SegRef3D software remains licensed under Apache-2.0. The following data licenses apply only to the respective demo image assets and their adaptations, not to SegRef3D software. The data providers do not endorse SegRef3D.
Source: The Cancer Imaging Archive (TCIA), Pancreas-CT (Version 2). One complete contrast-enhanced abdominal CT series, PANCREAS_0080, is included: 181 axial DICOM images, 512 × 512, 43,011,465-byte ZIP. Manual segmentation is not included.
Dataset citation: Roth H, Farag A, Turkbey EB, Lu L, Liu J, Summers RM. Data From Pancreas-CT (Version 2). The Cancer Imaging Archive, 2016. DOI: 10.7937/K9/TCIA.2016.tNB1kqBU.
Collection · CC BY 3.0 · TCIA Data Usage Policy · Original TCIA license and policy notice · Asset manifest.
Retain attribution, DOI, license and the TCIA data usage policy link when using or redistributing these data; require downstream users to retain them too. Follow TCIA restrictions, including no re-identification or contact of participants.
The official single-series ZIP, including LICENSE, is retained byte-for-byte. No slices were removed, resampled, cropped or re-encoded. TCIA describes these DICOMs as created from anonymized Analyze/NIfTI volumes, rather than original scanner exports. The supplied geometry gives X/Y 0.9765625 mm and Z 1 mm (position-derived); these values are preserved without substituting collection-level typical slice thickness. In the demo, screen down is posterior (back) and slices run caudal to cranial (LPS Z −180 to 0 mm). Decoded rows and slice order are reversed, with the affine updated consistently for viewing, masks and exports. This is a lossless voxel permutation, with no interpolation or change to the downloaded DICOM files.
The initial abdominal Window/Level (400/40) is a display-only preset; stored voxels and source metadata are unchanged. Images download from this site only on selection and are processed locally. SegCT/MRI request generation is local; Colab upload remains an explicit separate user action.
Series UID: 1.2.826.0.1.3680043.2.1125.1.41202274843063370955090296887703130.
Archive SHA-256: 6b9bff4c2a7bb77b564a13fb2fdcfa5a3afea25455ab066309fb5fce7c9a8065.
All 181 DICOM files decode; SOP Instance UIDs are unique and slice positions form a
continuous 1 mm interval. The manifest records per-file hashes and positions.
Peddie CP, Jones ML, Collinson LM. Cropped regions from Serial Block Face SEM of HeLa cell pellet with 10 nm pixels and 50 nm slices (benchmark dataset), EMPIAR-10478. Source · CC0 · EMPIAR licensing policy.
Only ROI_1416-1932-171 is used. 150 slices are sampled every other frame from 300 slices,
retaining source frames 1, 3, 5, … 299 (including the first image). Images were resized
from 2000 x 2000 to 512 x 512 with LANCZOS and encoded as optimized 8-bit grayscale PNG.
No additional cropping, alignment, or contrast adjustment. Sequential output filenames:
hela_0001.png–hela_0150.png. Effective Z spacing is 100 nm, twice the original 50 nm;
X/Y spacing remains 39.0625 nm. Image bytes: 25,418,095 (25.4 MB).
The manifest records the original frame number for every output image.
The SegRef3D maintainer supplied this BAP-derived set via Dropbox. The supplied archive contains only 132 PNG images and no README, dataset ID, brain ID, experiment ID, or physical spacing metadata. No more specific identifier or stain is assigned here.
Modifications: all 132 supplied images are retained at 707 × 553 pixels and 8 bits per RGB channel. Only their fully opaque alpha channel was removed, followed by optimized, lossless PNG compression. Every RGB pixel was verified unchanged after encoding. No downsampling, cropping, rotation, registration, contrast adjustment, or additional slice selection was performed. Image bytes decrease from 62,713,760 to 51,472,513 (51.5 MB). These BAP-derived assets, including the optimized images, remain CC BY-SA 4.0.
Order and limitations: original names image0001.png, image0003.png, …,
image0263.png are preserved in numeric order. The supplied set already omits all even
numbers; there are no additional gaps in this step-2 sequence. Natural and lexicographic
sorting agree for these zero-padded names. Contact sheets of all images show progressive
anatomy without an obvious reversal or abrupt ordering jump. Original section shifts,
tissue tears and slide artifacts remain. This review does not establish that every original
section is present. Source physical spacing metadata were not supplied. The demo starts
at slice 55 (image0109.png), visually selected for broad, clear left/right brain margins.
X/Y are calibrated by the user against an approximate 11.4 mm adult mouse brain reference
width, not a measurement of this BAP specimen. Z is an estimated 0.10 mm (100 µm) effective
interval: approximate 13.2 mm whole-brain AP coverage / 132 supplied images. It is not
source metadata or original histological section thickness. Physical volume remains unavailable
until XY calibration, and subsequent volumes are approximate. Register sections where necessary.
Scientific citation (BAP): Bohland JW et al. (2009). A Proposal for a Coordinated Effort for the Determination of Brainwide Neuroanatomical Connectivity in Model Organisms at a Mesoscopic Scale. PLoS Computational Biology 5(3): e1000334. https://doi.org/10.1371/journal.pcbi.1000334. This is the project’s general citation, not an invented identifier for this particular image set.
The HeLa 150-frame demo is reused byte for byte from the previously prepared SegRef3D demo. To regenerate the mouse demo from the supplied archive (Python + Pillow):
python lite-web/scripts/prepare-mouse-brain-demo.py path/to/source.zip
node --test "lite-web/tests/*.test.mjs"
Original archives are not shipped. Images load only after a user selects a demo and can then be cached by the browser. Microscopy demos start with empty masks each time they are selected; export your work before switching. User-selected local images continue to be processed locally without upload during normal SegRef3D Lite use.
Their existing CC BY 4.0 credits, preparation details and citations are unchanged; see the Lite README.