Human Connectom Project :
The minimal processing Pipeline
Human Connectom Project : The minimal processing Pipeline Van Essen DC, … The WU-Minn Human Connectome Project: an overview. Neuroimage. 2013 Marcus DS, ... Human Connectome Project informatics: quality control, database services, and data visualization. Neuroimage. 2013 Glasser MF,... The minimal preprocessing pipelines for the Human Connectome Project. Neuroimage. 2013 Sotiropoulos ... Advances in diffusion MRI acquisition and processing in the Human Connectome Project. Neuroimage. 2013 Smith SM,... Resting-state fMRI in the Human Connectome Project. Neuroimage. Uğurbil K, Xu J,... Pushing spatial and temporal resolution for functional and diffusion MRI in the Human Connectome Project. Neuroimage. 2013 Larson-Prior LJ, .... Adding dynamics to the Human Connectome Project with MEG. Neuroimage. connectome 300mT McNab JA,... The Human Connectome Project and beyond: initial applications of 300 mT/m gradients. Neuroimage. 2013 Setsompop K, ... Pushing the limits of in vivo diffusion MRI for the Human Connectome Project. Neuroimage. 2013 Marcus DS, ... Van Essen DC. Informatics and data mining tools and strategies for the human connectome project. Front Neuroinform. 2011 Jun
HCP overview
Objective : • • • • •
characterize brain connectivity and function and their variability in healthy adults. 2011-2016 2 years of preparation 3 year of scanning 1200 Subjects (22-35 years) : a population of adult twins and their non-twin siblings The state of art of acquisition → Share the data and the process (first release march 2013)
more than 100 investigators and technical staff from ten institutions in the consortium Scan : New 3T scanner (skyra) + new acquisition sequence Multi Band EPI
• • • • •
customize 3T skyra (gradient 100mT/m) (commercial version 80mT/m) Scan : ~ 3H30 per subjects in 4 or 5 session Structural : 40 mn : 0.7 iso + calibration scans (B0 B1+ B1-) resting states fMRI : 1H : 2 mm iso, TR 0.7 s ( 2x2 sesion RL / LR) task- fMRI : 1 H : 7 task working memory, reward processing, motor processing language, social cognition, relational processing and emotion processing session difusion MRI : 1H : 1.25 mm iso . 18 B0 3*90 DWI with Bvalue of 1000 2000 3000. repeated twice RL/LR same acquisition at 7T for subset of 200 subjects MEEG/EEG task and resting states for a subset of 100 subject Behaviour Genetic
• •
Head motion monitoring with a camera. cardiac and respiratory signals associated with each scan
• • • • •
Acquisition parameters optimisation
A new scanner 3T with Hig gradient
Multi Band SNR • • •
no SNR loss due to reduced data collection (a small SNR penalty associated with slice aliasing) BUT T1 relaxation effects from the shorter achievable Trs avec IPAT large signal fluctuations (amplify by diffusion encoding)
Acquisitions Details Structural T1 w (mprage) / T2w (sapce) at 0.7 iso optimisation of contrast parameters : FA, TI and TE B0 field maps / B1+ B1- (for readout distortion and bias correction) at 2mm iso B1- : receive field from a FLASH 32 head coil versus body coil B1+ : transmit field from flip-angle imaging (Yarnykh,2007) FA 50° TR1/TR2 20/120ms
fMRI multiband factor of 8, 2 mm iso, TR 0.7 s FA=52° no PF. TE 33ms matrix 104*90*72 for each session a Spin echo field map : 2 spin echo EPI with reversed phase encoding directions (2*3 images : 60s) A single band reference scan (for a better gray/white contrast) Always 2 sesion RL / LR
Diffusion multiband factor of 3, 1.25 mm iso . 18 B0 3*90 DWI with Bvalue of 1000 2000 3000. repeated twice RL/LR. Mono polar diffusion-enconding sheme. Direction-resversed images for each diffusion direction (whole sphere) B1- acquisition twice LR / RL Multiband acceleration how much acceleration could be achieved before significant slicecross-talk occurred, or undesirable artefacts arose in the multiband reconstruction due to interactions with head motion (up to 8) We determined that distortion-corrected L–R and R–L datasets are, in general, anatomically well aligned with each other, even in regions of different dropout — it is only the dropout that differs. Q sampling strategie
A short TR the basic single-voxel timeseries detection power is minimally affected by reduction in TR (at least in the range ~0.5-3 s), as the simple statistical gain from the increased number of data points is balanced by the loss in signal level (as TR falls below T1). Advantage : • the denser temporal sampling of physiological confounds; • the importance of temporal degrees-of-freedom for many analysis techniques (such as high-dimensional ICA) richer temporal characterisation of resting-state fluctuations.
Spatial resolution For DWI
For Tractography
Spatial resolution for fMRI : choice of spatial resolution SNR but also the point spread function of the BOLD effect at 3 T (Parkes et al., 2005), which was measured to be ~3.5 mm at full-width-at-half-maximum (FWHM) caused by the dominant draining vein contribution at this field strength (2mm at 7T) but A more specific cortical signal : → decrease cross-sulcal and cross-gyral timeseries correlation.
