Elsevier

NeuroImage

Volume 197, 15 August 2019, Pages 818-826
NeuroImage

Ciftify: A framework for surface-based analysis of legacy MR acquisitions

https://doi.org/10.1016/j.neuroimage.2019.04.078Get rights and content

Highlights

  • Ciftify allows for grayordinate-based (CIFTI format) analysis of non-Human Connectome Project (i.e. legacy) MR acquisitions.

  • The workflow and dependencies are distributed as a Docker container, following the BIDS-app interface.

  • Additional ciftify utilities aid in downstream analysis of CIFTI images.

  • This work offers a bridging solution for legacy data that will allow many researchers to adopt CIFTI format analyses.

Abstract

The preprocessing pipelines of the Human Connectome Project (HCP) were made publicly available for the neuroimaging community to apply the HCP analytic approach to data from non-HCP sources. The HCP analytic approach is surface-based for the cerebral cortex, uses the CIFTI “grayordinate” file format, provides greater statistical sensitivity than traditional volume-based analysis approaches, and allows for a more neuroanatomically-faithful representation of data. However, the HCP pipelines require the acquisition of specific images (namely T2w and field map) that historically have often not been acquired. Massive amounts of this ‘legacy’ data could benefit from the adoption of HCP-style methods. However, there is currently no published framework, to our knowledge, for adapting HCP preprocessing to “legacy” data. Here we present the ciftify project, a parsimonious analytic framework for adapting key modules from the HCP pipeline into existing structural workflows using FreeSurfer's recon_all structural and existing functional preprocessing workflows. Within this framework, any functional dataset with an accompanying (i.e. T1w) anatomical data can be analyzed in CIFTI format. To simplify usage for new data, the workflow has been bundled with fMRIPrep following the BIDS-app framework. Finally, we present the package and comment on future neuroinformatics advances that may accelerate the movement to a CIFTI-based grayordinate framework.

Introduction

The Human Connectome Project (HCP) has incorporated major technical advances at many steps of neuroimaging data acquisition and analysis (Van Essen et al., 2013; Glasser et al., 2016a). At the level of MR acquisition, the HCP used multi-band MR pulse sequences, which increased both the temporal and spatial resolution of MR data (Van Essen et al., 2012; Uğurbil et al., 2013). In addition, the HCP project took great care to utilize state-of-the-art approaches to correct for MR field bias and image distortions apparent across MR modalities (Glasser et al., 2016a). At the level of MR analysis, as part of the minimal preprocessing pipelines, the HCP introduced a novel Connectivity Informatics Technology Initiative file format (CIFTI; https://www.nitrc.org/projects/cifti/) for conducting analyses in the “grayordinate” framework. In CIFTI format, data from cerebral cortical gray matter is stored in relation to 2-dimensional surface meshes, whereas subcortical data is maintained in 3-dimensions, within the same file, by representing only subcortical gray matter voxels. The HCP also introduced a powerful visualization tool, Connectome Workbench, with an accompanying suite of command-line functions that allow for the manipulation of CIFTI, GIFTI and NIFTI format images.

A 2D cortical surface-based approach to MR analysis of cortical signals provides several advantages over the more commonly used 3D volume-based approach as previously described (Fischl, 2012; Argall et al., 2006; Goebel et al., 2006; Zijdenbos et al., 2002). Most notable benefits include better adherence to the inherent geometry of cortical surfaces, increased statistical power (Argall et al., 2006; Anticevic et al., 2008; Tucholka et al., 2012; Fischl et al., 1999; Jo et al., 2007; Coalson et al., 2018), removal of the deleterious effects of volume-based smoothing (which markedly erodes spatial localization (Coalson et al., 2018), superior visualization (Van Essen, 2012), and finally a simplified, more compact framework for multimodal analysis (such as analyses combining surface-based anatomical features, e.g., cortical thickness, and fMRI). If should be noted that these benefits have been reported using data with lower spatial and temporal resolutions than the HCP. For these reasons, surface-based analyses have provided new insights into how the human cortex is functionally organized within humans at both the population level (Glasser et al., 2016b; Yeo et al., 2011; Mueller et al., 2013; Margulies et al., 2016; Van Essen and Glasser, 2018) and individual level (Glasser et al., 2016b; Wang et al., 2015; Braga and Buckner, 2017).

