Structural Processing

A common processing stream for the T1-weighted anatomical scans was used for both the physiological and functional scanning sessions.

Defacing Structural Volumes

Prior to making the “raw” structural data available, defacing was performed to further anonymize the data. First, defacing of each pediatric template was performed manually to create a mask preserving the brain and scalp, but without facial regions. The defacing procedure was then performed as follows:

  • A copy of the subjects anatomical T1 or T2-weighted volume was made and bias corrected using the N4 algorithm [Tustison2010].
  • The defaced age-appropriate pediatric head template was then warped into the subject’s anatomical space using the SyN algorithm [Avants2008].
  • The computed transform was used to warp the non-face mask into subject space and then the mask was multiplied by the original (pre bias-correction) anatomical to give a “raw” volume that has been defaced, but has not yet had any other processing applied.

Structural Preprocessing

The structural volume was first bias corrected using the N4 algoithm [Tustison2010] as implemented in ANTs. The brain was then extracted using FSL’s brain extraction tool [Smith2002]. The brain mask was applied to the original structural volume and N4 bias correction was repeated on the brain-extracted volume.

Note

Brain extraction did not perform well on a handful of the youngest subjects (e.g. less than 6 months old). For these subjects, we will need to implement a revised approach)

Segmentation

A 3-class tissue segmentation was performed using FSL’s FIRST [Zhang2001]. No priors were used during segmentation.

Note

Segmentation results are currently poor for very young subjects (e.g. less than 6 months old). For these subjects, we will need a revised approach)

Registration to the Study Template

A series of four 2mm T1-Weighted age-specific templates in MNI coordinate space were generated using repeated non-rigid registration of inidividual T1W volumes as implemented in ANTs buildtemplateparallel.sh script.

The four templates correspond to the following age ranges:

  • 0-6 months
  • 6 months - 24 months (2 years)
  • 24 - 48 months (2-4 years)
  • 48 - 216 months (4-18 years)

Based on the subject’s age at scan, a nonlinear symmetric diffeomorphic image registration was used to transform the structural volume into the space of the study template. This algorithm is referred to as SyN [Avants2008] as implemented in ANTs under antsRegistration.

Note

The cmind_normalize.py module also has options to use other registration approaches such as FLIRT, but the data in the C-MIND database was registered using SyN.

Normalization to MNI space

Affine transformations from the study templates to MNI space were precomputed and then used across all subjects. These affine transformations are provided along with the templates. The template folders are labeled by the age ranges in months. The affine transformation from a 2mm study template to 2mm MNI space appropriate for a 5 year old would be the one in: /48to216/SyN/48to216_to_MNI_brain_2mm_ANTS_SyN_0GenericAffine.mat

The affine transformation is concatenated with the nonlinear transformation from anatomical to study template space computed above to generate a transformation directly from anatomical to MNI space without intermediate interpolation stages. See the help text for the ANTs tool antsApplyTransforms for more information on concatenating transforms.