Spatial Specificity of correlation map
Sulcus specificity of Resting-state components
Pre-processing
The cifti file format & grayordinates CIFTI format : •
Grayordinates : Combinations of cortical surface and subcortical volumetric parcels
• • •
dense timeseries : spatial grayordinate x time dense Connectome : seed points x spatial map of the connectivity dense scalar file (ica component | z-map …)
•
Header : contains the grayordinates space definition (l 1000=vertext 2152; line 91000 = 3.9 -1.4 10.6
•
normalisation ↔ resampling the sets of voxels in individually defined subcortical parcels to a standard set of voxels of the atlas parcel.
• • •
surface-constrained smoothing (→ adapted to cortex folding) subcortical parcel-constrained smoothing (→ no partial volume of surrounding structures)
•
save space (60 to 80%)
•
cifti → nifti or nifti + gifti
The minimal preprocessing pipelines
* Native Volume space (rigidly aligned to MNI aces)
PreFreeSurfer Pipeline
• • •
Undistorted “native” structural volume space (siemens correction not for all sequences, but use the Siemens gradient coefficient file) align T1 T2 B1 bias
readout distortion is different for T1 and T2 crop with fsl robustfov tool ACPC alignnment space (for visualisation and to remove rotation in the nifti sform) brain extraction with the template mask (after a non linear normalisation to MNI) more robust than BET readout distortion (very small same as EPI distortion) field map is register to T1 and T2 separatly. Boundary-based registration (BBR) cost function. Bias correction from T1*T2 : perserve gray matter inhomogeneity in T1 and T2 (different myelin content Fishl 2004) final output with spline interpolation (less blurring)
FreeSurefer Pipeline
• • •
segment the volume (ribon and CIFTI subcortical parcels) white and pial cortical surface definition standard surfaces registration (to fsaverage)
addon : Initial Mask 1 mm to 0.7 mm 0.7 T1 intensity normalized and white matter surface is adjusted gray matter surface is adjusted with T1 and T2 volume : T1w intensity normalisation with 4 sigmas (defautl is 3) → do not exclude lightly myelinated gray matter. Then T2w is used to erode the pial surface T1w is smoothed with a ribbon constrained approach (sigma 5mm) and pial surface is done again (with 2 sigmas)
improve pial surface
Extra Bias Field correction. Differenc with the Group average myelin maps (smooth of 14 mm) → substracted to the result.
PostFreeSurfer Pipeline
• • • • • • •
Make all nifti and gifti files for viewing wmparce 3 dilate 2 erosion → white matter mask for diffusion single spline interpolation to make T1 and T2 in MNI. registration to Conte69 surface template downsampling surfaces bidirectional resampling between high-resolution (164k) and low-resolution (32k) meshes. myelin maps
Conte69 : population-average surfaces which have correspondence between the left and right hemispheres VanEssen 2012
fMRIVolume pipeline
• • • • • •
remove spatial distortions realing volumes register to structural reduces the bias filed (from T1 estimation of different session!) normalizes intensity to a 4D mean (10 000) apply the final mask
Same correction of the gradient-nonlinearity-induced distortion (all images) realign on the single band reference image 12 parameters mvt (mvt+derived) demeaned and detrended for nuisance regression •
fMRISurface pipeline bring the timeseries to CIFTI space 2 mm FWHM smooth (surface and parcel constrained smoothing) (wheighted according to the partial volume inside the ribbon ) exclusion of noisy voxel (0.5 std above the 5mm sigma gaussian neighborhood) = voxel near the edge or with blood vessel. Gyral bias in high coefficient of variation is eliminated surface smoothin with vertex aera correction
SB Ref / MB, same dropout same distortion
dropout, which are different in L–R and R–L
distortion-corrected SBRef images after alignment to the structural images green : white-grey boundary from T1w. “ there is no obvious residual distortion that is different between these two images “
Consistent activation in the majority of the global contrasts, BUT some focused contrasts did not show consistent individual subject level activation. •
No orbital frontal activation in at least 50% of individual subjects in the emotion processing task;
•
no striatal activation in at least 50% of individual subjects in the comparison of reward versus baseline for the gambling task (SNR in subcortical regions);
•
no activation in parietal regions in the tools compared to other stimulus types contrast.
Improvment of surface analysis
Diffusion Pipeline
normalizes b0 intensity across runs removes EPI distortions : top up with extra spin echo EPI (RL LR) eddy-current-induced distortions + subject motion (eddy) corrects for gradient-nolinearities registers to structural
Gradient non-linearity for diffusion encoding
Advance processing Diffusion Qsampling Strategie
rFMRI post-processing : FIX • • • •
Data is provided from spatial (“minimal”) pre-processing, in both volumetric and grayordinate forms. Weak highpass temporal filtering (>2000s FWHM) is applied to both forms, achieving slow drift removal. MELODIC ICA is applied to volumetric data; artefact components are identified using FIX. Artefact and motion-related timecourses are regressed out of both volumetric and grayordinate data.
Optionally (and depending on further investigations), possibly also some combination of: • further motion cleanup/scrubbing; • further removal of physiological confounds based on physiological monitoring data; • removal of globally-related signals. (mean total signal or mean gray signal)
Results
