ChimeraX and IMOD: seeing and segmenting the 3D

The two tools you can't avoid in practice — IMOD to align and reconstruct, then a step-by-step tutorial for visualizing your MRC volume in ChimeraX, saving a figure, and tracing the structures out of it.

Once the principles are clear, real hands-on work runs through two pieces of software: IMOD (align the tilt series and reconstruct the 3-D volume) and ChimeraX (inspect that volume, turn it into figures, and trace the structures out of it). This page lays out how to install each and how to use it — both to see the volume and to segment the structures of interest out of it.

Installation and setup

Both are free and cross-platform (Windows / macOS / Linux).

IMOD: download the installer for your platform from bio3d.colorado.edu/imod and follow the instructions; it ships with etomo, 3dmod, tilt and the rest. For reconstructing large volumes, a CUDA-capable NVIDIA GPU helps a lot. If imod and etomo run from the command line, you’re set up.
ChimeraX: download from cgl.ucsf.edu/chimerax; free for academic use. A better GPU makes rendering large volumes smoother; if VRAM is tight, downsample the volume first.

With both installed, IMOD handles “tilt series → .mrc / .rec volume” and ChimeraX handles “see and segment the volume” — the two sections below cover how to use each.

IMOD: alignment and reconstruction

IMOD, developed at the Boulder Lab (David Mastronarde et al., University of Colorado), is the de-facto standard package for cryo-ET. It covers the whole path from a raw tilt series to a 3-D volume:

etomo — a graphical pipeline wizard that walks you through alignment and reconstruction;
alignment: a coarse alignment first, then a fine alignment using gold fiducials or patch tracking, estimating the geometry of each projection;
reconstruction: the tilt command runs weighted back-projection (WBP) or SIRT to produce the tomogram;
3dmod — browse the volume, draw models, and do segmentation and annotation.

Minimal workflow

Import the tilt series → coarse align → fine align (fiducials / patch) → reconstruct with tilt → view in 3dmod. etomo gives a panel for each step, and the commands can be scripted for batch runs.

Visualizing your MRC: a ChimeraX tutorial

Once reconstruction finishes, what you have is a 3-D volume file (.mrc or .rec) — a grid of density numbers you can’t see by eye. UCSF ChimeraX (free) renders it into a 3-D structure you can view, rotate, and turn into a figure. Walk this minimal path once:

Install and open. Download ChimeraX from the official site (Windows / macOS / Linux). Once it’s running, drag your volume into the window, or type open path/to/your.mrc in the command line at the bottom — ChimeraX detects it as a volume.
Set the isosurface threshold. A Volume Viewer panel pops up. Drag its threshold (level) slider: lower shows more (and more noise); higher keeps only the strongest density. Slide back and forth to find the level where the structure is just clear and the noise isn’t too bad — this step matters most.
Pick a display style. volume #1 style surface is a solid isosurface (best for overall shape); style mesh is a wireframe; style image is volume rendering.
Light it and outline it. lighting soft, or lighting full for shadows and more depth; add graphics silhouettes true to outline the structure for cleaner screenshots.
Colour. A flat colour like color #1 cornflowerblue, or colour by value/height so depth reads.
Rotate, clip, look. Drag with the mouse to rotate, view to reset; to see the interior, cut through it with a clipping plane (the Side View clip, or the clip command).
Save a figure. save figure.png supersample 3 exports a high-resolution screenshot; for a rotation movie, movie record … spin once … movie encode movie.mp4.

Tip

The 3-D reconstruction figures on this site’s results page (apoferritin, Vipp1) were rendered exactly this way. Dial in one good threshold and viewpoint, and every screenshot after that looks professional.

Segmenting the structures out

Seeing is only step one. An isosurface shows you the whole blob of density, but several things are usually tangled inside it — a stretch of membrane, a ribosome, a vesicle. Segmentation is the act of outlining each piece of structure in the volume, treating each as a separate object you can colour, hide, measure, and export on its own. Here are the two most common beginner routes: partition the density in ChimeraX, or hand-trace contours into a model in IMOD’s 3dmod.

Intuition

Segmentation answers “what is this piece of density, where is it, and how big?” The raw volume is just a soup of density values; segmentation puts a label on each spoonful, so you can lift a membrane out of the background, measure its area or volume on its own, or pull one organelle out and turn it around. Automatic or by hand, it all comes down to drawing boundaries in the volume and naming the regions.

Segmenting a density map in ChimeraX

ChimeraX ships with a Segment Map tool (menu Tools → Volume Data → Segment Map) that automatically cuts a density map into many regions: the algorithm finds local maxima in the density, then assigns the connected voxels around each maximum to it, so the whole volume is divided into small patches, each a candidate structural unit.