Although the benefits of surface-based over volume-based registration have been widely recognized for nearly a decade and despite the continued development and promulgation of surface-based tools by several groups (Fischl, 2012; Argall et al., 2006; Goebel et al., 2006; Zijdenbos et al., 2002; Van Essen, 2012), the adoption of surface-based analyses by the neuroimaging community has been slow. One reason for this inertia is that many widely used tools for MR preprocessing (e.g., SPM, FSL) as well as widely adopted pipeline tools (e.g., NIAK, CPAC) do not offer “out of the box” workflows with surface-based registration steps. These tools are not designed to support the CIFTI format, or provide visualization frameworks for combined 2D/3D geometry. The HCP consortium released its pipelines as part of the Minimal Preprocessing Pipeline GitHub project (Glasser et al., 2013). One key requirement of the Minimal Preprocessing Pipeline is acquisition of a high-resolution T2-weighted image, used to generate high-quality surface reconstructions (Glasser et al., 2013) and myelin maps (Glasser et al., 2014; Glasser and Van Essen, 2011), that was not typically acquired in legacy human MR protocols. Therefore, many ‘legacy’ acquisitions often collected without high-resolution T2 images or fieldmaps cannot be processed using the HCP's pipelines.

It is estimated that tens of thousands of participants are scanned annually as part of research studies, and this number has been growing for over 25 years (Smith, 2012). Thanks to data sharing consortia such as the International Neuroimaging Data-Sharing Initiative (INDI (Milham et al., 2018),) and OpenfMRI (Poldrack et al., 2013), thousands of legacy datasets are publicly available (Poldrack and Gorgolewski, 2014; Eickhoff et al., 2016). The National Institute of Mental Health (NIMH) have also committed to data sharing via the NIMH Data Archive (NDA; ndar.nih.gov). These legacy datasets are drawn from healthy individuals plus a wide variety of clinical populations and developmental stages. While some clinical and developmental populations are now being scanned according to HCP acquisition standards, it may be years before these sample sizes will be large enough to answer some of today's most pressing questions. The HCP requirements can be viewed either as a barrier or as an opportunity. If viewed as an opportunity, there is a concomitant need to develop tools for HCP-style analyses applicable to legacy MR data available today (and likely the near future) Therefore, maximally leveraging large legacy datasets is important for clinical research, especially for those attempting to characterize disease heterogeneity (Choudhury et al., 2014).

Here, we address the opportunity to leverage important innovations of the HCP pipelines to enable integration of two decades of existing legacy human neuroimaging data into the CIFTI grayordinate-based framework. To expand the utility of HCP-style methods, we present the ciftify package for grayordinate-based (CIFTI format) analysis of legacy acquisitions that have already been processed using FreeSurfer. (Datasets that use other cortical segmentation methods or have no surface-based processing at all are considered in the discussion.) Ciftify translates two key modules of the HCP Minimal Preprocessing Pipeline: the FreeSurfer-to-Connectome Workbench conversion, and the fMRI surface projection, into simple command line tools. Integrating these two steps will work from FreeSurfer-based anatomical outputs (generated from only T1w images) to convert current volume-based fMRI analysis pipelines into a grayordinate-based one. These tools allow researchers to move their analyses to the surface while ensuring the opportunity to analyze non-HCP quality data. Below, we describe the ciftify package and discuss its expected use case. We also introduce additional tools for running and interpreting group-level analyses in CIFTI format.

Section snippets

The ciftify preprocessing workflow and BIDS-app

A diagram of the ciftify preprocessing workflow is given in Fig. 1. As a precursor to ciftify, surfaces are generated from T1w anatomical images using FreeSurfer's recon_all function (Fischl, 2012), and fMRI runs are preprocessed using other software as discussed below. The BIDS-app Docker container will use FMRIPrep (Esteban et al., 2019a, 2019b) for these preprocessing steps if they have not already been run. Anatomical data is converted from FreeSurfer to CIFTI formats, and MNI inter-subject

Running analyses in CIFTI format

After data has been preprocessed into CIFTI format, the researcher needs to manipulate CIFTI files, in order to extract measures or run group analyses. For a researcher experienced only in volume-based analysis, this may present a challenge, as code may need to be adapted or rewritten. With recent software developments, most (if not all) calculations previously run in the volume are now possible on the surface. The Connectome Workbench command-line utility (wb_command) offers an extensive suite

