The HCP Pipelines reflect a concerted effort to improve the spatial accuracy of MRI data preprocessing so that the HCP Consortium and HCP Users can take full advantage of the high quality HCP data acquired for each of four modalities (structural MRI, resting-state fMRI, task-fMRI, and diffusion MRI). Another major goal was to improve the accuracy (and thereby validity) of cross-subject and cross-study spatial comparisons. This resulted in the development of the CIFTI Grayordinates standard space, which brings the advantages of surface-based cortical analyses into a whole-brain analysis framework.

An overarching purpose of CIFTI files is to allow spatial models of MRI data to better match the anatomical structures of the brain. The sheet-like cerebral cortex is better modeled as a surface mesh, whereas the globular subcortical nuclei are better modeled as volume parcels Glasser et al., 2013. (Cerebellar cortex is also a sheet, but unfortunately cannot yet be accurately segmented in individual subjects, so it is included as a volume parcel.) A space containing both cortical surface vertices and subcortical volume voxels is made up of grayordinates. The HCP uses a 2mm standard space made up of 91,282 grayordinates (2mm average spacing between surface vertices and 2mm isotropic voxels). Besides allowing for more precise analyses of brain MRI data, the grayordinates space markedly reduces the data storage, computational, and memory requirements for high spatial and temporal resolution data, by only storing the minimum data of interest.

• For structural analysis, one needs a high-resolution 3D (<=1mm) T1w image and either a high-resolution 3D (<=1mm) T2w image or a high-resolution 3D (<=1mm) FLAIR image. Higher than 1mm resolution is recommended if possible (e.g. <=0.8mm). FatSat is recommended for the T1w image. Scanner-based intensity normalizations (like Siemens PreScan Normalize) must be either on for both T1w and T2w images or off for both.

• For functional analysis, one needs an fMRI timeseries (if the scans are multi-band, it is highly recommended to save the single-band reference scan) and either a standard gradient echo field map (magnitude and phase difference) or a spin echo field map (two phase encoding direction reversed spin echo volumes with the same geometrical and echo spacing parameters as the fMRI timeseries). High spatial (<=2.5mm) and temporal (TR<=1s) resolution fMRI data is recommended.

• For diffusion analysis, one needs phase encoding direction reversed diffusion data. High spatial (<=1.5mm) and angular (>=120 directions) resolution diffusion data is recommended with multiple shells that include a higher bvalue (>=1500).

• Functional and diffusion preprocessing also require the scans needed for structural preprocessing.

• For the Siemens HCP 3T Connectom scanner, we found the 32-channel head coil and higher maximum gradient strength to be very helpful.

  • The use of Connectom when referring to the scanner itself is not a typographical error. Siemens considers the custom 3T scanner used for the Human Connectome Project to be the "Connectom" scanner.

Because the HCP pipelines aim to improve spatial accuracy and localization, any distortions present in the images need to be corrected. Accurate registration of EPI fMRI or diffusion data to the structural scans requires correction for geometric distortion in the EPI scans.

It may be possible to attain much of the distortion correction using a group average field map (e.g., from a different set of subjects), and this capability may become available in a future pipelines release.

These images are used for bias field correction, improvements in FreeSurfer-generated pial surfaces, and generation of cortical myelin maps. If myelin maps are important, a T2w scan is preferred over a FLAIR, whereas if lesion segmentation is more important, a user may prefer a FLAIR. Myelin maps are quite helpful for localizing some cortical areas, even on an individual subject basis. (Glasser & Van Essen, J Neuroscience, 2011; Van Essen & Glasser, Neuroimage, 2013).

Yes, but for the gradient nonlinearity correction (Jovicich et al., 2006) you'll need to get the correction coefficients for the WU-customized 'Connectom' scanner (using the "SC72C" gradient coil) from Siemens, since these coefficients are considered to be proprietary information. To access these coefficients, your institution must have a research agreement or be willing to sign a non-disclosure agreement with the vendor. Please contact:

For USA: Yulin Chang

For rest of world: Martin Stoltnow

The gradient unwarping code used by the HCP Pipelines is a customized version of the code originally published by Krish Subramaniam (but no longer actively maintained). The HCP-customized version is available at https://github.com/Washington-University/gradunwarp/releases.

