Brain parcellation atlases used for the registered brain overlay (Learn Cerebral v0.4)

1. Harvard-Oxford subcortical structural atlas, maximum-probability, 25% threshold, 1 mm, MNI152NLin6
   Retrieved from: https://github.com/neurodata/neuroparc
     atlases/label/Human/HarvardOxfordsub-maxprob-thr25_space-MNI152NLin6_res-1x1x1.nii.gz
   Original source: FSL (FMRIB, University of Oxford); Harvard Center for Morphometric Analysis.
   Licence: FSL licence, non-commercial use. Cite: Desikan et al. 2006; Frazier et al. 2005;
   Makris et al. 2006; Goldstein et al. 2007.
   Labels used: cerebral white matter + cortex (left/right), lateral ventricles, thalamus, caudate,
   putamen, pallidum, accumbens, hippocampus, amygdala, brain-stem.

2. Diedrichsen 2009 probabilistic atlas of the human cerebellum (anatomical lobules, MNI)
   Retrieved from: https://github.com/DiedrichsenLab/cerebellar_atlases
     Diedrichsen_2009/atl-Anatom_space-MNI_dseg.nii
   Licence: CC BY (per repository README). Cite: Diedrichsen J, Balsters JH, Flavell J, Cussans E,
   Ramnani N. A probabilistic MR atlas of the human cerebellum. NeuroImage 2009;46:39-46.
   Used as a single cerebellum mask (all lobules).

Method (tools/pipeline/register.py and labelbrain.py in the project folder):
 - Patient brain mask from CTA: median-filtered 14-75 HU tissue, bone excluded, largest component,
   closed and hole-filled.
 - 12-parameter affine (translation, rotation, scale, shear) registration of the atlas brain mask
   to the patient mask, maximising soft Dice within the CTA field of view (Powell, 4 starts).
   Result: Dice 0.930 (case 002), 0.909 (case 008).
 - Each patient brain voxel takes the registered atlas label (nearest labelled tissue where the
   atlas and patient masks differ). Ventricles = patient CSF (<16 HU) within the registered
   ventricle region, plus the eroded registered atlas ventricles.
 - Surfaces: marching cubes, Taubin smoothing and quadric decimation.

The registered parcellation is approximate (affine only) and is for educational orientation.
It is not a clinical segmentation.