Multimodal Integration and Inter-subject Registration surfer. nmr. mgh Multimodal Integration and Inter-subject Registration surfer.nmr.mgh.harvard.edu

Overview Intra-subject registration Affine transformations Registration, Automatic and Manual fMRI Integration fMRI Analysis Intro Registration Viewing on Volume and Surface ROI analyses Surface-based group analysis Inter-subject registration

Theory of Affine Spatial Transforms Anatomical (1x1x1.1mm, 256x256x128, Sag) Scanner Acquisition Functional (3x3x5mm, 64x64x30, Axial)

Theory of Affine Spatial Transforms Conformed Anatomical Space 1x1x1mm, 256x256x256, Cor Native Anatomical Space 1x1x1.1mm, 256x256x128, Sag “Anatomical Space” orig.mgz Surfaces Parcellations Segmentations

Theory of Affine Spatial Transforms “Anatomical Space” Native Functional Space 3x3x5mm, 64x64x30, Axial ??? Conformed Anatomical Space 1x1x1mm, 256x256x256, Cor

Multi-modal Intra-subject Registration FreeSurfer Anatomical (orig) Template Functional Note: Registering the template functional volume to the anatomical volume is sufficient to register the template to the surface.

FreeSurfer Registration and Template Volume FreeSurfer Subject-Specific Volumes Surfaces Thickness ROIs Template Volume fMRI DTI ASL PET … Registration Template Volume: In voxel-for-voxel registration with parameter map Best gray-white contrast

FreeSurfer Registration Matrix 4x4 Matrix As many as 12 DOF (usually 6) Simple Text file Coordinate system not easy to explain mgh-02407836-v2 3.437500 5.000000 0.150000 9.999985e-01 -1.428481e-03 -8.293565e-04 5.281680e-01 4.641568e-04 -2.388080e-01 9.710670e-01 -4.041043e+01 1.585159e-03 9.710652e-01 2.388064e-01 -1.376212e+00 0 0 0 1 round

FreeSurfer Registration Matrix 4x4 Matrix As many as 12 DOF (usually 6) Simple Text file Coordinate system not easy to explain mgh-02407836-v2 3.437500 5.000000 0.150000 9.999985e-01 -1.428481e-03 -8.293565e-04 5.281680e-01 4.641568e-04 -2.388080e-01 9.710670e-01 -4.041043e+01 1.585159e-03 9.710652e-01 2.388064e-01 -1.376212e+00 0 0 0 1 round  FreeSurfer subject name  Functional In-plane resolution mm  Functional Between-plane resolution mm  Intensity (for visualization)  Legacy

Automatic Intra-subject Registration bbregister \ -–subject bert \ –-mov mmtemplate.nii \ --bold \ --init-fsl \ –-reg register.dat  Command name  FreeSurfer subject name  Multimodal template volume  Multimodal contrast  Initialize with FSL-FLIRT  Output registration file BB = Boundary-based, about 5 min. Registers template to conformed anatomical of given subject (bert) Registration is initialized with FSL-FLIRT 6 DOF Initialization also with --init-spm and --init-header About 5 min

Manual Intra-subject Registration tkregister2 \ –-mov mmtemplate.nii \ –-reg register.dat \ --surfs  Command name  Multimodal template volume  registration file  Display white surf Note similarity to bbregister command Subject not needed (alread in register.dat file) tkregister2 --help

Manual Intra-subject Registration Visually inspect registration Manually edit registration (6 DOF) Cf Manual Talairach registration Green line is white surface tkregister2 --help

Tips Rigid = 6 DOF = No stretching Use CSF to get a sense of where the folds are Avoid using B0 distortion regions Avoid using ventricles Warning about “edge” of the brain Same Subject, Left-Right Flips

} Command-line Tools Automatic Registration: bbregister --help fslregister –help spmregister –help reg-feat2anat –help Manual Registration: tkregister2 --help Transformations: mri_vol2surf --help mri_vol2vol --help mri_label2vol --help mri_surf2vol --help } FreeSurfer Scripts

fMRI Integration Visualize individual fMRI results on surface volume ROI Volume Study: Count number of voxels above threshold in an anatomical ROI ROI Intensity Study: Average HRF inside of an ROI Surface-based fMRI group analysis

Hemodynamic Response (BOLD) Time-to-Peak (~6sec) Dispersion TR (~2sec) Equilibrium (~16-32sec) Undershoot Delay (~1-2sec)

Multiple Presentations/Averaging Individual Output: HRF Amp, HRF Var, p/z/t/F

fMRI Preprocessing Overview Motion Correction (MC Template) Use for registration template bbregister --mov mctemplate.nii --s subject --init-fsl --reg register.dat tkregister2 --mov mctemplate.nii --reg register.dat --surf Do not use nonlinear resampling to talairach/MNI space Do not spatially smooth (3d)

fMRI Analysis Overview First-Level (Individual) Analysis HRF Amplitude (or Contrast of Amplitudes) cope (FSL), CON (SPM), ces (FSFAST) Variance of Amplitude varcope (FSL), ??? (SPM), cesvar (FSFAST) Activation/Significance Maps: z t F sig (-log10(p)) All in alignment with MC Template!!!!

