Repository navigation
Projection and 3D
The printed atlas gives you one coronal plane at a time. These two views put the third axis back, so you can see how a structure runs through the brain.
Both are built from data already on the page — nothing extra is downloaded.
Every located label in the atlas, plotted as a scatter, with your selected structure highlighted in blue.
Every printed label in the atlas as one scatter, sagittal. The caudate putamen is picked out in blue — the run of labels the atlas prints for it, bregma +2.20 to −2.35. The four landmark rules are on, and the red guide marks where the plate viewer is sitting.
Two orientations:
| Button | Plot |
|---|---|
| sagittal · AP × DV | Side view — how deep a structure sits, and how far it runs front to back |
| top-down · AP × ML | From above — how far lateral it sits |
- Hover a dot to read its structure, plate and coordinates.
- Click a dot to jump to that structure's plate.
- Click the empty plot at an AP to scrub the plate viewer to that level.
- A vertical guide shows where the plate viewer is currently sitting.
- Landmarks adds bregma / lambda / interaural / occipital-crest reference rules; Skull outlines the skull around the cloud, widening the axes to fit it.
The caption under the plot tells you how many labels the selected structure has and on how many plates. If a structure has none, it says that instead of plotting nothing.
The 62 plates stacked where they actually sit — each section drawn at its own bregma. Needs WebGL 2. It builds once, the first time you open it.
| Mode | What it draws |
|---|---|
| Contours | The atlas's own drawn boundaries as a stack of sections |
| Volume | The same field, ray-marched as a solid |
| Labels | All 6,220 printed abbreviations as a stereotaxic point cloud |
The 62 drawn sections stacked where they sit, cut at the midline with Half. The blue dots are the selected structure's printed labels; the blue rectangle is the plate the viewer is on.
The same brain as Labels: all 6,220 printed abbreviations as a stereotaxic point cloud, with the selected structure's own labels in blue.
The stack is built from whichever source is selected above the plate — so you can stack the drawings, the Nissl sections or the myelin sections.
| Control | What it does |
|---|---|
| Density | How strongly each slice contributes — the 3D view's version of Contrast |
| Plates … – | Two sliders that clip the stack to a slab, anterior and posterior edge |
| Half | Cut at the midline to see inside |
| Ortho | Parallel projection — nothing is foreshortened, so equal lengths read equally at any depth. Use it when comparing positions rather than looking at the brain |
| Skull * | Wrap the stack in the CT skull surface, with a Bone opacity slider |
| Meshes | Draw structures as closed shapes instead of sections — see below |
| Surface | The interpolated brain surface as a faint shell. Only there once Meshes is on |
| Reset view | Back to the default viewpoint |
Everything above draws the atlas as sections — 62 of them, and the gaps between. Meshes draws it as shapes: one closed surface per structure, built offline from the same drawn extents the plate's outlines and colors come from, so a mesh and the outline on the plate under it are the same boundary read two ways.
Ticking Meshes reveals Surface beside it, and adds a Meshes group — Opacity and Color — to the folded Advanced half of the panel.
The meshes live in data/gerbil_atlas_volumes.json, about 20 MB. That is far too big to
carry in the page, so it is fetched the first time you tick the box and then kept — including
by the second pane of a Split, which reads the same cache.
- Served over
http://orhttps://, it is fetched for you. - Opened from disk as a
file://page it cannot be fetched at all, and a Load file button appears instead: point it atdata/gerbil_atlas_volumes.jsonin your own copy of the repository.
697 structures carry an area and get a mesh. The rest get nothing, and the note under the view says which of two reasons applies rather than leaving an unexplained blank: either the atlas names the structure but draws it no region on any plate, or it is a feature — a fissure, a vessel — that could never have one.
CPu as a mesh, standing in the contour stack. Two shapes and not one: nothing is cut at ML 0, so a bilateral structure comes out as the two components the atlas actually draws. The note under the view gives the volume — 49.87 mm³.
With the meshes on, the label cloud stands down. It would otherwise be a fog of dots duplicating shapes the mesh already draws. The exception is Labels mode, where the cloud is the rendering, so there both are shown.
Contours is the mode to pair meshes with. Volume re-marches the tissue in front of them and mostly buries them.
