build 2026-09-04 18:30
Or paste x, y, z into the x field — it splits automatically.
Voxels here are 4 × 4 × 40 nm — identical to µJump's minnie65, so a coordinate means the same thing in both tools and nothing needs converting.
Serial-section TEM of mouse primary visual cortex at 4 × 4 × 40 nm (Lee, Bonin, Reed, Graham, Hood, Glattfelder & Reid, Nature 532, 370–374 (2016); released through neurodata.io under the Open Data Commons Attribution Licence).
What this volume has that no other tool here does: the leptomeninges. A continuous sheet of flattened meningeal cells runs across the whole pial surface, with the subarachnoid space above it and a subpial vessel below, imaged at 4 nm. µJump, δJump and ηJump all start below the pia; βJump is hippocampus. The guided identification's pia / arachnoid branch exists for this dataset.
This dataset publishes no segmentation and no cell types. There is no classifier to agree or disagree with, so here the community is the classifier — and there are no meshes, no 3D models, no PowerPoint export and no connectivity, because none of those can exist without a segmentation. What you get instead is the image, a nucleus table, and the decision tree.
Four things to know before you trust a number. The public release is a ~33 µm slab of a thicker imaged block — quote 33, never 150 — so about 27 % of nuclei are cut by a face and their volumes are underestimates (they are tagged). Nuclei come from automatic detection, so positions are imperfect — correct them from the panel. The detector is size-tuned near 7.5 µm and finds roughly 46 % of the cells a direct count of mouse cortex predicts (1.6 × 105 nuclei/mm³), with the shortfall concentrated in small glial nuclei — so if you can see one it missed, add it. And rows tagged dense nucleoplasm are about one in three not a nucleus at all: in a pial vessel an erythrocyte is dense, oval, 6–8 µm across and as textured as nucleoplasm, and nothing the detector measures separates the two. Those are the rows most worth your attention.
Depth is measured, not fitted. The pial surface is inside this volume and was traced directly, so every nucleus carries a real distance below it. Cells above the pia sit in the leptomeninges and carry a negative depth. The layer boundaries themselves are standard mouse V1 values, not measured here.
Every filter except community identity and organelles runs against the 488 nuclei embedded in this page — no network needed to preview or export. Identity and organelles read the community backend, since neither can live in a static page.
Each box has its own “Colab notebook” button, which
writes a .ipynb you upload to Google Colab and run to download that box’s EM
imagery. Nothing is downloaded from this page — the notebook does the fetching.
Lee16 publishes no segmentation, so those two tickboxes are shown but disabled.
The dataset. Lee, Bonin, Reed, Graham, Hood,
Glattfelder & Reid, “Anatomy and function of an excitatory network in the visual
cortex” — Nature 532, 370–374 (2016),
PMID 27018655.
Public release at neurodata.io/data/lee16,
served from s3://open-neurodata/lee/lee16/image under the
Open Data Commons Attribution Licence v1.0
— ODC-By, which is not the same instrument as CC-BY and carries its own attribution
requirement. That release is image only: no segmentation, no meshes, no soma table, no cell
types. Every nucleus in this tool was detected by λJump's own notebook, not published by
the authors.
The leptomeningeal branch. The pia / arachnoid cell types in the guided tree come from Pietilä et al., “Molecular anatomy of adult mouse leptomeninges”, Nature (2023). Worth reading before using them: the pial, pial-sheath, inner-arachnoid and reticular fibroblasts share one fibroblast ultrastructure and are separated mainly by location, not by any single feature visible in one section. Only the arachnoid barrier cell has a real single-section EM signature — continuous tight junctions. The tree says so at each step rather than implying a confidence the images cannot support. The dural border cell is excluded here entirely: the dura was never imaged in this volume.
Cell classification from ultrastructure. Elabbady et al., “Perisomatic ultrastructure efficiently classifies cells in mouse cortex,” Nature 640, 478–486 (2025).
Brain ultrastructure overview: “Brain Ultrastructure: Putting the Pieces Together,” PMC7930431.
Neurovascular volume EM: “Public Volume Electron Microscopy Data: An Essential Resource to Study the Brain Microvasculature,” Frontiers in Cell and Developmental Biology 10, 849469 (2022).
EM ultrastructure teaching atlas (organelle and immune-cell morphology — not a peer-reviewed source; use it to learn to recognise structures, not to cite a claim): Jastrow’s EM Atlas.
How many cells should be here. The density this tool reports is about 46 % of the 1.60 ± 0.43 × 105 nuclei per mm³ counted directly across all cell types in mouse cortex (“Correlations of Neuronal and Microvascular Densities in Murine Cortex”, PMC4972024). The gap is not a mystery: the detector is size-tuned near 7.5 µm and glial nuclei run 5–7, so this table under-counts glia. It is measured, documented and checked on every run rather than left as a silent bias.
Read these before quoting a number. The slab is ~33 µm thick, so ~27 % of nuclei are cut by a face and their volumes are underestimates. 31 of the 822 sections are flagged as damaged and the 24 nuclei in or beside one are tagged. Detection is automatic and its dense-nucleoplasm pass runs at about 63 % precision because erythrocytes are indistinguishable from dense nuclei on every feature it measures — so do not quote a layer-1 or perivascular cell count from this tool until the community has reviewed those rows. That review is what λJump is for.
Charts built from the nuclei embedded in this page, plus one bulk read of the community’s current identifications (the same read the Filter tab’s identity dropdown uses). The charting library only loads once this tab is opened.
This dataset has no published cell types, so every chart is keyed on current community identity — a live consensus that can still change. The six most-identified names get their own bucket; anything else named folds into “Other”, and every nucleus nobody has classified yet is “Unclassified”.
The dense pass runs at about 63 % precision — roughly one dense row in three is an erythrocyte rather than a nucleus. A tall dense bar is where review is most worth doing, not a finding.
Detection is size-tuned near 7.5 µm, so the small end of every distribution here is thinner than the tissue really is — see the References tab.
For each identity (rows), what its single nearest neighbouring nucleus turns out to be (columns), as a percentage of that row. All 488 nuclei are used — no sampling needed at this size.
Each panel is one identity’s own distribution as a percentage of that identity, so a rarely-used identity is still readable next to a common one. Depth is measured from the traced pial surface; negative is above it.
Image data: Lee et al. (2016), released via neurodata.io under the Open Data Commons Attribution Licence v1.0 (ODC-By). Nucleus detections, depths and layer assignments are λJump’s own and are not part of that release.