atlastrack is rebuilt as a registered, anatomically-contextualised atlas,
and is now a tract atlas (atlas_type() returns "tract") carrying a
centerline per tract for 3D rendering.
Tracts are now built from AtlasTrack's probabilistic tract maps rather
than its distributed label volume. That volume is a winner-take-all
projection, so every tract was dilated far beyond its core and nested
subdivisions cannibalised their parents — R_SLF retained only 26% of its own
high-probability core, the rest claimed by R_pSLF. Rendering it painted
whole lobes rather than tracts. Each tract is now thresholded at 15% of its
own maximum probability and merged highest-probability-wins, cutting the
labelled volume roughly four-fold while keeping all 35 tracts.
AtlasTrack lives in its own template space, pitched ~24° from fsaverage, so the previous release rendered tracts spilling outside the brain outline. The tract cores are now affine-registered through the AtlasTrack T1 — a direct intensity T1→T1 alignment against fsaverage, rather than inferring the transform from the tract extent — and resampled into fsaverage space.
Each tract is reduced to a principal-curve centerline, drawn as a 3D tube and as a 2D projection over a grey fsaverage silhouette. Feeding the thresholded cores rather than the winner-take-all volume is what makes the curve fits well conditioned: a tract dilated across half a lobe has no meaningful principal axis.
2D projections are taken at five fixed anatomical positions, specified in RAS millimetres rather than voxel indices so they stay correct if the reference volume changes: axial at +17, +2 and −11 mm, coronal at −17 mm and sagittal at −12 mm, each projected over a ±8-slice slab. Tracts are projections through the slab; the silhouette framing them is a true slice at the midpoint. The sagittal position sits far enough laterally to cut the cortex as a gyral ribbon rather than tangentially through the medial wall, while staying medial enough to retain the brainstem the corticospinal tract descends into (the brainstem does not extend past −17 mm). The silhouette includes the brainstem, cerebellar cortex and deep grey alongside the cortical ribbon, so tracts leaving the cerebrum have anatomical context; white matter is left unfilled so the tracts remain visible against it.
The aggregate whole-brain fibre masks (AllFib*), the fornix-cut labels
(Fxcut) and the empty optic-radiation labels (OR) are dropped. CC is
dropped for the same reason: it is an aggregate rather than a tract, holding
its own forceps (Fmaj 96%, Fmin 88% inside it) so it drew the same fibres
twice, and being a sheet rather than a bundle its principal curve wandered
diagonally across both hemispheres instead of following a pathway. That
leaves 34 named tracts.
Each view draws only the tracts substantially represented in it, rather than every tract that happens to intersect the slab, which left each panel cluttered and each tract repeated. Tracts are compared per family so left and right stay together, and every tract is drawn in at least one view.
Requires ggseg.extra with the tract-projection orientation, centering, cortex-slice and atlas-name fixes; earlier versions mis-place tract projections relative to the anatomical reference.
brain_polygons format
(ggseg.formats 0.0.3). The atlases now render without sf and its
GDAL/GEOS/PROJ system libraries, enabling wasm and air-gapped installs.
Plots are unchanged.