Two of the camera controls treat meshes differently, which is worth knowing before it surprises you. Half cuts the meshes and the shell along with the stack. The Plates slab does not: it clips the tissue only, so a mesh and the shell stay whole however far in you pull it — which is what makes the Surface figure below legible.
| What you have selected | What the view draws |
|---|---|
| A structure with its own mesh | that mesh |
| A structure printed inside a shared region — A1, in Au1 (A1/AAF) | the mesh filed under the name the label leads with, Au1, which is the region the plate outlines too |
| A division | one mesh per member, up to 140; past that the largest of them, and the note says how many were left out |
| Nothing selected, but a filter down to 40 results or fewer | one mesh per result |
A division has no mesh of its own and never will — it is its members' meshes standing together, exactly as its outline on a plate is their outlines unioned. The note adds them up: the hippocampal formation below is 24 meshes, 64.13 mm³; the cerebral cortex is 65 meshes of its 68 structures — the other three the atlas draws no region for — and 333.96 mm³.
Surface appears once Meshes is on, and wraps everything in the interpolated brain surface as a faint shell. It gives a structure a body to sit in: on its own a mesh floats, and against 62 contour rings it is hard to place.
The same CPu with Surface on, and the Plates slab pulled in to two sections so the shell is not competing with the whole stack. The shell is the brain outline interpolated the same way the structures are — the outer boundary of the same field.
The shell is depth-tested against whatever the meshes wrote, so it follows Opacity below without being told: solid meshes occlude it, and translucent ones let it lie over them.
Surface is the one 3-D setting that does not ride in a link. Meshes, the opacity and the color are all carried; the shell is not, so someone opening your link gets the meshes without it and ticks the box themselves.
At the default 92% and above, the meshes are solid enough to sort themselves: one depth-writing pass, nearest surface wins. Turn it down and they are composited back to front instead — which is the whole of what lets a structure inside another one show through it.
The hippocampal formation at 35%. At the default the outer subfields hide the inner ones completely; composited, the whole nested set reads at once.
Compositing is ordered by each mesh's own center against the camera. The shapes here are compact and it holds for them, but two that genuinely interleave can still come out in the wrong order. Double-click the slider to put it back to 92%.
Three answers to what a mesh's color means, and none of them is anatomy.
| Option | What decides the color |
|---|---|
| One per structure (default) | A hue hashed from the structure's own name. The same abbreviation is the same color in every session and either pane — but 360 degrees and 697 names, so two can land on the same hue |
| Plate colors | The plate's own coloring lifted into the third dimension: the eight slots solved so that no two regions touching on a plate are alike, and the same color on every plate a region is drawn on |
| Selection | The selected structure in the color the plate marks it with, and a division's members all in the division's one color |
The hippocampal formation as 24 meshes in One per structure. This is the case the default exists for: one structure is one shape and any color draws it, but a division has to say which of its members you are looking at before it says anything else.
The same 24 meshes in Plate colors — eight colors rather than 24, and each one the color that structure wears on the section underneath it. This is the mode that makes the plate and the 3-D view one picture. Adjacency in depth was never part of that solve, so two meshes that meet only in the gap between two sections can still come out alike.
Selection is the mode for "where is this", rather than "which of these am I looking at".
&mh=1 carries the meshes, &mo= the opacity and &mc= the color; &mh2, &mo2 and &mc2
do the same for pane B. A link carrying any of them opens with Advanced already unfolded,
so a folded panel is never the reason the picture looks the way it does.
| Action | How |
|---|---|
| Orbit | Drag |
| Pan | Shift-drag or right-drag |
| Zoom | Wheel |
- Nothing here is a segmentation. The contours are the atlas's drawn boundaries; the labels are where abbreviations are printed.
- The streaking in the volume view is arithmetic, not anatomy. Sections are 350 µm apart but a plate pixel is about 17.5 µm — sampling along the brain is 20× coarser than across it. The stack really is discrete, and the volume interpolates between slices.
- A single structure reads as a sparse arc, not a shape. Most structures carry only a handful of labels — the median is six.
- A mesh is interpolated, not segmented. The atlas is 62 drawings and nothing between them. The meshes fill the gap at 50 µm, so six planes in seven are arithmetic — a linear guess along the brain, not a measurement. Where a structure really changes shape inside 350 µm, and thin laminae do, the mesh cannot show it.
- Some meshes are hulls rather than shapes. A structure drawn on only one or two plates is never sampled along AP at all, so it is closed as a convex hull per component: a claim about where it is, not about what shape it is. 433 of the 697 follow the drawn boundaries; the other 264 are hulls, and the note says which you are looking at.
- For thin layered structures a label is often printed just outside the structure itself.