recon-all of freesurfer

10 min read

recon-all is a batch program and runs >30 steps.  It easily takes 30 hours to finish one subject.

Use tkmedit CON14 T1.mgz and tksurfer CON14 lh inflated to visualize images. More about visualization with freesurfer.

Note: commands are in green. They are copied from the log files.

View images for every step here

Before you start, you need to put the structural images into certain directory hierarchy and set environment variable. Assume folder “structural” is where subjects’ structural images are. Under “structural”, you have folders “SUBJ1”, “SUBJ2”, “CON14” etc for every subject. You will have to put your structural images to “structural/SUBJ1/mri/orig/001.mgz”. If you have multiple images for this subject, use “002.mgz” etc. (If you original file format is not mgz, check out how to convert image formats).


Then set environment variable in linux shell:

  1. motion correction, and conform
    input: mri/orig/00?.mgz
    output: mri/orig.mgz (through mri/rawavg.mgz)
    If you have multiple structural images (001.mgz, 002.mgz, etc), then motion correction will be done and rawavg.mgz is created. If you have only one image (001.mgz), it simply copies the file and rename to rawavg.mgz.
    cp CON14/mri/orig/001.mgz CON14/mri/rawavg.mgz
    Then conform using mri_convert:
    mri_convert CON14/mri/rawavg.mgz CON14/mri/orig.mgz --conform
    Conform is to make voxel size 1x1x1mm and same number of voxels (256) along x,y,z directions (iso). Data is written to orig.mgz.
  2. non-uniform intensity normalization
    input: mri/orig.mgz
    output: mri/nu.mgz
    mri_nu_correct.mni --i orig.mgz --o nu.mgz --n 2
    Non-parametric Non-uniform intensity Normalization (N3), corrects for intensity non-uniformity in MR data, making relatively few assumptions about the data. This runs the MINC tool ‘nu_correct’. By default, four iterations of nu_correct are run. The flag -nuiterations specification of some other number of iterations.
  3. talairach transformation
    input: mri/nu.mgz
    output: transform/
    talairach_avi --i nu.mgz --xfm transforms/
    This computes the affine transform from the orig volume to the MNI305 atlas using the MINC program mritotal (see Collins, et al, 1994) through a FreeSurfer script called talairach. Several of the downstream programs use talairach coordinates as seed points. You can/should check how good the talairach registration is using tkregister2 --s subjid --fstal. tkregister2 allows you to compare the orig volume against the talairach volume resampled into the orig space. Run “tkregister2 –help” for more information. Creates the files mri/transform/ and talairach.xfm.
    This step produces a long log.
  4. intensity normalization
    input: mri/nu.mgz
    output: mri/T1.mgz
    mri_normalize -g 1 nu.mgz T1.mgz
    Performs intensity normalization of the orig volume and places the result in mri/T1.mgz. Attempts to correct for fluctuations in intensity that would otherwise make intensity-based segmentation much more difficult. Intensities for all voxels are scaled so that the mean intensity of the white matter is 110. If there are problems with the normalization, users can add control points. See also Normalization2.
  5. skull strip
    input: mri/nu.mgz
    output: mri/brainmask.mgz, mri/
    mri_em_register -skull nu.mgz /usr/local/freesurfer/average/RB_all_withskull_2008-03-26.gca transforms/talairach_with_skull.lta
    Removes the skull from mri/T1.mgz and stores the result in mri/ and mri/brainmask.mgz. Runs the mri_watershed program. If the strip fails, users can specify seed points (-wsseed) or change the threshold (-wsthresh, -wsmore, -wsless). The -autorecon1 stage ends here.
  6. EM (GCA) Registration (-<no>gcareg)
    input: mri/nu.mgz, mri/brainmask.mgz
    output: mri/transforms/talairach.lta
    mri_em_register -mask brainmask.mgz nu.mgz /usr/local/freesurfer/average/RB_all_2008-03-26.gca transforms/talairach.lta
    Computes transform to align the mri/nu.mgz volume to the default GCA atlas found in FREESURFER_HOME/average (see -gca flag for more info). Creates the file mri/transforms/talairach.lta. The -autorecon2 stage starts here.
  7. CA Normalize (-<no>canorm)
    input: mri/nu.mgz, mri/brainmask.mgz
    output: mri/norm.mgz

