Re: thickness analysis display
Hi Bruce, thanks for the explanation of the p value display. I understand now and can see what is happening using the matlab commands: t = linspace(-4,4,1000); % t-test values p = tpdf(t,20); % corresponding p values with df=20 (requires stats toolbox) plot(t,p); hold on plot(t,-log10(p),'r') This works for the p values. What do you use to display the t values output using the --t option of mri_surfglm? I'm trying to use these to determine whether the p values indicate greater or lesser thickness between groups. In my case, the 2-tailed t-critical values for df=15 can be obtained with the matlab stats toolbox: alpha = [.05 .01 .001] -tinv(alpha/2,15) ans = 2.1314 2.9467 4.0728 So might this give the following values, given that fslope = (2.9467 + 1) / 4.0728: fthresh = 2.1314 fmid= 2.9467 fslope = 0.9690 I'm using the heat scale (stat positive) to display this and it has both blue and red regions. Does tksurfer automatically assume a 2-tailed distribution and when entering the values above, display for +/- fthresh/fmid? When displaying the results of a thickness analysis, what do you do to correct for multiple comparisons? Is there a sound argument for resel correction? Thanks in advance, Darren - Original Message - From: Bruce Fischl [EMAIL PROTECTED] To: [EMAIL PROTECTED] Sent: Thursday, October 31, 2002 2:52 AM Subject: Re: thickness analysis display Hi Darren, the -log10(p) transform simply means you're looking at the exponent, with increasing values meaning more significant. For example, if your threshold were set to: fthresh = 1 fmid = 2 fslope = 1 nothing with p1e-1 would be colored, the mid point of the color scale would be p=1e-2 and the full saturation would be fmid+1/fslope=1e-3. If fslope is too confusing (which it is to most people) you can set fmax instead. Note that in your example you should be using log10 not log: -log10(0.05) ans = 1.3010 so setting fthresh = 1.3 would give you an effective threshold of .05 (instead of .1 in the example above). cheers, Bruce Bruce Fischl email: [EMAIL PROTECTED] Mass. General Hosp. NMR Center.tel:(617)-726-4897 Rm. 2328, Building 149, 13th Street fax:(617)-726-7422 Charlestown, MA 02129 USA On Wed, 30 Oct 2002, Darren Weber wrote: Dear Doug et al, I've completed mris_surfglm on the cortical thickness, comparing two groups. For a linear regression, I would normally check the assumptions of linearity, homoscedasticity and normality in the residuals, but for 250,000 analyses, it's a daunting, if not impossible, task! So, lets assume the linear model applies. I need to confirm and clarify a few points on how to display/threshold the stats. Firstly, I followed the --help option examples and saved the output stats using: --sigt ./glm/group-sigt-rh.w paint I can now load these values as an overlay on the target subject surface of choice, great. Now the questions arise as to what quantity this is and how to use the overlay threshold GUI. Firstly, what quantity is it? According to the --help option: #Save the signficance (p-value) of the t-ratio of the contrast. The #value is actually the -log10 of the significance with the same #sign as the t-ratio from which it was computed. The significance #is computed from a double-sided t-test and is NOT corrected for #multiple comparisons across space. fmt is the format (see OUTPUT #FORMATS). So, the output values are not t or p, but -log10(p). Why is this so? I don't understand the advantage of doing this transform on the p-values. If so, and this should be simple, do we calculate a threshold of p = .05 as: p = .05; thresh = -log(p) thresh =2.9957 This would be a 2-tail threshold for any value of p. How would you obtain a 1-tailed threshold? Perhaps multiply p by 2, ie: thresh = -log(p*2) thresh =2.3026 OK, so we get a threshold value somehow. I'm not sure how to use the overlay threshold GUI, which has three slides and a slope value. The tksurfer manual does not describe the threshold controls, it only contains the following description: Overlay File Display Options To select which overlay to show, use the View-Overlay Layer submenu. The field names are automatically set to the file name loaded. You can change this name by typing a new one into the information area in the Tool window. The display for the current overlay can be configured in the View-Configure-Configure Overlay Display dialog. You can select the color scale to use with the radio buttons in the top area of the dialog. The Truncate option can be checked to turn off the display of negative values. Check the Reverse option to reverse the sign of the values as they are drawn in the color scale. View- Inverse and Complex are reserved for future
