scitomo/fibsem-microalgae-atlas3d-esb

Microalgae FIB-SEM Volume — Atlas3D ESB

Summary

This dataset is a three-dimensional focused ion beam–scanning electron microscopy (FIB-SEM) acquisition of microalgae. The sample type was confirmed by the acquisition contributor, Pierre-Henri Jouneau (CEA/DRF/IRIG/DEPHY/MEM/LEMMA). The species, specimen preparation and biological origin have not been established in the information available for this release; they should not be inferred from the images alone.

The published primary image stack consists of 1,667 successive ESB images of 10,048 × 10,048 pixels, stored as a multiscale, losslessly compressed OME-NGFF 0.5 / Zarr v3 volume. Each source image is a 16-bit TIFF with a nominal lateral sampling of 4 nm × 4 nm (field of view 40.192 µm × 40.192 µm). The source pixel values are preserved exactly at pyramid level 0, with no spatial registration, gain correction, filtering, normalization or denoising performed by the conversion.

FIB-SEM collects a sequence of images of freshly milled specimen surfaces. This is serial surface imaging, not projection-based electron tomography. The images can be used for volume-microscopy analysis, restoration research and as a reference object for synthetic tomographic experiments, with the modality differences stated explicitly.

Acquisition and contributors

Field	Value
Specimen	Microalgae; species undetermined
Acquisition contributor	Pierre-Henri Jouneau
Affiliation	CEA/DRF/IRIG/DEPHY/MEM/LEMMA
Instrument	ZEISS Crossbeam 550
Acquisition system	Fibics Atlas3D / AtlasEngine
Detector	ESB
Modality	FIB-SEM serial surface imaging
Primary stack	1,667 images, source acquisition indices 161–1827
Voxel-array shape (ZYX)	1667 × 10048 × 10048
Pixel data type	uint16
Nominal XY sampling	4 nm × 4 nm
Nominal Z increment	4 nm (see Z-calibration caveat)
Specimen preparation, fixation and embedding	Not documented for this release

The original archive also includes 160 earlier images in an acquisition directory explicitly named ESB/Bad, indexed 1–160, and eight auxiliary keyframe/overview TIFFs. These are not included in the released primary Zarr volume. The excluded-image inventory is provided without redistributing their pixels. Assignment to the Bad group is inherited from the acquisition folder, not an independent quality diagnosis.

Files and format

volume.ome.zarr/             OME-NGFF 0.5 / Zarr v3 multiscale uint16 volume
  0/                        Native-resolution, pixel-exact primary image stack
  1/                        XY downsampled 2×
  2/                        XY downsampled 4×
  3/                        XY downsampled 8×
metadata/
  acquisition.json          Acquisition/geometry summary, including Z caveats
  conversion.json           Software versions, converter fingerprint and settings
  slices.csv                Slice-index correspondence, Z records and hashes
  excluded_bad.csv          Inventory of excluded acquisition slices
  qc_preview.json           Preview information, if generated
preview/                    Orthogonal-plane images and optional HTML QC preview

Arrays use ZYX dimension order. All four scales preserve the original number of Z slices. Levels 1–3 are derived from repeated 2 × 2 XY integer block means with half-up rounding; level 0 contains the original decoded uint16 pixels. The Zarr v3 store uses sharding and lossless Blosc/Zstandard compression. See metadata/conversion.json for precise implementation settings.

metadata/slices.csv records the mapping between zero-based volume_index and the original acquisition z_index, as well as source-relative filenames, vendor-reported Z coordinates and per-image checksums. The original source TIFF files are not part of this distribution.

Spatial calibration and known limitations

The Z axis is nominally uniform, but the vendor-reported Z coordinates are not. The OME-Zarr scale metadata express a regular 4 nm nominal Z-index interval, relative to the first retained image. They must not be interpreted as an exact measured physical Z-coordinate system. For a measurement-sensitive analysis, consult metadata/slices.csv for each image's recorded z_pos_um and assess the acquisition coordinate calibration independently.

The first and last retained images have reported vendor Z coordinates of approximately 0.723810 µm and 7.108759 µm, a coordinate span of approximately 6.385 µm. The median increment between consecutively indexed primary images is about 3.842 nm; this is an observation about the reported coordinates, not proof of the actual physical slice thickness.

Three unusual adjacent-image recorded Z increments were detected in metadata screening:

• Acquisition indices 1053 → 1054: about −0.175 nm.
• Acquisition indices 1097 → 1098: about +7.860 nm.
• Acquisition indices 1372 → 1373: about −0.028 nm.

No source acquisition indices are missing or duplicated in the retained 161–1827 sequence. The conversion does not establish stage/world orientation, physical Z calibration or image registration.

Quality control

• The dataset preparer reports exhaustive verification of all 1,667 decoded source TIFF images against the OME-Zarr data, including the three derived multiscale levels. This verifies that conversion preserved source pixels and implemented the intended downsampling; it does not independently assess specimen integrity.
• Visual inspection of the generated orthogonal preview images found clearly resolved internal cellular structures and no obvious major image-quality problem. This is a qualitative observation, not a quantified guarantee of image alignment, milling regularity or artifact absence.
• The irregular vendor-reported Z increments remain documented rather than corrected or silently resampled.

Suggested uses

This volume may be useful for 3D electron-microscopy visualization, microscopy restoration and denoising, structure-preserving representation learning, 3D segmentation research (without provided labels), and construction of explicitly synthetic forward-model tests for computational imaging and tomography.

It is not a cryo-electron tomography tilt series and is not experimental cryo-ET ground truth. The image formation, contrast and artifact characteristics of FIB-SEM and cryo-ET differ substantially; simulation results from this dataset should not be represented as experimental cryo-ET results.

Rights, attribution and citation

Rights holder and acquisition contributor: Pierre-Henri Jouneau — CEA/DRF/IRIG/DEPHY/MEM/LEMMA.

Data dedication: Creative Commons CC0 1.0 Universal. The acquisition contributor and reported rights holder, Pierre-Henri Jouneau, verbally authorized redistribution of this acquisition under CC0 1.0 Universal, including commercial reuse, as confirmed to the dataset publisher on 9 October 2026.

Under CC0, attribution is not legally mandatory. For scientific traceability, users are nevertheless asked to acknowledge Pierre-Henri Jouneau for the acquisition and to cite this dataset's Hugging Face repository (and, when available, its EMPIAR deposition) in publications that substantially use the data. This request does not impose an additional legal restriction on reuse.

An EMPIAR deposition is planned by the acquisition contributor; no EMPIAR accession number is available for this release. This card will be updated with the accession if and when the record becomes public.

Reproducibility

The conversion metadata include converter version and checksum, library versions, compression/shard/chunk configuration, and acquisition slice correspondence. The original TIFF file hashes and decoded pixel hashes are recorded in metadata/slices.csv unless source-file hashing was explicitly disabled during conversion; inspect metadata/conversion.json for the recorded setting.

The primary TIFF pixel values can be independently compared with OME-Zarr level 0; the higher-resolution pyramid levels are deterministic integer XY reductions. The conversion software and verification procedure are described in the accompanying Scitomo/Atlas3D converter documentation.