    mri_ca_normalize -mask brainmask.mgz nu.mgz /usr/local/freesurfer/average/RB_all_2008-03-26.gca transforms/talairach.lta norm.mgz
    Further normalization, based on GCA model. Creates mri/norm.mgz.

  8. CA Register (-<no>careg)
    input: mri/norm.mgz, mri/brainmask.mgz
    output: talairach.m3z
    mri_ca_register -align-after -nobigventricles -mask brainmask.mgz -T transforms/talairach.lta norm.mgz /usr/local/freesurfer/average/RB_all_2008-03-26.gca transforms/talairach.m3z
    Computes a nonlinear transform to align with GCA atlas. Creates the file mri/transform/talairach.m3z.
    This step is very loooooong (10 hours)
  9. Remove neck (-<no>rmneck)
    input: mri/nu.mgz
    output: mri/nu_noneck.mgz
    mri_remove_neck -radius 25 nu.mgz transforms/talairach.m3z /usr/local/freesurfer/average/RB_all_2008-03-26.gca nu_noneck.mgz
    The neck region is removed from the NU-corrected volume mri/nu.mgz. Makes use of transform computed from prior CA Register stage. Creates the file mri/nu_noneck.mgz.
  10. EM Registration, with Skull (-<no>skull-lta)
    input: mri/nu_noneck.mgz
    output: mri/transforms/talairach_with_skull.lta
    mri_em_register -skull -t transforms/talairach.lta nu_noneck.mgz /usr/local/freesurfer/average/RB_all_withskull_2008-03-26.gca transforms/talairach_with_skull.lta
    Computes transform to align volume mri/nu_noneck.mgz with GCA volume possessing the skull. Creates the file mri/transforms/talairach_with_skull.lta.
  11. CA Label (-<no>calabel)
    input: mri/norm.mgz, mri/transforms/talairach.m3z
    output: mri/, mri/aseg.mgz, (through mri/aseg.auto_noCCseg.mgz)
    mri_ca_label -align -nobigventricles norm.mgz transforms/talairach.m3z /usr/local/freesurfer/average/RB_all_2008-03-26.gca aseg.auto_noCCseg.mgz
    mri_cc -aseg aseg.auto_noCCseg.mgz -o CON14
    Labels subcortical structures, based in GCA model. Creates the files mri/ and mri/aseg.mgz.
  12. ASeg Stats (-<no>segstats)
    ?Computes statistics on the segmented subcortical structures found in mri/aseg.mgz. Writes output to file stats/aseg.stats.
  13. Normalization2 (-<no>normalization)
    input: mri/norm.mgz, mri/aseg.mgz, mri/brainmask.mgz
    output: mri/brain.mgz
    mri_normalize -aseg aseg.mgz -mask brainmask.mgz norm.mgz brain.mgz
    mri_mask -T 5 brain.mgz brainmask.mgz brain.finalsurfs.mgz
    Performs a second (major) intensity correction using only the brain volume as the input (so that it has to be done after the skull strip). Intensity normalization works better when the skull has been removed. Creates a new brain.mgz volume. The -autorecon2-cp stage begins here. If -noaseg flag is used, then aseg.mgz is not used by mri_normalize.
  14. WM Segmentation (-<no>segmentation)
    input: brain.mgz
    output: wm.seg.mgz
    mri_segment brain.mgz wm.seg.mgz
    mri_edit_wm_with_aseg -keep-in wm.seg.mgz brain.mgz aseg.mgz wm.asegedit.mgz
    mri_pretess wm.asegedit.mgz wm norm.mgz wm.mgz
    Attempts to separate white matter from everything else. The input is mri/brain.mgz, and the output is mri/wm.mgz. Uses intensity, neighborhood, and smoothness constraints. This is the volume that is edited when manually fixing defects. Calls mri_segment, mri_edit_wm_with_aseg, and mri_pretess. To keep previous edits, run with -keepwmedits. If -noaseg is used, them mri_edit_wm_aseg is skipped.
  15. Cut/Fill (-<no>fill)
    input: wm.mgz
    output: filled.mgz
    mri_fill -a ../scripts/ponscc.cut.log -xform transforms/talairach.lta -segmentation aseg.auto_noCCseg.mgz wm.mgz filled.mgz
    This creates the subcortical mass from which the orig surface is created. The mid brain is cut from the cerebrum, and the hemispheres are cut from each other. The left hemisphere is binarized to 255. The right hemisphere is binarized to 127. The input is mri/wm.mgz and the output is mri/filled.mgz. Calls mri_fill. If the cut fails, then seed points can be supplied (see -cc-crs, -pons-crs, -lh-crs, -rh-crs). The actual points used for the cutting planes in the corpus callosum and pons can be found in scripts/ponscc.cut.log. The stage -autorecon2-wm begins here. This is the last stage of volumetric processing. If -noaseg is used, then aseg.mgz is not used by mri_fill.
  16. Tessellation (-<no>tessellate)
    input: filled.mgz, norm.mgz
    output: surf/lh.orig.nofix (through filled-pretess255.mgz)
    mri_pretess ../mri/filled.mgz 255 ../mri/norm.mgz ../mri/filled-pretess255.mgz