Re: mris_convert
Hi Bruce, thankyou, I had some minor bugs in the output asc file. It was in meters, but more importantly the vertices started at one, where FS vertices start at zero. Cheers, Darren - Original Message - From: Bruce Fischl [EMAIL PROTECTED] To: [EMAIL PROTECTED] Sent: Monday, October 21, 2002 11:42 PM Subject: Re: mris_convert Hi Darren, I think tkmedit/tksurfer can read .asc files natively, but if not mris_convert should be able to convert them to binary. cheers, Bruce Bruce Fischl email: [EMAIL PROTECTED] Mass. General Hosp. NMR Center.tel:(617)-726-4897 Rm. 2328, Building 149, 13th Street fax:(617)-726-7422 Charlestown, MA 02129 USA On Mon, 21 Oct 2002, Darren Weber wrote: Dear Bruce et al, I wonder if we can convert ascii to binary surf files? I have some scalp and skull surfaces that can be saved into an ascii file, is it possible to convert this to the freesurfer surf file format? I would like to display these surfaces using tkmedit to check their locations in the MRI volume, given that they are generated from a volume with the orientation and dims of the freesurfer volume. Take care, Darren -- Darren Weber, PhD Student Cognitive Neuroscience, School of Psychology Flinders University of SA, GPO Box 2100, Adelaide, SA 5001, Aust. Ph: (61 8) 8201 3889, Fax: (61 8) 8201 3877 http://203.3.164.46/~dlw/homepages/index.html
Re: mris_make_template etc.
G'day Bruce et al, the template is uploading to ftp://floyd.psy.flinders.edu.au/pub/weber/rh.wholebrain.tif (~5Mb, the permissions should allow anonymous ftp access). Could you take a look at it? I guess it's OK but I have no idea what it should look like, maybe it makes sense in the light of your experience and insight? If you notice any problems, please let me know. I have been able to view it with http://www.irfanview.com/ Thanks, Darren - Original Message - From: Bruce Fischl [EMAIL PROTECTED] To: [EMAIL PROTECTED] Sent: Saturday, October 19, 2002 1:18 AM Subject: Re: mris_make_template etc. Hi Darren, yes, I think what youu are doing is essentially correct. Don't worry about the .gcs or tissue_params files, they have nothing to do with the spherical morphing. The .tif file you create should be a normal multi-frame tif, with 9 frames in it, made up of 3 sets of 3 (mean,stderror,dof). These are statistics on the geometric properties of the group of subjects you gave to mris_make_template in a spherical parameterization (longitude and lattitude). You should be able to view them in anything that knows how to read multi-frame tif files I would think. cheers, Bruce p.s. Note that the dof frame has only 1 nonzero element (the 1st), which should be the # of subjects that make up the template. Bruce Fischl email: [EMAIL PROTECTED] Mass. General Hosp. NMR Center.tel:(617)-726-4897 Rm. 2328, Building 149, 13th Street fax:(617)-726-7422 Charlestown, MA 02129 USA On Fri, 18 Oct 2002, Darren Weber wrote: Dear Bruce et al, could you please confirm the following is reasonable? I am trying to register whole brain surfaces across subjects, hence creating a study specific template (if anyone is interested in the template, let me know). First, run the following to create an individual subject (C01) template: mris_make_template rh sphere c01 /usr/local/freesurfer/average/rh.wholebrain.tif Are any other arguments required? This command, as above, creates the output file only, no other files like those in the /usr/local/freesurfer/average/ area are created. What are the *.gcs or tissue_parms.txt files? The .tif is not a graphics file - what does tif stand for? The rh.wholebrain.tif is the same size as other template files, so that should be a good sign that it contains the required data. Then, after spherical transforms for all subjects, the rh.sphere surfaces are registered to this individual subject template, using a csh script containing: set avg=/usr/local/freesurfer/average/rh.wholebrain.tif mris_register -curv -w 0 ${sub}/surf/rh.sphere $avg ${sub}/surf/rh.sphere.reg What are the -curv and -w options and how do they affect the registration? That completes the first pass of the registration process. Then create a new wholebrain template, using all subjects' sphere.reg files: set sub=c01 c02 c03 c04 c05 c06 c07 c08 c09 c10 p02 p04 p05 p06 p07 p08 p09 mris_make_template rh sphere.reg $sub /usr/local/freesurfer/average/rh.wholebrain.tif This should create a template that is the average of all these subjects. Then the registration process is repeated to overwrite the sphere.reg files for each subject, using the new averaged template. Is this reasonable? If it has worked correctly, it should now be possible to use mris_surfglm on the rh.sphere.reg files. Thanks, Darren -- Darren Weber, PhD Student Cognitive Neuroscience, School of Psychology Flinders University of SA, GPO Box 2100, Adelaide, SA 5001, Aust. Ph: (61 8) 8201 3889, Fax: (61 8) 8201 3877 http://203.3.164.46/~dlw/homepages/index.html