Template and Map Functional Template Template+ Map

Volume Viewing tkmedit subject orig.mgz –aux brain.mgz –aparc+aseg \ -overlay sig.nii –reg register.dat \ -fthresh 2 –fmax 4 sig.nii – significance map in native functional space. could have been z, t, or F map as well. register.dat – freesurfer registration file fthresh – lower threshold (value depends on map). You can change this in the interface. fmax – saturation threshold. (value depends on map). You can change this in the interface. aparc+aseg – display aparc+aseg.mgz. Can load this from the interface too.

Volume Viewing Red/Yellow + Blue/Cyan - Seg Opacity ROI Average ROI Count place holder

Sampling onto the Surface White/Gray Pial

Sampling onto the Surface White/Gray Pial White/Gray Pial Half Way Average Projection Fraction --projfrac 0.5

Sampling on the Surface

Sampling on the Surface: Projection Fraction -0.1 0.0 (white) +0.1 +0.3 +0.5 +0.7 +0.9 +1.0 (pial) +1.1

Surface Viewing Resample HRF Contrast Significance to left hemisphere mri_vol2surf \ --mov sig.nii \ --reg register.dat \ --hemi lh \ --projfrac 0.5 \ --o lh.sig.mgh  map in native functional space  freesurfer registration file  hemisphere  projection fraction (half)  output (Nvertices-x-1 mgh format) Note similarity to bbregister and tkregister2 commands. Load HRF Contrast Significance as overlay tksurfer subject lh inflated -aparc –overlay lh.sig.mgh

Surface Viewing Red/Yellow +,Blue/Cyan - Parcellation Outline ROI Average ROI Count

ROI fMRI Analyses: Intensity Volume Full Anatomical ROI Functionally Constrained ROI Volume

Average HRF within Full Anatomical ROI Eg, average functional HRF amplitudes from all voxels inside of post-central gyrus (red) regardless of whether a voxel is significant or not. Resampling puts ces into voxel-for-voxel registration with aseg.mgz

Average HRF within Full Anatomical ROI Resample HRF Contrast to anatomical space mri_vol2vol \ --mov ces.nii \ --reg register.dat \ --interp nearest \ --fstarg \ --o ces.anat.mgh  Command name  HRF map in functional space  FreeSurfer Registration File  Nearest neighbor interpolation  Specify anatomical output space  Output file in anatomical space Note similarity to bbregister, tkregister2, and mri_vol2surf commands. Average HRF Contrast within ROIs mri_segstats \ --seg $SUBJECTS_DIR/subject/mri/aseg.mgz \ --ctab $FREESURFER_HOME/FreeSurferColorLUT.txt \ --i ces.anat.mgh --sum ces.aseg.stats Resampling puts ces into voxel-for-voxel registration with aseg.mgz

Average HRF within Full Anatomical ROI Average HRF Contrast within ROIs mri_segstats \ --seg $SUBJECTS_DIR/subject/mri/aseg.mgz \ --ctab $FREESURFER_HOME/FreeSurferColorLUT.txt \ --i ces.anat.mgh --sum ces.aseg.stats Notes: --seg is the segmentation (eg, aseg.mgz, aparc+aseg.mgz, etc) --ctab is matching color lookup table Output File: ces.aseg.stats simple text file with same format aseg.stats multiple subjects can be combined with asegstats2table

Average HRF within a Functionally Active area inside of an Anatomical ROI Resampling puts ces into voxel-for-voxel registration with aseg.mgz Eg, average functional HRF amplitudes from voxels inside of post-central gyrus (red) for voxels that have p<.01 (sig>2) regardless of sign (yellow or blue), or p<.01 (sig>2) for positive activation (yellow only), or p<.01 (sig>2) for negative activation (blue only)