    mri_tessellate ../mri/filled-pretess255.mgz 255 ../surf/lh.orig.nofix
    mris_extract_main_component ../surf/lh.orig.nofix ../surf/lh.orig.nofix
    This is the step where the orig surface (ie, surf/?h.orig.nofix) is created. The surface is created by covering the filled hemisphere with triangles. Runs mri_tessellate. The places where the points of the triangles meet are called vertices. Creates the file surf/?h.orig.nofix Note: the topology fixer will create the surface ?h.orig.
  17. Orig Surface Smoothing (-<no>smooth1, -<no>smooth2)
    input: surf/lh.orig.nofix
    output: surf/lh.smoothwm.nofix
    mris_smooth -nw ../surf/lh.orig.nofix ../surf/lh.smoothwm.nofix
    After tesselation, the orig surface is very jagged because each triangle is on the edge of a voxel face and so are at right angles to each other. The vertex positions are adjusted slightly here to reduce the angle. This is only necessary for the inflation processes. Creates surf/?h.smoothwm(.nofix). Calls mris_smooth. Smooth1 is the step just after tessellation, and smooth2 is the step just after topology fixing.
  18. Inflation (-<no>inflate1, -<no>inflate2)
    input: surf/lh.smoothwm.nofix
    output: surf/lh.inflated.nofix, lh.sulc, lh.curv, lh.area
    mris_inflate -no-save-sulc ../surf/lh.smoothwm.nofix ../surf/lh.inflated.nofix
    Inflation of the surf/?h.smoothwm(.nofix) surface to create surf/?h.inflated. The inflation attempts to minimize metric distortion so that distances and areas are perserved (ie, the surface is not stretched). In this sense, it is like inflating a paper bag and not a balloon. Inflate1 is the step just after tessellation, and inflate2 is the step just after topology fixing. Calls mris_inflate. Creates ?h.inflated, ?h.sulc, ?h.curv, and ?h.area.
  19. QSphere (-<no>qsphere)
    input: lh.inflated.nofix
    output: lh.qsphere.nofix
    mris_sphere -q ../surf/lh.inflated.nofix ../surf/lh.qsphere.nofix
    This is the initial step of automatic topology fixing. It is a quasi-homeomorphic spherical transformation of the inflated surface designed to localize topological defects for the subsequent automatic topology fixer. Calls mris_sphere. Creates surf/?h.qsphere.nofix.
  20. Automatic Topology Fixer (-<no>fix)
    input: qsphere.nofix
    output: lh.orig
    mris_fix_topology -mgz -sphere qsphere.nofix -ga CON14 lh
    mris_euler_number ../surf/lh.orig
    mris_remove_intersection ../surf/lh.orig ../surf/lh.orig
    Finds topological defects (ie, holes in a filled hemisphere) using surf/?h.qsphere.nofix, and changes the orig surface (surf/?h.orig.nofix) to remove the defects. Changes the number of vertices. All the defects will be removed, but the user should check the orig surface in the volume to make sure that it looks appropriate. Calls mris_fix_topology. Creates surf/?h.orig (by iteratively fixing surf/?h.orig.nofix).
  21. Final Surfaces (-<no>finalsurfs)
    input: lh.orig
    output: lh.white, lh.pial, lh.thickness, lh.curv
    mris_make_surfaces -noaparc -mgz -T1 brain.finalsurfs CON14 lh
    Creates the ?h.white and ?h.pial surfaces as well as the thickness file (?h.thickness) and curvature file (?h.curv). The white surface is created by “nudging” the orig surface so that it closely follows the white-gray intensity gradient as found in the T1 volume. The pial surface is created by expanding the white surface so that it closely follows the gray-CSF intensity gradient as found in the T1 volume. Calls mris_make_surfaces.
  22. [need to double check, b/c in the log this step is not here, also find command such as
    mris_calc -o lh.area.mid lh.area add lh.area.pial
    mris_calc -o lh.area.mid lh.area.mid div 2
    mris_calc -o lh.volume lh.area.mid mul lh.thickness]
    Cortical Ribbon Mask (-<no>cortribbon)
    Creates binary volume masks of the cortical ribbon, ie, each voxel is either a 1 or 0 depending upon whether it falls in the ribbon or not. Saved as ?h.ribbon.mgz. Uses mgz regardless of whether the -mgz option is used. The -autorecon2 stage ends here.
  23. Repeat step 17
    output: lh.smoothwm
    mris_smooth -n 3 -nw ../surf/lh.white ../surf/lh.smoothwm
  24. Repeat step 18
    input: lh.smoothwm
    output: lh.inflated
    mris_inflate ../surf/lh.smoothwm ../surf/lh.inflated
    mris_curvature -thresh .999 -n -a 5 -w -distances 10 10 ../surf/lh.inflated
  25. Spherical Inflation (-<no>sphere)
    input: lh.inflated
    output: lh.sphere
    mris_sphere ../surf/lh.inflated ../surf/lh.sphere