Re: skull stripping
Hi Bob et al, I prefer mri_strip_skull to mri_watershed. I think the former was a previous method that has now been replaced by the latter. Please find attached a small csh script (process_anatomy.csh) that I find useful for doing the skull strip and segmentation (note that mri_watershed is hashed out). I use another script to loop over subjects (process_allanatomy.bash). Kind regards, Darren - Original Message - From: Bob Thoma [EMAIL PROTECTED] To: [EMAIL PROTECTED] Sent: Thursday, October 17, 2002 3:58 AM Subject: skull stripping All, I've been having some difficulty with skull stripping, it doesn't appear to be removing enough. I have tried changing the remove skull settings under expert preferences, but have had no luck. I found a previous email with a similar problem. Bruce, I believe you replied as such: are you using a talairaching procedure? If so, please check that it worked properly. If it fails, everything after it fails as well. Please also check your T1 volume by bringing it up in tkmedit. The vast majority of the white matter should have an intensity of exactly 110. If this is not the case,then the intensity normalization didn't work properly, possibly because of a talairach failure, or for some other reason. The segmentation itself usually works pretty well, so I suspect that it is one of the previous steps that is causing you problems. cheers, Bruce I have two questions, one, how can i check to see if the talairaching procedure worked properly or how to not run the talairaching procedure. Second, where do I locate the white matter intensity that you stated to be 110. Or if anyone has any other suggestions, I would be grateful. Thanks Natalie _ Protect your PC - get McAfee.com VirusScan Online http://clinic.mcafee.com/clinic/ibuy/campaign.asp?cid=3963 process_anatomy.csh Description: Binary data process_allanatomy.bash Description: Binary data
registration of whole cortical sheet
Dear Bruce et al, my query is about how to register cortical surfaces across X subjects, given that we have modified the usual left/right surface separation. Some time ago we discussed how to extract a complete cortical surface, rather than the left/right hemi surfaces. The purpose of this being to use a complete cortical surface for ERP/MEG source modelling methods. This involves a couple of modifications to the wm volume and the usual inputs to the fill white matter process (see attached shell script for details). In my case, the resulting surfaces are called rh.*, although they actually contain both left and right hemispheres, joined by the corpus callosum (not strictly cortex, but unavoidable in this approach). For source modelling, these surfaces are nice. Having obtained these surfaces for nearly 20 subjects, we've raised the possibility of doing a cortical thickness analysis for two groups, controls and PTSD patients. This requires that any vertex A is located in the same coordinate system and location for all subjects, so they must be coregistered (Fischl et al, 1999, Human Brain Mapping 8:272-284; http://www3.interscience.wiley.com/cgi-bin/issuetoc?ID=67501785). Normally, I assume the registration process morphs each left/right hemi cortex into two unit spheres (rather than morphing each hemi cortex onto the left/right hemisphere of one unit sphere). So, many of the examples in Fischl et al (1999) illustrate one hemisphere or another. Thus, we normally have lh.sphere and rh.sphere, two unit spheres, one for each hemi cortex surface. Further, the spherical sulcal patterns of each lh.sphere and rh.sphere are then further morphed (non-linear) to match the patterns of some average template. Obviously the whole cortical surfaces we have are not separated into lh/rh, rather the rh contains both. My concern here is with this template - I don't know much about it. I assume the template (average7) is specific for each left/right hemi-cortex. It might not be a good template for our data, which contains the whole cortex, including the sagittal fissure and corpus callosum. Can you recommend a template for registration of these subjects' surfaces? Is it reasonable to use one subject surface as a template for all others? If so, what do you think of a two stage process, first registering all subjects to subject X, then averaging across all subjects to create a study specific average, then reregister all subjects to this average? Many thanks for your consideration, Darren -- Darren Weber, PhD Student Cognitive Neuroscience, School of Psychology Flinders University of SA, GPO Box 2100, Adelaide, SA 5001, Aust. Ph: (61 8) 8201 3889, Fax: (61 8) 8201 3877 http://203.3.164.46/~dlw/homepages/index.html process_fillwm.csh Description: Binary data
