Overview

CIVET is a human brain image-processing pipeline for fully-automated corticometric, morphometric and volumetric analyses of magnetic resonance (MR) images. The pipeline is composed of a set of Perl scripts and high-level C/C++ programs.

CIVET starts by a 9 or 12-parameter linear/affine registration of the MR images (T1-weighted, T2-weighted and Proton Density weighted) from native to stereotaxic space, using the average MNI ICBM152 model (or any other volumetric model that is available) as the target of registration. Once the images have been transformed to stereotaxic space, the images are corrected for radio-frequency non-uniformities and a brain mask is computed. A non-linear transformation is then computed from the subject in stereotaxic space to the model.

The next step in CIVET is to perform the tissue classification into white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF), either based on the T1-weighted image only, or on multispectral data (T1, T2, PD). The classification is comprised of two steps: a discrete classification based on tag points, followed by the evaluation of the partial volumes for the tissue classes.

The brain is split into the left and right hemispheres for the purpose of surface extraction. The boundary between cortical GM and subcortical WM (referred to as the white matter surface) is first extracted. The pial surface, or the boundary betwen the cortical GM and the extra-cortical CSF (referred to as the GM surface) expands outwards from the WM surface to the CSF. Each default surface (per hemisphere) is composed of 81,920 triangles (polygons) and 40,962 vertices. High resolution surfaces are also available at 327,680 polygons and 163,842 vertices.

The surfaces are registered to the MNI ICBM152 surface template (or any other surface model that is available). Surface registration as such allows for group comparisons by interpolating vertex-based cortical measurements at reference vertices on the template. The cortical thickness is computed by evaluating the distance, in mm, between the original WM and GM surfaces transformed back to the native space of the original MR images, then interpolated onto the surface template. Vertex-based areas and volumes are, on the other hand, computed directly on the resampled surfaces and measure local variations of area/volume contraction and expansion relative to the vertex distribution on the surface template.

Regional maps are produced based on a lobar parcellation of the surfaces (major lobes, AAL, or DKT-40). Other measurements such as mean curvature, gyrification index, and total cortical area are also computed.

A number of figures, tables, and reports are automatically produced to assist in the process of quality control and data analysis.

What’s new in CIVET-2.0.0?

The major improvements are outlined below.

  • Marching-cubes algorithm:
A marching-cubes algorithm has been implemented to replace the deformable ellipsoid model for the extraction of the white matter surface to produce a high-quality surface without surface bridges.
  • White surface t1-gradient calibration:
The white surface is adjusted locally to the maximum gradient position of the t1 MR image at the border between white and gray matter. The direct consequence of the calibration step is to reduce significantly the sensitivity of the N3 parameters (choice of N3 spline distance) on the tissue classification algorithm, making results almost independent of the chosen spline distance.
  • New N3 version:
The new N3 version (1.11.0 and up), which accounts correctly for damping, is invariant to voxel sizes and allows processing of images at different resolutions (commonly 0.5mm or 1.0mm).
  • High-resolution surfaces:
High-resolution surfaces, at 163,842 vertices and 327,680 triangles, offer increased definition of the cortical surfaces.
  • Surface blurring fwhm:
The value of the fwhm for surface blurring has been corrected due to a bug in CIVET-1.1.12. (Older versions of CIVET like 1.1.11 were not affected by this bug.) The value of fwhm in CIVET-2.0.0 is correct. The value of fwhm in CIVET-1.1.12 was off by a factor of sqrt(2). For example, a specified fwhm=30mm would give an effective fwhm=21.2mm in CIVET-1.1.12.
  • Mean curvature:
The mean curvature is wrong in CIVET-1.1.12 by a factor of 2 due to the same bug that affected the surface blurring fwhm. To obtain the correctly scaled values of mean curvature, multiply the values from CIVET-1.1.12 by 2. Also, don’t forget to compensate for the sqrt(2) factor (see above) when blurring mean curvature. In CIVET-2.0.0, everything is correct.
  • Classified tissue volumes:
The total volumes in the file classify/cls_volumes.dat are now masked by the gray surfaces. They include cerebrum gray matter (cortical and sub-cortical), white matter and CSF. The values exclude the cerebellum, the brainstem and the skull. These values in CIVET-1.1.12 were not masked. These measurements are reported in native space.
  • ANIMAL segmentation:
The ANIMAL volumetric segmentation has been reinstated to provide regional volumes of sub-cortical structures.

