and thoughts-
Joakim
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.
Hi Joakim,
if the kernel is too small then you will have less noise reduction than you might with a bigger kernel, but of course if it's enough to detect the effect you're looking for, then you are fine.
And yes, those should use surface smoothing by default (Doug can correct me if I'm wrong)
cheers Bruce
On Thu, 9 Dec 2010, Joakim Vinberg wrote:
Bruce,
Thanks for the excellent advice!
Two immediate follow-up questions come to mind:
I've read the analyses behind the matched filter theorem, and my follow-up question is if there would be any deleterious effects by using a fwhm below the size of a theorized ROI (as is often the case). I've actually done many previous analyses with no smoothing at all, so I am curious about using small smoothing kernels ("just enough" smoothing to eliminate some noise and allow for exploration of ROIs of various sizes). I have not found particular analyses to suggest this approach, and am curious if anyone else has any suggestions.
Secondly, it seems that both mris_preproc, and mri_surf2surf use surface smoothing by default, and I just wanted to make sure I have understood the Wiki correctly.
Thanks!
Joakim
-----Original Message----- From: Bruce Fischl [mailto:fischl@nmr.mgh.harvard.edu] Sent: Thursday, December 09, 2010 4:51 PM To: Joakim Vinberg Cc: freesurfer@nmr.mgh.harvard.edu Subject: Re: [Freesurfer] guidelines for spatial smoothing
Hi Joakim
yes, I believe your first point is true as you are implictly only smoothing within gray matter on the surface, as opposed to volume smoothing that includes CSF and even skull at larger kernel sizes.
W.r.t the optimal kernel size, that's a harder question. In general you smooth to (1) reduce noise, and (2) account for residual misregistration. The matched filter theorem says that your kernel should match the hypothesized size of the effect you are looking for, so there is no general right answer. The best kernel size is smaller if you are expecting focal
effects, and bigger if they are more diffuse.
sorry, I wish there was a better answer than that.
cheers Bruce
On Thu, 9 Dec 2010, Joakim Vinberg wrote:
Hi FreeSurfer community:
I wanted to ask for general thoughts on spatial smoothing of data.
In particular, some searching of the archives reveals that a bit of smoothing can help the quality of the data, as well as the
inter-subject
registration of significance volumes.
FreeSurfer can perform both volume-based smoothing and surface-based smoothing. Theoretical concerns and some testing seem to indicate that smoothing along the surface yields more robust results and more consistent results on the data in practice-is this correct?
Secondly, I have read some of the general thoughts about smoothing
(more
localized region, less smoothing; more subjects, less smoothing). Does anybody have some more practical insight into appropriate smoothing kernels for anatomical data (e.g. thickness), and functional data (typically BOLD, typically recorded at 3 x 3 x 4mm)? I'm interested in subject pool sizes between 10 and 50.
Thanks very much in advance for all of the help and thoughts-
Joakim
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
-
Bruce Fischl -
Joakim Vinberg