Changes in version 0.0.2.9001                      

Base-R plot() (breaking)

plot.ggseg_atlas() is reimplemented with base graphics
(graphics::polygon() / graphics::polypath()), and ggplot2 is dropped
from Imports — the package no longer depends on ggplot2 for its own
plotting.

  - plot() now returns the atlas invisibly rather than a ggplot object;
    it is called for its side effect. Each spatially separate piece
    (e.g. a hemisphere surface or a slice) is drawn in its own panel,
    arranged in a near-square grid for a legible overview of the atlas.
    Code that captured the return value to add ggplot2 layers
    (plot(atlas) + ...) must be updated.
  - The show.legend argument is removed; the base-R plot draws no
    legend. Extra arguments in ... are forwarded to the underlying
    polygon() / polypath() primitives (e.g. lwd, border).
  - vdiffr is dropped from Suggests; the plot tests no longer snapshot
    SVG.

Lighter dependency tree

  - Dropped the dplyr and tidyr Imports in favour of base-R equivalents,
    shrinking the recursive dependency tree from 32 to 20 packages (also
    removes tibble, pillar, purrr, stringi, stringr, tidyselect,
    generics, magrittr and more). Returned data objects keep the
    tbl_df/tbl classes so they continue to integrate with tibble/dplyr
    workflows, but tibble is no longer required at install time.
  - print() for a ggseg_atlas now shows the first 10 core rows by
    default (atlases can have hundreds of regions); pass n to control
    how many rows print, e.g. print(dk(), n = 50).

Bundled SUIT cerebellar atlas

  - New suit() bundled atlas — the SUIT cerebellar parcellation (lobules
    + deep nuclei) from ggsegSUIT, stored in the sf-optional polygon
    (geom) format with 3D vertices (lobules) and meshes (nuclei).
    ggseg.formats now ships one atlas of each kind: dk() (cortical),
    aseg() (subcortical), tracula() (tract), suit() (cerebellar).

Unified geom slot (breaking)

Atlas 2D geometry now lives in a single atlas$data$geom slot whose class
(sf or brain_polygons) determines the rendering path. The parallel sf
and polygons slots are gone — conversion between the two is lossless, so
only one representation is ever stored.

  - New accessors: atlas_geom(), atlas_polygons(),
    atlas_geometry_type(), is_atlas_sf(), is_atlas_polygon(). atlas_sf()
    now converts from the polygon representation when needed and is the
    single interception point for ggseg plotting. atlas_geom() falls
    back to a legacy sf slot, so atlases built before this change keep
    working. Reverse dependencies should call these accessors rather
    than reaching into atlas$data.
  - ggseg_data_cortical() / ggseg_data_subcortical() /
    ggseg_data_cerebellar() / ggseg_data_tract() now take a single geom
    argument. A released sf argument is still accepted via ...
    (converted to polygons via sf_to_polygons()) with a deprecation
    warning.
  - as_polygon_atlas() / as_sf_atlas() set the single geom slot.
  - migrate_atlas_files() rewrites atlases to the single geom slot
    (polygons by default; keep_sf = TRUE stores sf).

sf-optional atlas format

Foundation work for the sf-optional milestone — see
ggsegverse/ggseg.formats#4.

  - New brain_polygons representation: a nested tibble keyed by label,
    with a geometry list-column containing per-view, per-ring point
    coordinates (view, x, y, group, subgroup). Renderable directly by
    geom_polygon() via the subgroup aesthetic (which handles holes
    through grid::pathGrob even-odd fill).
  - Geometry round-trips between sf and brain_polygons losslessly. The
    sf-side conversion uses sfheaders (pure Rcpp, no GDAL/GEOS/PROJ
    system libraries), enabling wasm builds and air-gapped installation
    paths. The low-level converters are internal; the public API is the
    atlas-level as_sf_atlas() / as_polygon_atlas() and the atlas_sf() /
    atlas_polygons() accessors.
  - ggseg_data_cortical(), ggseg_data_subcortical(),
    ggseg_data_cerebellar(), and ggseg_data_tract() now accept a
    polygons = argument alongside sf =. When only sf is supplied, the
    polygons slot is derived automatically; the two slots are kept in
    sync so existing callers see no change.
  - as_polygon_atlas() and as_sf_atlas() convert between the sf-backed
    and polygon-only forms at the atlas level.
  - migrate_atlas_files() walks a package's data/ directory and rewrites
    every ggseg_atlas .rda to the polygon format. Intended for
    downstream atlas-package maintainers across the ggsegverse
    ecosystem.
  - validate_data_labels() checks 2D label coverage against whichever 2D
    source is present (sf or polygons), preserving the same 80%/90%
    thresholds.

sfheaders joins Imports. sf moves from Imports to Suggests. The package
can now be installed without GDAL / GEOS / PROJ system libraries —
enabling wasm builds and air-gapped installs. Functions that genuinely
need sf (e.g. validate_sf(), as.data.frame.ggseg_atlas(),
plot.ggseg_atlas(), the atlas_view_* repositioning helpers) check
requireNamespace("sf") at entry and error with a clear pointer to
as_polygon_atlas() if sf is unavailable. The bundled dk, aseg, and
tracula atlases still carry their sf slots, so callers who have sf
installed see no behavioural change.