Average HRF only within the Functionally Active Area of an Anatomical ROI, Count Voxels above threshold in each ROI Resample HRF Contrast Significance to anatomical space mri_vol2vol \ --mov sig.nii \ --reg register.dat \ --interp nearest \ --fstarg \ --o sig.anat.mgh Average HRF Contrast within functionally constrained ROIs mri_segstats \ --seg $SUBJECTS_DIR/subject/mri/aseg.mgz \ --ctab $FREESURFER_HOME/FreeSurferColorLUT.txt \ --i ces.anat.mgh --sum ces.aseg.mask.stats \ --mask sig.anat.mgh --mask-thresh 2 --mask-sign abs

Average HRF only within the Functionally Active Area of an Anatomical ROI, Count Voxels above threshold in each ROI mri_segstats \ --seg $SUBJECTS_DIR/subject/mri/aseg.mgz \ --ctab $FREESURFER_HOME/FreeSurferColorLUT.txt \ --i ces.anat.mgh --sum ces.aseg.mask.stats \ --mask sig.anat.mgh --mask-thresh 2 --mask-sign abs Volume in stats file is volume above threshold (may be 0) Sign is important for Average! abs, pos, or neg pos will always result in positive HRF average neg will always result in negative HRF average abs ???? Careful to avoid circularity Can use a different contrast to mask

Surface-based Group Analysis mris_preproc \ --hemi lh \ --o lh.fsaverage.ces.mgh \ --iv subject1/ces.nii subject1func/register.dat \ --iv subject2/ces.nii subject2func/register.dat \ --iv subject3/ces.nii subject3func/register.dat \ ... After that, everything else is the same as a thickness study … mris_fwhm --i lh.fsaverage.ces.mgh --fwhm 10 \ --o lh.fsaverage.ces.sm10.mgh --cortex mri_glmfit --surf fsaverage lh –cortex \ --y lh.fsaverage.ces.sm10.mgh ...

Tutorial Registration – manual and automatic registration fMRI Integration (Sensorimotor Paradigm) Individual Volume view sig Surface view sig ROI analysis with and without functional constraint Group mris_preproc ROI analysis (asegstats2table)

Another type of problem: inter-subject uni-modal registration Surface-based (2D) registration does an excellent job of aligning cortical folds, but does not apply to non-cortical structures (e.g. basal ganglia). Volumetric (3D) registration applies to the entire brain but does not, in general, align folding patterns. Goal: combined their strength

Why aligning folds in the volume is hard… Bruce’s brain affine transformed and mapped on Doug’s Affine – does not provide the a good initialization; does not bring folds close enough for the basin of attraction Affine transform of surfaces from one subject mapped to another.

pial surface WM surface Template Affine Nonlinear Mention (red , yellow)

Combined volumetric and surface-based registration (CVS) Spherical alignment Elastic propagation of cortical registration results in the 3D volume Volumetric alignment of sub-cortical regions sub-cortical alignment using cortical correspondence as initialization

Resulting morph Template Elastic CVS

Template HAMMER FLIRT CVS Examples of results obtained in Inter-Subject Experiment 2. The first column displays the template segmentation and the second, third and fourth columns represent the deformed segmentation volumes resulting from FLIRT (affine), HAMMER and our new combined surface- and intensity-based registration. In all the images, the surfaces shown are those of the template (pial surfaces in red and gray/white surfaces in yellow).

Extended Jaccard Coefficient (23) measures computed for Experiment 1 Extended Jaccard Coefficient (23) measures computed for Experiment 1. The last column of the cortical measures represents the overlap measure computed in the surface space. The vertical lines represent the standard error of the mean of the measurement. Extended Jaccard Coefficient measures: 20 cortical and 21 sub-cortical labels. (The vertical lines represent the standard error of the mean of the measurement.) G.M. Postelnicu*, L . Zöllei*, B. Fischl: "Combined Volumetric and Surface Registration", IEEE Transactions on Medical Imaging (TMI), Vol 28 (4), April 2009, p. 508-522

mri_cvs_register --mov subjid registering the subject to, by default, the CVS atlas space make sure that the SUBJECTS_DIR for subjid is correctly set Optional Arguments -- template subjid : subjid for template subject -- templatedir dir : recon directory for template (default is SUBJECTS_DIR) --outdir dir : output directory for all the results (default is SUBJECTS_DIR/subjid/cvs) … and many more: use --help -I like how you have "moving" in italics. Will you cover how commands like vol2vol and flirt are usually asking for a mov, ref, template, etc.? Some of those terms are interchangeable and it can be confusing when different commands use different terms.

