Differences between revisions 11 and 12
Deletions are marked like this. Additions are marked like this.
Line 16: Line 16:
A: It depends on your processor speed and machine performance (notice that dual core or hyperthead won't speed up one analysis significantly), give a look at the link below in the section "Step-wise directives": http://surfer.nmr.mgh.harvard.edu/fswiki/recon-all A: It depends on your processor speed and machine performance (notice that dual/quad/hex core or hyperthead won't speed up one analysis significantly), give a look at the link below in the section "Step-wise directives": http://surfer.nmr.mgh.harvard.edu/fswiki/recon-all
Line 21: Line 21:
A: FreeSurfer run its process in a non-parallel environment, so you won't have benefit from a dual core machine for a single case analysis. However if you have many cases you can start two FreeSurfer recon-all process in the same machine and theoretically you can reduce by half the time to analyze your group of cases. A similar procedure can also be used in quad-core environment. Note that to take benefit of a multi-core environment you need to use a SMP kernel in your OS. A: FreeSurfer run its process in a non-parallel environment, so you won't have benefit from a dual/quad/hex core machine for a single case analysis. However if you have many cases you can start two FreeSurfer recon-all process in the same machine and theoretically you can reduce by half the time to analyze your group of cases. A similar procedure can also be used in quad/hex-core environment. Note that to take benefit of a multi-core environment you need to use a SMP kernel in your OS.
Line 28: Line 28:
Use a Quad-Core processor Intel Q6600 Use a Quad-Core processor Intel Q6600 (a i7 processor or Xeon 7400 series are better)
Line 34: Line 34:
AMD Opteron Quad-Core Barcelona Model 2350 or Model 2347 AMD Opteron Quad-Core or Phenon X4
Line 65: Line 65:
=== How to run recon-all in a ssh terminal ===
A: Run
  '''recon-all -s subjid -all &''' <<BR>>
  '''disown -a'''

 

top


  • FAQ


  • The purpose of this FAQ is to provide a Wiki space where users can add the most frequent questions to avoid asking already answered questions in the support list.

General

How can I help this FAQ?

A: If you are able to edit pages go ahead! If you don't have permission to write on the Wiki send an e-mail to ppj at netfilter dot com dot br

How long does it take to finish a reconstruction?

A: It depends on your processor speed and machine performance (notice that dual/quad/hex core or hyperthead won't speed up one analysis significantly), give a look at the link below in the section "Step-wise directives": http://surfer.nmr.mgh.harvard.edu/fswiki/recon-all

Also you can contribute to our running time statistics: https://surfer.nmr.mgh.harvard.edu/fswiki/ReconAllRunTimes

How can I reduce the time of recon-all in a group of patients?

A: FreeSurfer run its process in a non-parallel environment, so you won't have benefit from a dual/quad/hex core machine for a single case analysis. However if you have many cases you can start two FreeSurfer recon-all process in the same machine and theoretically you can reduce by half the time to analyze your group of cases. A similar procedure can also be used in quad/hex-core environment. Note that to take benefit of a multi-core environment you need to use a SMP kernel in your OS.

I want to use freesurfer to quantity cortical thickness, and I want to use the software under Linux. What equipment for computer do you recommend?

This answer requires constant updates:

For Best Performances:

Use a Quad-Core processor Intel Q6600 (a i7 processor or Xeon 7400 series are better)

Install at Least 4GB of Memory

If you prefer AMD you can use:

AMD Opteron Quad-Core or Phenon X4

Other alternative much more expensive is: http://www.apple.com/macpro/ with 8 cores.

The number of cores is roughly the number of studies you can process simultaneously. Notice that each process will take 20-24 hrs.

Keep in mind that you need to use the fastest memory in order to achieve maximum benefit from multi-core architecture.

Is it possible to run FreeSurfer in Ubuntu Linux?

A: Yes. Ubuntu Linux is basically a Debian distro, so you should use FreeSurfer RH9 version. Depending on your video card you should disable the DRI using the option NoDRI in the Device section of your X configuration file. Notice that in older Ubuntu version there's a bug that prevents NoDRI from working.

What are the advantages of FreeSurfer over VBM?