    Inflates the orig surface into a sphere while minimizing metric distortion. This step is necessary in order to register the surface to the spherical atlas. (also known as the spherical morph). Calls mris_sphere. Creates surf/?h.sphere. The -autorecon3 stage begins here.
  26. Ipsilateral Surface Registation (Spherical Morph) (-<no>surfreg)
    input: lh.sphere
    output: lh.sphere.reg
    mris_register -curv ../surf/lh.sphere /usr/local/freesurfer/average/lh.average.curvature.filled.buckner40.tif ../surf/lh.sphere.reg
    Registers the orig surface to the spherical atlas through surf/?h.sphere. The surfaces are first coarsely registered by aligning the large scale folding patterns found in ?h.sulc and then fine tuned using the small-scale patterns as in ?h.curv. Calls mris_register. Creates surf/?h.sphere.reg.
  27. [can’t find this step, but find command mris_jacobian ../surf/lh.white ../surf/lh.sphere.reg ../surf/lh.jacobian_white ]Contralateral Surface Registation (Spherical Morph) (-<no>contasurfreg)
    Same as ipsilateral but registers to the contralateral atlas. Creates lh.rh.sphere.reg and rh.lh.sphere.reg.
  28. Average Curvature (-<no>avgcurv)
    input: lh.sphere.reg
    output: lh.avg_curv
    mrisp_paint -a 5 /usr/local/freesurfer/average/lh.average.curvature.filled.buckner40.tif#6 ../surf/lh.sphere.reg ../surf/lh.avg_curv
    Resamples the average curvature from the atlas to that of the subject. Allows the user to display activity on the surface of an individual with the folding pattern (ie, anatomy) of a group. Calls mrisp_paint. Creates surf/?h.avg_curv.
  29. Cortical Parcellation (-<no>cortparc, -<no>cortparc2)
    input: mri/aseg.mgz, surf/lh.sphere.reg
    output: label/lh.aparc.annot
    mris_ca_label -aseg ../mri/aseg.mgz CON14 lh ../surf/lh.sphere.reg /usr/local/freesurfer/average/lh.curvature.buckner40.filled.desikan_killiany.2007-06-20.gcs ../label/lh.aparc.annot
    Assigns a neuroanatomical label to each location on the cortical surface. Incorporates both geometric information derived from the cortical model (sulcus and curvature), and neuroanatomical convention. Calls mris_ca_label. -cortparc creates label/?h.aparc.annot, and -cortparc2 creates /label/?h.aparc.a2005s.annot.
  30. Parcellation Statistics (-<no>parcstats)
    input: stats/lh.aparc.stats, label/lh.aparc.annot
    output: label/aparc.annot.ctab
    mris_anatomical_stats -mgz -f ../stats/lh.aparc.stats -b -a ../label/lh.aparc.annot -c ../label/aparc.annot.ctab CON14 lh
    Runs mris_anatomical_stats to create a summary table of cortical parcellation statistics for each structure, including 1. structure name 2. number of vertices 3. total surface area (mm^2) 4. total gray matter volume (mm^3) 5. average cortical thickness (mm) 6. standard error of cortical thicknessr (mm) 7. integrated rectified mean curvature 8. integrated rectified Gaussian curvature 9. folding index 10. intrinsic curvature index. For -parcstats, the file is saved in stats/?h.aparc.stats. For -parcstats2, the file is saved in stats/?h.aparc.a2005s.stats.
  31. Repeat step 29 (Cortical Parcellation)
    mris_ca_label -aseg ../mri/aseg.mgz CON14 lh ../surf/lh.sphere.reg /usr/local/freesurfer/average/lh.atlas2005_simple.gcs ../label/lh.aparc.a2005s.annot
  32. Repeat step 30 (Parcellation statistics)
    mris_anatomical_stats -mgz -f ../stats/lh.aparc.a2005s.stats -b -a ../label/lh.aparc.a2005s.annot -c ../label/aparc.annot.a2005s.ctab CON14 lh
  33. Repeat step 16-32 for right hemisphere
  34. ASeg Stats
    mri_segstats –seg mri/aseg.mgz –sum stats/aseg.stats –pv mri/norm.mgz –excludeid 0 –brain-vol-from-seg –brainmask mri/brainmask.mgz –in mri/norm.mgz –in-intensity-name norm –in-intensity-units MR –etiv –subject CON14 –surf-wm-vol –ctab /usr/local/freesurfer/ASegStatsLUT.txt
  35. Cortical ribbon mask
    mris_volmask –label_left_white 2 –label_left_ribbon 3 –label_right_white 41 –label_right_ribbon 42 –save_ribbon –save_distance CON14
  36. AParc-to-ASeg
    mri_aparc2aseg –s CON14 –volmask
    mri_aparc2aseg –s CON14 –volmask –a2005s
  37. WMParc
    mri_aparc2aseg –s CON14 –labelwm –hypo-as-wm –rip-unknown –volmask –o mri/wmparc.mgz –ctxseg aparc+aseg.mgz
  38. mri_segstats –seg mri/wmparc.mgz –sum stats/wmparc.stats –pv mri/norm.mgz –excludeid 0 –brain-vol-from-seg –brainmask mri/brainmask.mgz –in mri/norm.mgz –in-intensity-name norm –in-intensity-units MR –etiv –subject CON14 –surf-wm-vol –ctab /usr/local/freesurfer/