mri_cvs_register Optional Arguments (cont) --step1 Only do step 1 (spherical registration). --step2 Only do step 2 (elastic registration). --step3 Only do step 3 (volumetric registration). --noaseg Do not use aseg volumes in the volumetric registration pipeline (default is 0). Setting this option could shorten significantly the time of registration, however, might also take away from the accuracy of the final results.

mri_cvs_register Optional Arguments (cont) --nocleanup Do not delete temporary files (default is 0). --keepelreg Do not delete elastic registration (default is 0) outcome. --cleanall Recompute all CVS-related morphs that might have been computed prior to the current CVS run (def = 0). --cleansurfreg Recompute CVS-related surface registration morphs that might have been computed prior to the current CVS run (def = 0). --cleanelreg Overwrite /recompute the CVS-related elastic registration morph that might have been computed prior to the current CVS run (default is 0). --cleanvolreg Overwrite / recompute CVS-related volumetric morphs that might have been computed prior to the current CVS run (default is 0).

CVS atlas

related commands mri_cvs_check mri_cvs_data_copy mri_vol2vol checking whether all files needed for a successful CVS registration are present mri_cvs_data_copy copying the CVS-relevant recon directories over to a new location mri_vol2vol applying the CVS registration morph to files corresponding to the moving subject

Applying CVS morphs mri_vol2vol applying CVS morph to aseg file mri_vol2vol --targ templateid --m3z morph.m3z \ --noDefM3zPath --mov asegvol \ --o asegvol2CVS --interp nearest \ --no-save-reg applying morph to corresponding diffusion file mri_vol2vol --targ templateid --m3z morph.m3z --noDefM3zPath --reg 2anat.register.dat --mov diffvol --o diffvol2CVS paths $FREESURFER_HOME/subjects/cvs_avg35

Applying CVS morphs applyMorph: use with older style CVS morphs (.tm3d) applying CVS morph to aseg file applyMorph --template templateid --transform morph.tm3d\ vol asegvol asegvol2CVS nearest applying morph to corresponding diffusion file applyMorph --template templateid –transform morph.tm3d vol diffvol2anat diffvol2CVS linear paths $FREESURFER_HOME/subjects/cvs_avg35

Application of CVS to tractography Goal: fiber bundle alignment Study: compare CVS to methods directly aligning DWI-derived scalar volumes Conclusion: high accuracy cross-subject registration based on structural MRI images can provide improved alignment Zöllei, Stevens, Huber, Kakunoori, Fischl: “Improved Tractography Alignment Using Combined Volumetric and Surface Registration”,  NeuroImage 51 (2010), 206-213

Average tracts after registration mapped to the template displayed with iso-surfaces FLIRT FA-FNIRT CVS The probability maps were thresholded at 0.1

Mean Hausdorff distance measures for three fiber bundles CST ILF UNCINATE

Functional MRI analysis in CVS space Collaboration with Kami Koldewyn, Joshua Julien and Nancy Kanwisher at MIT

Ongoing development Improve CVS capability to register ex-vivo to in-vivo acquisitions Implemented MI-based volumetric registration (for CVS step 3) to accommodate intensity profile differences Qualitative preliminary results on 4 subjects L. Zöllei, Allison Stevens, Bruce Fischl: Non-linear Registration of Intra-subject Ex-vivo and In-vivo Brain Acquisitions, Human Brain Mapping, June 2010 L. Zöllei, B. Fischl, Automatic segmentation of ex-vivo MRI images using CVS in FreeSurfer, HBM 2011

Subject 1 Target (in-vivo) Masked target 2-step CVS CVS with MI

registration tool summary mris_register fslregister: bet + flirt bbregister mri_robust_register mri_cvs_register mri_nl_align Some require recons! -with the registration tool summary slide, will you cover the strengths and weaknesses of each or when it is most appropriate to use them? Like bbregister and cvs can only be used if you have surfaces, fsl takes into account X while robust_register takes into account Y so you would choose one over the other in such Z situations.

registration morph summary .dat, .lta, .xfm, .fslmat: encode rigid and affine transformations mri_vol2vol sphere.reg: encodes spherical morph mris_resample .m3z, .tm3d: encode nonlinear volumetric morphs mri_vol2vol, applyMorph Some require recons!

Summary Multi/Cross-modal map (HRF Amplitude, FA) Multimodal Integration requires a Template A Template is: Same size as multimodal map In Voxel-to-voxel alignment with map Has better anatomical contrast Baseline functional Low-B DTI Usually a MC template Volume and Intensity ROI Analyses Functionally-constrained ROI