MRI scanners use gradients in the magnetic field to define the space of the image. These gradients are intended to be linear (i.e. have a constant slope). In practice this is not attainable, resulting in spatial distortions. The HCP 3T Connectom scanner has greater distortions than do more standard scanners because of design tradeoffs for this customized scanner.

All scanners have some level of gradient distortion. Therefore, for maximal anatomical fidelity and for comparing data across scanners (which will very likely have different gradient distortion levels), we encourage researchers to obtain gradient distortion coefficients and perform gradient distortion correction (GDC) (Jovicich et al., 2006). For Siemens scanners, the gradient coefficients are available on the scanner (C:\MedCom\MriSiteData\GradientCoil\coeff.grad).

Users of the HCP Pipelines may opt not to perform gradient distortion correction. For example, if a scanner is used in which the images are collected in a region of the magnet bore in which the gradients have very linear performance over the imaging field of view. The HCP Pipelines provide options which can be used to turn off gradient distortion correction.

We hope to release the FreeSurferNHP pipeline, along with monkey and chimpanzee anatomical templates, in the future. The image acquisition requirements are the same as for humans, except that spatial resolution should be higher, if possible.

At younger ages, children's heads are different enough from adults' heads to make the initial alignment stages of the HCP Pipelines less robust when using adult volume templates. It may be necessary to make age-specific templates for children under a certain age. Once the initial alignments are robustly achieved, the HCP Pipelines will perform similarly to adults (assuming typical T1w contrast is present).

We are planning to examine the robustness of the current pipelines in 4-6 year olds as part of the piloting phase for the LifeSpan project.

Currently it is necessary to acquire the structural data on a 3T scanner (mainly because of SAR limitations on 7T scanners, making the T2w or FLAIR scans hard to acquire). Then fMRI or diffusion data can be acquired on the 7T scanner. Gradient distortion correction is required if combining 3T and 7T data. Additionally, there may be further pipeline modifications directed at combining 3T and 7T data in the same subjects.

The current release of the HCP Pipeline Scripts aligns surfaces to template spaces using MSM; specifically the refined 2018 version (Robinson et al., Neuroimage, 2018), which builds upon (Robinson et al., Neuroimage, 2014). MSM offers improvements over standard FreeSurfer-based alignment in that there is substantially less surface distortion and improved alignment of cortical areas across subjects.

The PostFreeSurfer phase of Structural Preprocessing uses cortical folding only-based registration (MSMSulc).

Subsequent to Functional Preprocessing and ICA+FIX processing, re-registration is performed in the MSMAll pipeline using cortical folding along with myelin map and resting state functional information (MSMAll).

The MSM code is available at: https://github.com/ecr05/MSM_HOCR. For compiled MSM binaries see: https://github.com/ecr05/MSM_HOCR/releases.

The HCP Pipelines provide a framework for substantially improved spatial localization of MRI data, particularly across subjects and studies. That said, further substantial improvements are possible though better MSM surface registration based on areal features rather than just cortical folding patterns (as used by MSMSulc and FreeSurfer). We hope to release another MSM-based pipeline for areal-feature-based registration (using myelin maps and resting state networks) together with HCP data aligned with this pipeline in the future.

Glasser MF, Sotiropoulos SN, Wilson JA, Coalson TS, Fischl B, Andersson JL, Xu J, Jbabdi S, Webster M, Polimeni JR, Van Essen DC, Jenkinson M, WU-Minn HCP Consortium. The minimal preprocessing pipelines for the Human Connectome Project. Neuroimage. 2013 Oct 15;80:105-24. PubMed PMID: 23668970; PubMed Central PMCID: PMC3720813.

Please read the following post to the HCP-Users email list.

Search the hcp-users archive first. If you don't find the answer from searching, then subscribe to the HCP-Users email list at http://humanconnectome.org/contact/#subscribe on the bottom right of the page, then send a message to the list at [email protected].