Check out recon-all manual in freesurfer wiki.


作者:北京师范大学 龙宇航,longyuhangwork@163.com代码来源(见本页底部):周思远 在使用wtc计算脑间神经同步后,我们需要在多个频率段、多个通道组合上对神经同步值进行统计检验,因此当进行频段选择时,面临多重比较的问题。为了解决多重比较的问题,可以采取基于参数或非参数检验的多重比较矫正的方法。由于基于非参数检验的多重比较矫正对数据的分布形态没有严格要求,因此具有更广泛的应用场景 (Maris and Oostenveld, 2007)。本文即介绍基于随机置换的非参数检验的方法 (Zheng et al., 2020; Long et al., 2021)。 在寻找感兴趣的效应时,我们采取了基于频率簇(Cluster)的方法,即在频率方向寻找连续显著的Cluster,该方法比基于最强效应点的方法具有更为优秀的抗噪音能力。值得注意的是,我们并没有沿着通道的方向去寻找连续显著的通道簇,这是因为沿着通道方向寻找Cluster容易受到生理噪音的影响。 下面进入具体的实操部分。假设本例招募了22对组1被试及22对组2被试,每对被试分别进行3种条件的任务,因此本例是2(组别,被试间因素)*3(条件,被试内因素)的实验设计。本例对神经同步值进行2*3的混合方差分析,并关注交互作用。 具体来讲,进行置换检验需要进行以下几个步骤:1. 重采样;2. 对随机样本进行计算及统计;3. 计算真实样本的统计量;4. 真实样本与随机样本的对比。下面依次进行介绍。 1. 重采样...
Xu Cui
1 min read