Re: registration of whole cortical sheet
Thanks, Bruce. mris_make_template --help stops at 'valid options are:' :-( One of the input parameters is a surface name - I suppose 'orig', 'white', 'pial' etc are valid. I'm not sure what surface to specify; is it 'sphere' or 'inflated' or something else? Cheers, Darren - Original Message - From: Bruce Fischl [EMAIL PROTECTED] To: [EMAIL PROTECTED] Sent: Friday, October 11, 2002 12:12 PM Subject: Re: registration of whole cortical sheet Hi Darren, that's pretty much the way I generated the lh and rh templates we use. There's a binary named mris_make_template,which you probably have, that will take a bunch of surface models and construct a template (.tif) from them. You'll have to name the individual surface lh or rh, since it's expecting a hemisphere. The mris_register binary is actually smart enough to change it's default parameters if the target is from a single surface (it uses a more rigid morph). Good luck, Bruce On Fri, 11 Oct 2002, Darren Weber wrote: Dear Bruce et al, my query is about how to register cortical surfaces across X subjects, given that we have modified the usual left/right surface separation. Some time ago we discussed how to extract a complete cortical surface, rather than the left/right hemi surfaces. The purpose of this being to use a complete cortical surface for ERP/MEG source modelling methods. This involves a couple of modifications to the wm volume and the usual inputs to the fill white matter process (see attached shell script for details). In my case, the resulting surfaces are called rh.*, although they actually contain both left and right hemispheres, joined by the corpus callosum (not strictly cortex, but unavoidable in this approach). For source modelling, these surfaces are nice. Having obtained these surfaces for nearly 20 subjects, we've raised the possibility of doing a cortical thickness analysis for two groups, controls and PTSD patients. This requires that any vertex A is located in the same coordinate system and location for all subjects, so they must be coregistered (Fischl et al, 1999, Human Brain Mapping 8:272-284; http://www3.interscience.wiley.com/cgi-bin/issuetoc?ID=67501785). Normally, I assume the registration process morphs each left/right hemi cortex into two unit spheres (rather than morphing each hemi cortex onto the left/right hemisphere of one unit sphere). So, many of the examples in Fischl et al (1999) illustrate one hemisphere or another. Thus, we normally have lh.sphere and rh.sphere, two unit spheres, one for each hemi cortex surface. Further, the spherical sulcal patterns of each lh.sphere and rh.sphere are then further morphed (non-linear) to match the patterns of some average template. Obviously the whole cortical surfaces we have are not separated into lh/rh, rather the rh contains both. My concern here is with this template - I don't know much about it. I assume the template (average7) is specific for each left/right hemi-cortex. It might not be a good template for our data, which contains the whole cortex, including the sagittal fissure and corpus callosum. Can you recommend a template for registration of these subjects' surfaces? Is it reasonable to use one subject surface as a template for all others? If so, what do you think of a two stage process, first registering all subjects to subject X, then averaging across all subjects to create a study specific average, then reregister all subjects to this average? Many thanks for your consideration, Darren -- Darren Weber, PhD Student Cognitive Neuroscience, School of Psychology Flinders University of SA, GPO Box 2100, Adelaide, SA 5001, Aust. Ph: (61 8) 8201 3889, Fax: (61 8) 8201 3877 http://203.3.164.46/~dlw/homepages/index.html
Re: mris_surfglm