A:

  1. FS uses geometry to do inter-subject registration, which experience has shown results in a much better matching of homologous cortical regions than volumetric techniques.
  2. FS allows you to look at the two components of volume separately (thickness and surface area). It has been found that these two do not necessarily track one another, and in the worst case where one is increasing and the other decreasing the volume change can be 0.
  3. The target that FS uses for registration (the white matter surface geometry) is completely invariant to gm atrophy, so gm changes won't result in a different registration.

Reconstruction

Q: It seems that there is a 5mm thickness upper limit in my volumes, is it normal? How can I change this?

A: We did this to prevent noncortical regions such as the basal ganglia from corrupting the thickness measure through averaging. With the ?h.cortex.label it is probably no longer needed, however you can use mris_thickness -max <max thick> to generate a thickness with a different max. Take a look in the example below:

mris_thickness -max 10 bert lh newlh.thickness

The file newlh.thickness will be created inside the surf directory of your subject. But we don't think the true thickness is ever that much except in pathological cases like dysplasia and other disorders of cortical development.

Why in many subjects the insular cortical surface seems so thick? Is the convoluted nature of the Insula that causes that?

A: It's not the convoluted nature of the Insula. It's the fact that extreme capsule is so thin that it frequently isn't very apparent on MR, and so there appears to be continuous gray matter from the basal ganglia into the cortex. We think version 4.0 fixes this.

I am trying to measure the cortical thickness of specific ROI how can I do this?

A: You can save the ROI to a label file and then use:

  • mris_anatomical_stats -l <label file>

How to run recon-all in a ssh terminal

A: Run

  • recon-all -s subjid -all &
    disown -a

FreeSurfer File Meanings

What are the sulc and curv overlays (in QDEC) showing us?

A: The 'sulc' conveys information on how far removed a particular vertex point on a surface is from a hypothetical "mid-surface" that exists between the gyri and sulci. This surface is chosen so that the "mean" of all these displacements is zero. The 'sulc' gives a indication then of linear distance and displacements: how "deep" and how "high" are brain folds

In FreeSurfer, gyri have negative 'sulc' values, are colored green, and indicate how far "down" a point has to travel to reach this "mid-surface". Sulci have positive 'sulc' values, are colored red, and indicate how far "up" a point needs to travel to reach the mid-surface.

The 'curv' conveys information on the curvature (not distance) at a specific vertex point. The color conveys the sign, and is just an arbitrary choice. The sharper the curve, the higher the value (positive or negative). Areas with positive curvature, are colored red, and correspond to curvatures in sulci, i.e. curving "up". Areas with negative curvature are colored green, and correspond to curves pointing "down", i.e. gyri.

So, in a nutshell, the difference is that the 'curv' files contain information about curvatures, and the 'sulc' files contain information about displacement.

FreeSurfer & Matlab

How can I make an histogram of cortical thickness?

A: In Matlab you can use the read_curv() function, as in the example below.

  • thick = read_curv ('/usr/local/freesurfer/subjects/bert/surfer/lh.thickness');
    hist (thick,100);


    Notice that there are many zero values that refer to areas where there's no cortical surface. The best procedure is to create a new array without the zero values and then make the histogram.

QDEC

I am using Qdec to examine the anatomical differences between two groups of subjects. The surface-based measures I can select are thickness, area, area.pial, sulc, curv, and jacobian_white. Could anybody tell me what anatomical features the later three (sulc, curv, and jacobian_white) actually measure?


sulc = "average convexity" from our 1999 reconII paper(https://surfer.nmr.mgh.harvard.edu/ftp/articles/fischl99b-recon2.pdf).
Essentially measures the depth/height of each point above the average surface.

curv = smoothed mean curvature.
jacobian_white = the jacobian of the spherical transform. Measures the amount of distortion needed to warp a subject into register with the atlas.
You might want to look at the slides downloadable from the top of this page:

http://surfer.nmr.mgh.harvard.edu/fswiki/FsTutorial
which contain pictures showing the meaning of 'curv' and 'sulc'.

UserContributions/FAQ (last edited 2019-12-14 14:15:45 by DougGreve)