Skip to content

Atlases used in Clinica

This page describes the different atlases used in the pipelines of Clinica.

Volume atlases

Processing of T1-weighted MRI & PET images

These atlases, all defined in MNI space, are mainly used when performing volumetric processing of T1 and PET images, as done in the t1-volume-* and pet-volume pipelines.

  • AAL2 [Tzourio-Mazoyer et al., 2002, 2015] is anatomical atlas based on a single subject. It is the updated version of AAL, which is probably the most widely used cortical parcellation map in the neuroimaging literature. It was built using manual tracing on the spatially normalized single-subject high-resolution T1 volume in MNI space (Holmes et al., 1998). It is composed of 120 regions covering the whole cortex as well as the main subcortical structures.
  • AICHA [Joliot et al., 2015] is a functional atlas based on multiple subjects. It was built using parcellation of group-level functional connectivity profiles computed from resting-state fMRI data of 281 healthy subjects. It is composed of 345 regions covering the whole cortex as well as the main subcortical structures.
  • Hammers [Hammers et al., 2003; Gousias et al., 2008] is an anatomical atlas based on multiple subjects. It was built using manual tracing on anatomical MRI from 30 healthy subjects. The individual subjects parcellations were then registered to MNI space to generate a probabilistic atlas as well as a maximum probability map. The latter was used in the present work. It is composed of 69 regions covering the whole cortex as well as he main subcortical structures.
  • LPBA40 [Shattuck et al., 2008] is an anatomical atlas based on multiple subjects.. It was built using manual tracing on anatomical MRI from 40 healthy subjects. The individual subjects parcellations were then registered to MNI space to generate a maximum probability map. It is composed of 56 regions covering the whole cortex as well as he main subcortical structures.
  • Neuromorphometrics is an anatomical atlas based on multiple subjects. It was built using manual tracing on anatomical MRI from 30 healthy subjects. The individual subjects parcellations were then registered to MNI space to generate a maximum probability map. It is composed of 140 regions covering the whole cortex as well as he main subcortical structures. Data were made available for the “MICCAI 2012 Grand Challenge and Workshop on Multi-Atlas Labeling”.

The main difference between LBPA40, Hammers and Neuromorphometrics atlases is the degree of detail (i.e. the number of regions) of the anatomical parcellation.

Processing of DWI images

These atlases, all defined in MNI space, are mainly used when processing DWI images, as done in the dwi-processing-dti pipeline.

  • JHUDTI81 [Hua et al., 2008; Wakana et al., 2007]: This atlas contains 48 white matter tract labels that were created by manually segmenting a standard-space average of diffusion MRI tensor maps from 81 subjects.
  • JHUTracts[0|25|50] [Mori et al., 2005]. This atlas contains 20 white matter tract labels that were identified probabilistically by averaging the results of deterministic tractography run on 28 subjects. Several thresholds of these probabilistic tracts are proposed (0%, 25%, 50%).

Surface atlases

These atlases are mainly used when processing T1-weighted images with the t1-freesurfer-* pipeline and PET images with the pet-surface pipeline.

  • Desikan [Desikan et al., 2006]: This atlas is a subdivision of the cerebral cortex into gyri and contains 34 regions per hemisphere. It was built using a dataset of 40 MRI scans from which 34 cortical ROIs were manually identified in each of the individual hemispheres.
  • Destrieux [Destrieux et al., 2010]: This atlas is a subdivision of the cerebral cortex into gyri and sulci, and contains 74 regions per hemisphere. It was built on anatomical MRI of 24 healthy subjects from which 74 cortical ROIs were manually identified in each of the individual hemispheres.

Tip

Easily access the papers cited on this page on Zotero.