第二十期 fNIRS Journal Club 通知 2021/06/26,10:00am

同时用fNIRS和EEG测量脑信号有哪些好处?技术上应该注意什么?美国斯坦福大学李日辉博士,将为大家讲解他做过的一个同时用fNIRS和EEG测量的实验。热烈欢迎大家参与讨论。 时间: 北京时间2021年6月26日上午10:00地点: https://zoom.com房间号: 856 9352 0230密码: 695930 李博士要讲解的文章如下: Li, Li, Roh, Wang, Zhang (2020) Multimodal Neuroimaging Using Concurrent EEG/fNIRS for Poststroke Recovery Assessment:...
Xu Cui
7 sec read

Calculate phase difference between two general signals (e.g. HbO…

In a recent fNIRS journal club (vedio recorded here), Dr. Tong talked about their work on the phase difference between oxy and deoxy Hb,...
Xu Cui
1 min read

3 Replies to “recon-all of freesurfer”

  1. Hi,
    my name is Martina and I’m a student of biomedical engineering in Rome. I’m doing a brain imaging research for my final exam.
    Thank you for your useful info on your website.
    I’m using freesurfer for registration and segmentation of brain images instead of FSL tools (flirt,bet and fast) and then continue to process the data with FSL. But I have a problem: after converting the image generated with freesurfer mri_convert to “.nii” format it does not overlap with FSL’s reference MNI152.
    I know that freesurfer uses talairach reference. Is it true? May it be the problem?
    How can I superimpose properly the images?
    Do you know any workaround?
    Thank you very much.
    Martina Andellini

  2. Dear Xu,

    Thanks for the explanation, would it be possible to help me on the following?

    I want to get the segmented cortex from the Talairach space to the standard space and convert it to a mask so that I can use it to extract the cortex from other scans (T2 and T2*) which have already been co-registered to the original T1 image in its standard space.

    Many thanks

Leave a Reply to martina Cancel reply

Your email address will not be published. Required fields are marked *