Hi Doug et al, from looking at the help on mris_surfglm it appears that vertices in the analyses are not as dense as those from the subject surfaces. I'm guessing that using the ico surface with 7 iterations of refinement (lots of vertices!), you then locate the nearest vertex in each subject for each ico vertex. So, the GLM analyses for any given ico vertex use data from different vertex numbers across subjects - are these vertex indices and ico vertex correspondence available as an output option or are they stored in a temp file or something? Say, for each vertex number of ico, a row vector of subject vertex indices, in the order they are input to mris_surfglm. Also, if the above assumption is correct, the implication may be that some slight variations in registration (alignment accuracy might vary to within 1 - 10 vertices, perhaps) could have an impact on the results? Is there any local vertex neighbourhood averaging of thickness values when extracting the individual subject values for the GLM? Cheers, Darren - Original Message - From: Doug Greve [EMAIL PROTECTED] To: [EMAIL PROTECTED] Sent: Friday, October 11, 2002 2:13 AM Subject: Re: mris_surfglm Hi Darren, the ico and average7 are about the same computationally. However, it is more convenient to use average7 it has a surface that you can directly render the results on. With ico, you'd still have to run mri_surf2surf to convert it to another surface. I'm not sure what you're trying to do in terms of the surface registration... doug Darren Weber wrote: Dear Doug et al, surfglm looks good. I have processed 10 controls and 8 patients such that each surface contains a whole cortical surface, without the usual split through the corpus callosum. How would I use the surface registration process? Should I use 'ico' as the average or can I use one of the subjects? Take care, Darren -- Darren Weber, PhD Student Cognitive Neuroscience, School of Psychology Flinders University of SA, GPO Box 2100, Adelaide, SA 5001, Aust. Ph: (61 8) 8201 3889, Fax: (61 8) 8201 3877 http://203.3.164.46/~dlw/homepages/index.html -- Douglas N. Greve, Ph.D. MGH-NMR Center [EMAIL PROTECTED] Phone Number: 617-724-2358 Fax: 617-726-7422
mris_surfglm
Dear Doug et al, surfglm looks good. I have processed 10 controls and 8 patients such that each surface contains a whole cortical surface, without the usual split through the corpus callosum. How would I use the surface registration process? Should I use 'ico' as the average or can I use one of the subjects? Take care, Darren -- Darren Weber, PhD Student Cognitive Neuroscience, School of Psychology Flinders University of SA, GPO Box 2100, Adelaide, SA 5001, Aust. Ph: (61 8) 8201 3889, Fax: (61 8) 8201 3877 http://203.3.164.46/~dlw/homepages/index.html
Re: RF bias correction with mri_convert
I am referring to a command line option to mri_convert: --unwarp-gradient-nonlinearity ... - Original Message - From: Bruce Fischl [EMAIL PROTECTED] To: [EMAIL PROTECTED] Sent: Thursday, August 29, 2002 9:30 PM Subject: Re: RF bias correction with mri_convert p.s. what we do is not really bias field estimation/correction, but something much more aggressive, really a pre-segmentation, which accounts for bias fields as well as tissue inhomogeneity. If what you want is really the bias field for some other purpose, you're probably better of using the EM stuff from the FSL (unless you can measure it directly with your scanner, which is the best solution!). Bruce On Thu, 29 Aug 2002, Stephen Smith wrote: Hi - no, it doesn't - that is part of the initial freesurfer processing (which estimates white matter and then estimates bias field on the basis of white matter points), not the format conversion. ttfn, Steve. On Thu, 29 Aug 2002, Darren Weber wrote: Dear Bruce et al, does mri_convert automatically apply RF bias field estimation/correction? If it can do the RF bias correction, what is the best command line option for this? Take care, Darren -- Darren Weber, PhD Student Cognitive Neuroscience, School of Psychology Flinders University of SA, GPO Box 2100, Adelaide, SA 5001, Aust. Ph: (61 8) 8201 3889, Fax: (61 8) 8201 3877 http://203.3.164.46/~dlw/homepages/index.html Stephen M. Smith Head of Image Analysis, FMRIB Oxford University Centre for Functional MRI of the Brain John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK +44 (0) 1865 222726 (fax 222717) [EMAIL PROTECTED] http://www.fmrib.ox.ac.uk/~steve
Re: RF bias correction with mri_convert