Some consequences of these major improvements are:

  • Reduced dependence of results on the N3 spline distance:
A value of the spline distance of 200 is suggested for scans at 1.5T while a distance of 100 or 125 is suitable for 3T scans.
  • New average population surfaces:
New average population surfaces must be used for surface registration and statistical analyses because the projection of the vertices onto the brain is different between the marching-cubes algorithm and the deformable ellipsoid model previously used. Consequently, it is not possible to do vertex-to-vertex comparisons of results obtained with CIVET-2.0.0 vs CIVET-1.1.12 since the surface registration targets are different.

Overall, CIVET-2.0.0 provides a major overhaul of the surface extraction and associated algorithms to provide high-quality cortical surfaces and more precise measurements of cortical thickness. New functionalities have been added, for instance: processing at different voxel sizes, high-resolution surfaces, ANIMAL segmentation, improved quality control tools, etc. The user should also find this new version easier to use since results are mostly invariant to the N3 spline distance. Finally, fewer subjects are lost due to processing failures (e.g. mis-registration).

Finally, the computational resources required by CIVET 2.0.0 are shown below. At comparable conditions (1mm voxel size and low-res surfaces), CIVET 2.0.0 offers a reduction in disk space due to the elimination of some unnecessary intermediate files and cropping of the non-linear transformation deformation field. In terms of computational time, CIVET 2.0.0 is very competitive throughout the various options, from low-res to hi-res surfaces at different voxel size resolution. (In the figures, the suffix _mc means marching-cubes for CIVET 2.0.0.)

Full summary of major changes between CIVET 1.1.12 and CIVET 2.0.0: 

    *************************************************************************
    New in September, 2014 quarantine, version 2.0.0

    1 - marching-cubes algorithm for extraction of initial white surface
    2 - new calibration of white surface at GM-WM t1-gradient
    3 - improved node movement in surface fit program
    4 - corrected functional for Laplacian constraint in surface
        fit program (for gray surface)
    5 - improved surface registration 
    6 - new surface registration models based on marching-cubes
    7 - simplified CIVET QC pipeline
    8 - corrected pve classification at 0.50mm volume template
    9 - all changes listed below for 1.1.13
    10 - ANIMAL volumetric lobar segmentation
    11 - corrected fwhm for surface smoothing (was off by sqrt(2) in 1.1.12)

    Note: This version requires the package surface-extraction-3.0.2
          or higher.

    More to come soon:

    1 - classification and surface extraction in native space.
    2 - intermediate model for linear and non-linear volume registration.
    3 - Plug-in for alternate population templates
    4 - removal of self-intersections in .sm file for surface
        registration (Maxime Boucher)
    5 - Longitudinal surface registration and QC

    *************************************************************************
    New in July, 2013 quarantine, version 1.1.13 (not released)

    1 - bug fixed in N3 to be invariant to voxel size (requires N3-1.12.0
        and EBTKS-1.6.4)
    2 - high-resolution surfaces at 320k polygons
    3 - surface registration on high-resolution surfaces, true
        multi-resolution algorithm (faster)
    4 - improved medial cut through the corpus callosum
    5 - processing at voxel sizes of 0.5mm and 1.0mm
    6 - speed optimizations for pve
    7 - speed optimizations for Laplacian field at 0.5mm
    8 - cropping of non-linear xfm grid file to save disk space
        and to reduce peak memory
    9 - masking of hippocampus and amygdala for icbm152nl_09s model (0.5mm)
    10 - masking of hippocampus and amygdala for ADNIhires model (0.5mm)
    11 - improved schedule for white matter surface extraction
         (smoother convergence, scaled starting ellipsoid)
    12 - new ICBM surface average for scaled ellipsoid
    13 - clean-up of connected pieces of white matter (partial
         volumes) 
    14 - new pipeline to build average surfaces from CIVET output

Next Section: Basic Usage of CIVET

CIVET Home