Discussion

The brain imaging field has lacked a framework for leveraging the HCP preprocessing philosophy for non-HCP legacy data. The ciftify project presented here helps to close this gap and encourages wider adoption of surface-based analysis of MR data, specifically using the CIFTI file format, for groups working with non-HCP legacy datasets. Allowing researchers to adapt the preprocessing pipelines they are most familiar with allows for 1) easier uptake of surface-based analysis insofar as we remove

Research data statement

All templates and code are available at https://github.com/edickie/ciftify and archived at https://doi.org/10.5281/zenodo.2651201.

Acknowledgements

ANV acknowledges support related to the present work the Canadian Institutes of Health Research, Ontario Mental Health Foundation, and Centre for Addiction and Mental Health Foundation, the Ontario Ministry of Research and Innovation, the Canada Foundation for Innovation. MFG and DCVE acknowledge support from NIH (F30 MH097312 to MFG, RO1 MH-60974 to DCVE, and 3R24 MH108315 to DCVE).

References (65)

  • D.C. Van Essen et al.

    The brain analysis library of spatial maps and atlases (BALSA) database

    Neuroimage

    (2017)
  • B. Fischl

    FreeSurfer

    Neuroimage

    (2012)
  • B. Fischl et al.

    Cortical Surface-Based Analysis. II: Inflation, Flattening, and a Surface-Based Coordinate System

    (1999)
  • M.F. Glasser et al.

    The minimal preprocessing pipelines for the Human Connectome Project

    Neuroimage

    (2013)
  • M.F. Glasser et al.

    Trends and properties of human cerebral cortex: correlations with cortical myelin content

    Neuroimage

    (2014)
  • D.N. Greve et al.

    Accurate and robust brain image alignment using boundary-based registration

    Neuroimage

    (2009)
  • C. Hutton et al.

    Image distortion correction in fMRI: a quantitative evaluation

    Neuroimage

    (2002)
  • H.J. Jo et al.

    Spatial accuracy of fMRI activation influenced by volume- and surface-based spatial smoothing techniques

    Neuroimage

    (2007)
  • D.S. Marcus et al.

    Human Connectome Project informatics: quality control, database services, and data visualization

    Neuroimage

    (2013)
  • S. Mueller et al.

    Individual variability in functional connectivity architecture of the human brain

    Neuron

    (2013)
  • E.C. Robinson et al.

    MSM: a new flexible framework for Multimodal Surface Matching

    Neuroimage

    (2014)
  • E.C. Robinson et al.

    Multimodal surface matching with higher-order smoothness constraints

    Neuroimage

    (2018)
  • G. Salimi-Khorshidi et al.

    Automatic denoising of functional MRI data: combining independent component analysis and hierarchical fusion of classifiers

    Neuroimage

    (2014)
  • T. Tong et al.

    Functional density and edge maps: characterizing functional architecture in individuals and improving cross-subject registration

    Neuroimage

    (2017)
  • A. Tucholka et al.

    An empirical comparison of surface-based and volume-based group studies in neuroimaging

    Neuroimage

    (2012)
  • K. Uğurbil et al.

    Pushing spatial and temporal resolution for functional and diffusion MRI in the Human Connectome Project

    Neuroimage

    (2013)
  • A.M. Winkler et al.

    Permutation inference for the general linear model

    Neuroimage

    (2014)
  • A.M. Winkler et al.

    Multi-level block permutation

    Neuroimage

    (2015)
  • A. Abraham et al.

    Machine learning for neuroimaging with scikit-learn

    Front. Neuroinf.

    (2014)
  • B.D. Argall et al.
    (2006)
  • J.M. Bjork et al.

    The ABCD study of neurodevelopment: identifying neurocircuit targets for prevention and treatment of adolescent substance abuse

  • R.M. Braga et al.

    Parallel interdigitated distributed networks within the individual estimated by intrinsic functional connectivity

    Neuron

    (2017)
  • Cited by (80)

    • Using synthetic MR images for distortion correction

      2023, Developmental Cognitive Neuroscience
    View all citing articles on Scopus
    View full text