Hi Bruce, et al, The Markov random field approach with FSL FAST is intensive (~ 2000 sec on Athlon 1700, 1Gb PC2700 on VIA KT333A) and the results appear to be good. Of course, it could be better to measure the quantity on the scanner. Would this approach require access and modification of the kspace images before reconstruction? Naturally we could all rest easily if the scanner engineering and signal processing took care of the problem ;-) We don't have a research scanner and any access to any of that :-( It would be nice to evaluate this latter physical solution with that from FAST. Thanks for clarifying the mri_normalize process. It will be interesting, although overkill, to see what happens when FreeSurfer is fed one of the volumes after RF correction with FAST. In particular, whether it will improve the segmentation any. Another interesting experiment might be to segment the WM with freesurfer and use that volume as a prior input to FAST. Again, overkill, but anything that could improve the automation of the first segmentation stage might decrease the latter manual editing. The cost is the extra processing overheads. Yes, there is some other work to be done on just the RF bias corrected volume. Cheers, Darren - Original Message - From: Bruce Fischl [EMAIL PROTECTED] To: [EMAIL PROTECTED] Sent: Thursday, August 29, 2002 9:30 PM Subject: Re: RF bias correction with mri_convert p.s. what we do is not really bias field estimation/correction, but something much more aggressive, really a pre-segmentation, which accounts for bias fields as well as tissue inhomogeneity. If what you want is really the bias field for some other purpose, you're probably better of using the EM stuff from the FSL (unless you can measure it directly with your scanner, which is the best solution!). Bruce On Thu, 29 Aug 2002, Stephen Smith wrote: Hi - no, it doesn't - that is part of the initial freesurfer processing (which estimates white matter and then estimates bias field on the basis of white matter points), not the format conversion. ttfn, Steve. On Thu, 29 Aug 2002, Darren Weber wrote: Dear Bruce et al, does mri_convert automatically apply RF bias field estimation/correction? If it can do the RF bias correction, what is the best command line option for this? Take care, Darren -- Darren Weber, PhD Student Cognitive Neuroscience, School of Psychology Flinders University of SA, GPO Box 2100, Adelaide, SA 5001, Aust. Ph: (61 8) 8201 3889, Fax: (61 8) 8201 3877 http://203.3.164.46/~dlw/homepages/index.html Stephen M. Smith Head of Image Analysis, FMRIB Oxford University Centre for Functional MRI of the Brain John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK +44 (0) 1865 222726 (fax 222717) [EMAIL PROTECTED] http://www.fmrib.ox.ac.uk/~steve
RF bias correction with mri_convert
Dear Bruce et al, does mri_convert automatically apply RF bias field estimation/correction? If it can do the RF bias correction, what is the best command line option for this? Take care, Darren -- Darren Weber, PhD Student Cognitive Neuroscience, School of Psychology Flinders University of SA, GPO Box 2100, Adelaide, SA 5001, Aust. Ph: (61 8) 8201 3889, Fax: (61 8) 8201 3877 http://203.3.164.46/~dlw/homepages/index.html
eeg-toolbox release
Hi, there is a new release of the matlab eeg-toolbox/mri-toolbox available via http://eeg.sourceforge.net/. This release incorporates a new mesh_shrinkwrap function (alpha release) for generating a surface from an Analyze MRI volume. If given the Analyze version of the original input to freesurfer (eg, mri_convert -oid 1 0 0 -ojd 0 1 0 -okd 0 0 1 orig orig.asc), it will output a scalp in the same space as the freesurfer cortex asc file (created by freesurfer command, mris_convert). It is possible to use the FSL BET program to get a skull from volume from this orig volume, which can be tesselated also. Please see the mesh_3shell_script for an example of this process. This release also updates the mesh plot functions, to provide overlays and transparency. For recipients who would like to hear about or discuss releases of this toolbox, please subscribe to the eeg-users email list. This will be the last announcement for this toolbox that will be sent to related email lists. I sincerely apologise to regular users of these lists and their administrators for this 'off-topic' announcement, but I hope the distribution of these tools might benefit other users of scan, freesurfer, spm, fsl, etc tools. All communication related to this matlab toolbox will hereafter be conducted in [EMAIL PROTECTED] All the best with your work. Kind regards, Darren -- Darren Weber, PhD Student Cognitive Neuroscience, School of Psychology Flinders University of SA, GPO Box 2100, Adelaide, SA 5001, Aust. Ph: (61 8) 8201 3889, Fax: (61 8) 8201 3877 http://203.3.164.46/~dlw/homepages/index.html
test
apologies for any inconvenience some of my email to the list has not been working