ACTA-DIRECT Dataset v2 — micro-CT tooth scans

Source: Gonzalez Valenzuela, Ricardo Eugenio (creator), with Erwin Berkhout,
Henk Marquering, Pascal Mettes and Bruno Loos (supervisors). ACTA-DIRECT
Dataset v2. Vrije Universiteit Amsterdam, 2025.
https://doi.org/10.48338/VU01-H5ALYJ
Names, roles, publisher and year as registered in the DOI's DataCite record
(checked 2026-10-12); the record gives the creator's affiliation as Academic
Center for Dentistry Amsterdam.
Licence stated by the deposit: Creative Commons Attribution 4.0 International
(CC BY 4.0), https://creativecommons.org/licenses/by/4.0/
Acquisition (dataset paper): Gonzalez Valenzuela RE et al., BMC Oral Health
2024, https://doi.org/10.1186/s12903-024-05076-x — SCANCO MicroCT40, 72
micrometre voxels.

Files used (original/1_Micro-CT_Scans/), each one extracted, anonymised tooth:
  0164_tooth.nii (SHA-256 a3ce5bcde49e650d9ac3bf86ed503f06e530645486dda13eb2d5f89399967cb0)
    Universal #15; #2 is its explicit mirror.
  0158_tooth.nii (SHA-256 ca8bcc6895f3470210b134ba2d019289ef35fd65401c77582f588a92f6d5bc19)
    Universal #3; #14 is its explicit mirror.
  0059_tooth.nii (SHA-256 c85b81696ba76c5a31205e392e4f2a663705207e84bda309abeec0cd54c1e868)
    Universal #32; #17 is its explicit mirror.
  0056_tooth.nii (SHA-256 ee39089b25266ed610d344a91e9f601cf3fe2e68d326b66e31a4099566225450)
    Universal #18; #31 is its explicit mirror.
  0217_tooth.nii (SHA-256 8537ae810354ee8b243d4c4b83672f6c13c0bd7488c8c72098ca7e2454744d92)
    Universal #21; #28 is its explicit mirror.
  0039_tooth.nii (SHA-256 4bc58a8bc20951d429cf6df12213c23a3c8be62c75171e2c22732fc1c0b75fe1)
    Universal #30; #19 is its explicit mirror (added 2026-10-15).

The dataset records no tooth type, number or side. ToothForm assigned each
tooth's class and side from our in-house morphology review (three independent
visual analyses plus a morphometric script, then a consensus; for 0039, crown,
root and canal morphometrics with CT sections, then the owner's review of the
rendered views); this is an inference, not a source label and not
expert-confirmed. The side is the side
as imaged, because the scans' image handedness is not documented. Each mirrored
position is a reflected copy of the same tooth with reversed triangle winding,
not another specimen. The teeth are six different specimens, not a matched set.

Changes made by ToothForm: enamel, dentin and pulp were segmented automatically
from the published volumes on their native 72 micrometre grid (intensity
thresholds, seeded watershed and partial-volume boundary labelling). Pulp
includes canal stretches traced only where the voxels are measurably darker
than the surrounding dentin (for 0059, sub-voxel apical canals are drawn at a
0.216 mm minimum width; for 0056, 6,587 scan-hard voxels were relabelled as
pulp this way); tiny enclosed voids were filled; small label refinements are
listed per tooth in the app's model notes. Labels were
converted to millimetre surface meshes, placed by one proper rotation and
translation per tooth, simplified with meshoptimizer for display tiers, and
given smoothed display normals (shading only). For tooth 0056 the surfaces
were also smoothed with every vertex kept within half a voxel. No crown repair,
surface patching, idealised anatomy or nonuniform scaling was added. The labels
are not expert annotations and make no clinical claim. Tissue colours and
translucency are illustrative display colours.

Full provenance and reproduction: Research/ACTA_DIRECT_PROVENANCE.md.