On brain templates and atlases, and how to build them

Niko Sirmpilatze

Contents

  • On brain templates and atlases
  • Image registration primer
  • How to build a brain atlas
  • brainglobe-template-builder

Where did I learn all this?

On templates and atlases

What is a brain template?

  • a standard reference for brain anatomy
  • a common coordinate system - a “space”
  • aliases: reference, standard

Multiple reference images - same space

What are templates used for?

  • average results from multiple subjects within a study
  • compare results across studies (e.g. coordinate-based meta-analysis)
  • measure deviations from the “norm”:
    • example: Deformation-based morphometry

Lerch et al., 2011. Methods Mol Biol.

Some “famous” templates

👨 Human

🐭 The Allen Mouse Brain Common Coordinate Framework

What is an atlas?

A Titan condemned to hold up the heavens or sky for eternity after the Titanomachy

The most superior (first) cervical vertebra (C1) of the spine and is located in the neck.

A collection of maps; it is typically a bundle of maps of Earth or of a region of Earth.

A Brain Atlas:

is composed of serial sections along different anatomical planes of the healthy or diseased developing or adult animal or human brain where each relevant brain structure is assigned a number of coordinates to define its outline or volume. Brain atlases are contiguous, comprehensive results of visual brain mapping and may include anatomical, genetic or functional features.

Annotations

aliases: labels, regions, parcellations, segmentations, structures

🐭 The Allen Mouse Brain Atlas

👨 Brodmann areas (1909)

source: Wikipedia

👨 The HCP Multimodal Parcellation (2016)

Glasser et al., 2016. Nature

Atlases: from minimal to comprehensive

Minimal

  • 1 reference image
  • 1 flat annotation image

🐒 Macaque D99 atlas, source: AFNI

Comprehensive

  • multi-modal and/or -scale templates
  • multiple/hierarchical annotations

🐒 Macaque CHARM atlas, source: Jung et al., 2021

How are atlases used

Wang et al., 2020. Cell

Too many atlases!

source: https://xkcd.com/927

Remedies

Atlases are at the heart of BrainGlobe

Image registration primer

Linear transformations

Nonlinear transformations

3 params per voxel (3D vector field), source: 3blue1brown

Symmetric Diffeomorphic (SyN): Avants et al., 2011. Neuroimage

Image registration as an optimisation problem

  • Degrees of freedom (DOF): rigid (6), affine (12), nonlinear (n_voxels x 3)

  • Constraints: maximize similarity, minimize deformation

  • Similarity metrics: mutual information, cross-correlation etc.

  • Optimisation algorithms: Nowdays mostly gradient descent/ascent

The “scale pyramid”
  • First match rough features, then refine
  • First linear, then nonlinear
  • Increase DOF as we go down the pyramid

Advanced Normalisation Tools (ANTs)

ANTS

  • implemented in C++
  • uses ITK (Insight Segmentation and Registration Toolkit)
  • command-line interface
  • includes the canonical SyN algorithm implementation
  • provides full control over the optimisation process:
    • transform types, DOF, similarity metrics, “scale pyramid” traversal, etc.
  • Two main commands: antsRegistration and antsApplyTransforms
  • Provides some shell scripts for convenience:
    • e.g. antsMultivariateTemplateConstruction2.sh
  • very powerful, but not very user-friendly

How to build a brain atlas

Building an atlas step-by-step

1. Acquire 3D whole-brain images

  • Medical imaging: CT, MRI
  • 3D light microscopy: STPT, fMOST, LSFM

2. Build a template

  • Pick a single reference subject, or
  • Build an unbiased population average

3. Annotate brain areas

  • Needs manual curation from experts
  • Can incorporate data-driven approaches

4. Package and distribute

  • Publish on a public repository
  • Make it easy to use (docs, API)

What’s our plan?

Step Task Method
1. 3D whole-brain imaging AMF + collaborating imaging facilites
2. Build a template braiglobe-template-builder
3. Annotate regions Collaborating experts (with napari widgets?)
4. Package for BrainGlobe Improved bg-atlasgen (Alessandro & Viktor)
5. Distribution bg-atlasapi

Biased population average

Unbiased population average

Fonov et al., 2011. Neuroimage

Iterative averaging sharpens image

Wang et al., 2020. Cell

Single-subject vs unbiased average

Wang et al., 2020. Cell

Implementations of unbiased template construction

  • ANTs: antsMultivariateTemplateConstruction2.sh (uses SyN for nonlinear registration)
  • optimized_antsMultivariateTemplateConstruction by CoBra Lab
    • more sensible defaults to the ANTs script
    • optimal traversal of “scale pyramid” (depending on image size)
    • allows resuming if interrupted
    • designed for parallelisation on HPC clusters (with SLURM)
    • integrates masking during registration

brainglobe-template-builder

Why is it needed?

Standing on the shoulders of giants (or turtles all the way down):

Added value:

  • GUI via napari widgets
  • Preprocessing of images
  • tiff > nifti conversion
  • Feed into the template construction pipeline
  • Visualise results (QC)

Necessary preprocessing

Demo time!

brainglobe-template-builder

NIU team meeting | 2024-01-17