NEWS

---

ggsegJHU 3.0.0

Breaking changes

• jhu() is removed. It was an 11-region unified atlas that collapsed the tracts and merged the hemispheres. jhu_tracts() covers the same anatomy at full resolution (20 tracts, hemispheres separate) and jhu_wm() carries the 48-region ICBM-DTI-81 parcellation, so jhu() only ever offered a coarser view of data both accessors already provide.
• jhu_tracts() is now a tract atlas (atlas_type() returns "tract", and the object is a tract_atlas) rather than a subcortical one. Each tract carries a principal-curve centerline, drawn as a 3D tube for ggseg3d() and as a 2D projection over an anatomical silhouette. Code that dispatched on subcortical_atlas will no longer match it.
• The uncinate fasciculus and the temporal part of the superior longitudinal fasciculus were swapped in previous releases. FSL orders the uncinate (volume indices 17, 18) before the temporal SLF (19, 20), and the metadata assumed the reverse, so both tracts carried the other's name. Verified against the volumes: labels 17/18 sit at MNI (±35, +7, −15), the limen insulae, which is uncinate; 19/20 at (±43, −47, +4), posterior temporal, which is SLF-temporal. Any analysis that selected uncf or sltf from an earlier release selected the wrong tract.

Pipeline improvements

• Tracts are now built from JHU's probabilistic maps rather than the distributed maxprob-thr25 volume. That volume keeps only the single highest-probability tract at each voxel, so tracts sharing a corridor erode each other: the left temporal SLF survived as 76 voxels of a bundle spanning several centimetres, far too little to fit a meaningful centerline. Each tract is now thresholded at 25% of its own maximum and merged, raising the labelled volume roughly four-fold.
• Overlaps are resolved on each tract's probability relative to its own peak rather than on the raw value. JHU's per-tract maxima range from 43% to 97%, and raw comparison hands every shared corridor to whichever tract is more reliably reconstructed rather than to whichever the voxel belongs to. Worst-case retention rises from 17% to 37% of a tract's core, and the total labelled volume is unchanged — only the assignment differs.
• 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 +30, +8 and −14 mm, coronal at −18 mm and sagittal at −30 mm, each projected over a ±8-slice slab. Tracts are projections through the slab; the silhouette framing them is a true slice, taken from wherever in the slab holds the most cortex.
• 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. Left and right stay together, and every tract is drawn in at least one view.
• Tracts are coloured per bundle, with left and right sharing a hue, giving 10 distinguishable colours. Earlier releases coloured by the coarser group (projection/association/limbic/commissure), leaving only four colours for 20 tracts. FSL ships no colour table for this atlas, so the palette is chosen rather than inherited.

jhu_wm()

jhu_wm() stays a subcortical atlas — the ICBM-DTI-81 labels are white-matter parcels rather than pathways — but its build was broken and is repaired:

• The right superior longitudinal fasciculus was missing entirely, and the other 47 regions were eroded. The labels were projected with protect_cortex = TRUE, which shields cortex and cerebral white matter from being overwritten. That guard suits the deep grey structures the subcortical pipeline was built around, and is exactly wrong here: it protected a white-matter atlas out of the only tissue it describes. The projection now runs with the guard off, and all 48 regions survive.
• Regions are coloured per structure, with left and right sharing a hue, for 27 distinguishable colours. Previously all 48 shared five group colours.
• The FreeSurfer anatomy the labels are projected into is now correctly kept as grey context rather than promoted into the atlas. It had been leaking into core and the palette, so atlas_regions() reported cortical parcels as though they were JHU white-matter labels.
• The sagittal views read properly. A sagittal cut runs along the cortical ribbon rather than across it, so the thin slabs used for axial and coronal caught the ribbon only where it happened to sit in-plane and broke the outline into fragments; the sagittal slabs are now wider. The silhouette itself is a single slice, so widening alone could not fill it — ggseg.extra now picks that slice by cortex content, which moves the mid-sagittal reference off the interhemispheric fissure (x=134 rather than 128) and cuts the medial wall as a ribbon instead of grazing it.
• Region outlines are smoothed, and pieces below 20 px² are dropped, so the parcels read as shapes rather than as rasterised voxel staircases.
• The build script no longer aborts against current ggseg.extra: it used project_volume_anatomical()'s removed output_file return field, plus the deprecated views argument and write_ctab().

ggsegJHU 2.1.0

Breaking changes

• icbm() has been removed. It was a legacy variant of the ICBM-DTI-81 white matter atlas with no 2D geometry, undivided hemispheres, and no group column. Use jhu_wm() instead, which covers the same anatomy with both 2D and 3D geometry.
• jhu_labels() (introduced in 2.0.0) is now jhu_wm().

Pipeline improvements

• Both atlases are now resampled onto the FreeSurfer cvs_avg35_inMNI152 aparc+aseg grid before the 2D pipeline runs. The JHU NIfTIs and cvs_avg35_inMNI152 are both already in MNI152 1mm space, so resampling uses the existing xforms via --regheader (no mri_coreg step, which would binarise the white-matter labels into a non-brain mask and distort the fit). This produces:
  • smoother region boundaries (trilinear resampling + argmax replaces the voxel-grid jaggedness of the raw 1mm masks);
  • a true cortical-reference outline behind the white-matter regions on every 2D slice, anatomically aligned with the labels.
• Custom view ranges replace the default subcortical views so brainstem and lateral structures are not clipped.
• The core table now carries a group column for both atlases.

ggsegJHU 2.0.0

Breaking changes

• jhu and jhu_3d data objects have been removed
• Use jhu_tracts() and jhu_wm() accessor functions instead
• ggseg.formats is now a hard dependency

Minor changes

• Added icbm() atlas (formerly in ggsegICBM package)
• Package URLs updated to ggsegverse GitHub organisation

ggsegJHU 1.0.01

• Adapted atlases to ggseg <= 1.6.0