Region geometry operations

  - New atlas_region_op() combines two sets of region geometry with a
    boolean operation per view (difference, intersection, union,
    symdifference), writing the result to a new region. Boolean ops need
    a geometry engine, so this helper always requires sf; a polygon-only
    atlas is rehydrated for the operation and the result returned in
    polygon form.
  - atlas_region_contextual() now operates on whichever 2D
    representation an atlas carries (sf and/or polygons) and keeps both
    in sync — it needs no sf for a polygon-only atlas. It also gains an
    ignore.case argument.
  - The atlas manipulation helpers no longer leave a stale polygons slot
    behind a freshly rewritten sf slot; the two 2D representations stay
    consistent after every operation.

sf-free view manipulation

  - atlas_context_remove(), atlas_view_remove(), atlas_view_keep(),
    atlas_view_remove_region(), atlas_view_remove_small(),
    atlas_view_gather(), and atlas_view_reorder() now run on
    polygon-only atlases with no sf installed. Filtering, polygon area
    (shoelace), and view repositioning are implemented in pure R against
    the brain_polygons coordinate table; the polygon results match the
    sf path to floating-point precision. sf-backed atlases continue to
    use the existing sf code path unchanged.
  - atlas_views() reads view names from polygons when sf is absent.
  - atlas_region_keep() and atlas_region_remove() no longer drop 2D
    geometry on polygon-only atlases (they previously rebuilt from the
    sf slot only).
  - View helpers now warn about "no 2D geometry" (rather than "no sf
    data") only when an atlas carries neither sf nor polygons.

                 Changes in version 0.0.2 (2026-04-03)                  

Deep cerebellar nuclei support

  - ggseg_data_cerebellar() gains an optional meshes parameter for deep
    cerebellar structures (e.g. dentate, interposed, fastigial nuclei)
    that are not on the SUIT cortical surface. Surface regions use
    vertices (shared SUIT mesh), deep structures use individual meshes
    (like subcortical atlases).
  - Validation now checks the union of vertices + meshes labels against
    core when both are present, rather than requiring each to cover all
    labels independently.
  - rebuild_atlas_data() preserves cerebellar data type and handles
    mixed vertices + meshes correctly.

Bug fixes

  - reposition_views() now handles sfc_GEOMETRY (mixed geometry types)
    by casting to MULTIPOLYGON before coordinate operations.
  - atlas_view_gather() is more robust against non-sf or empty sf data,
    preventing errors in subcortical and tract pipelines.

                 Changes in version 0.0.1 (2026-02-18)                  

Initial CRAN release. Extracts and formalises the atlas data structures
that were previously embedded in ggseg and ggseg3d.

Unified ggseg_atlas S3 class

  - ggseg_atlas() constructor with typed data containers for cortical,
    subcortical, tract, and cerebellar atlases.
  - Type-checking predicates: is_ggseg_atlas(), is_cortical_atlas(),
    is_subcortical_atlas(), is_tract_atlas(), is_cerebellar_atlas().
  - Coercion with as_ggseg_atlas(), as.data.frame(), and as.list()
    methods.
  - plot() method for quick atlas visualisation via ggplot2.

Accessors

  - atlas_type(), atlas_regions(), atlas_labels(), atlas_palette(),
    atlas_sf(), atlas_vertices(), atlas_meshes(), and atlas_views() for
    querying atlas contents without reaching into slots.

Atlas manipulation

  - Pipe-friendly region operations: atlas_region_keep(),
    atlas_region_remove(), atlas_region_rename(),
    atlas_region_contextual().
  - Metadata enrichment with atlas_core_add().
  - View management: atlas_view_keep(), atlas_view_remove(),
    atlas_view_remove_region(), atlas_view_remove_small(),
    atlas_view_gather(), atlas_view_reorder().

Bundled atlases

  - Ships three ready-to-use atlases: dk() (Desikan-Killiany cortical),
    aseg() (FreeSurfer subcortical), and tracula() (white matter
    tracts).
  - get_brain_mesh() and get_cerebellar_mesh() provide 3D surface meshes
    for rendering.

Legacy conversion

  - convert_legacy_brain_atlas() and unify_legacy_atlases() bridge old
    ggseg/ggseg3d atlas objects to the unified format.
  - Deprecated wrappers (brain_atlas(), brain_regions(), etc.) ease
    migration from the old API.

FreeSurfer I/O

  - read_freesurfer_stats(), read_atlas_files(), and
    read_freesurfer_table() for reading FreeSurfer statistics into R.