Minimal workflow:

Set a good threshold first. Just as for viewing, dial the isosurface threshold in Volume Viewer to where the structure is clear and noise is controlled — segmentation works on the density above that threshold, and too low a threshold carves noise into regions.
Run Segment Map. Open the tool, select your volume, click Segment. It computes a set of regions and colours each one differently.
Isolate the structure you want. Regions tend to come out fragmented. Use the tool’s merge/group controls to join the adjacent regions that belong to one structure (a stretch of membrane, one particle); hide or delete the regions you don’t want, leaving only the one of interest.
Colour, hide, view alone. Colour the kept region on its own and hide the rest of the density, and you get a clean figure with “only the target structure” left.
Measure and export. Selecting a region lets you read its voxel count (which converts to a volume) and see its position and extent; you can also save that region on its own as a new volume or surface mesh for downstream analysis or figures.

Depth

Segment Map’s “watershed by local maxima” works well on compact, high-contrast things (isolated particles, clear vesicles), but on long, thin, low-contrast membranes it tends to fragment or break them. In practice treat it as a starting point: auto-segment the big pieces, then clean up the boundaries by hand with colour/merge/hide. For a more accurate surface, export the region’s surface and tune it with ChimeraX’s surface display options (solid / mesh, transparency, smoothing) so the membrane or organelle outline reads cleanly.

Hand-segmenting in IMOD’s 3dmod

When the structure is too thin or too low-contrast for the automatic tools to be trusted, the most reliable route is to trace contours by hand. IMOD’s 3dmod is built for this: you page through the tomogram slice by slice, drawing a line around the boundary of a membrane or organelle on each one, and the stack of lines builds up a 3-D model object — then a surface is generated from the contours. That is a hand-made segmentation.

Minimal workflow:

Open the tomogram. 3dmod your.rec (or open it from etomo). A ZaP slice window appears; scroll through the slices (along Z) with the wheel, and a Model state is ready for drawing.
Enter modeling mode. Switch the window to Model mode (not Movie mode). Now a mouse click drops a model point.
Trace a contour along the boundary. Page to a slice where the target is clear, click points around the edge of the membrane / organelle to form a closed contour. Page to the next slice and trace another. Slice after slice, the stack of contours outlines the structure’s 3-D shape.
Separate objects. One object corresponds to one kind of structure. To segment several things (say outer membrane, inner membrane, a vesicle), create multiple objects, give each its own colour, and trace each separately.
Generate a surface mesh from the contours. Once traced, have 3dmod “skin” the stack of contours into triangles (a mesh), giving a continuous 3-D surface — the solid shape of the membrane or organelle.
Save the model, read it back as a segmentation. Save the model as IMOD’s .mod file. The model is a segmentation: each object = one labelled structure, usable to measure area / volume, do statistics, or export the surface for rendering in ChimeraX.

Tip

Hand-tracing is slow but the most controllable — weak signal and complex membrane structure often leave no other option. The trick is that you don’t have to draw every slice: trace a contour every few slices, let interpolation fill the gaps, then go back and fix the key slices. It saves a lot of clicking.

Automatic segmentation tools (good to know)

Tracing dozens of membranes through a cell by hand is clearly unworkable, so a set of automatic / semi-automatic tools learn to segment membranes and organelles for you — common ones include MemBrain (membrane-focused), EMAN2 (which has a convolutional-network tomogram-segmentation workflow), and the commercial Dragonfly. Most ask you to label a small batch by hand for training, then generalize the model to the full dataset; the output is again labels or surfaces for membranes / organelles. They are named here only as pointers — exact usage and quality depend on the data, so follow each tool’s own documentation.

One thread: the cleaner the reconstruction, the easier the segmentation

Whether you go through ChimeraX or 3dmod, how hard segmentation is depends directly on how clean the volume itself is. In a noisy volume where the missing wedge smears structure along z, membrane boundaries flicker in and out — automatic tools fragment them and a hand-tracer can’t find the edge. In a volume that has been denoised and missing-wedge-compensated, membranes are clear and continuous and particle boundaries are crisp, and segmentation gets far easier.

That is exactly the point of the training / restoration step: a self-supervised model (CryoCARE, DeepDeWedge, and this site’s own reconstruction methods) cleans the volume up and hands the missing wedge to a learned prior, and then you segment in ChimeraX / 3dmod. A good reconstruction is the precondition for a good segmentation.

How the two connect

In one line: IMOD takes you “from projections to a volume,” and ChimeraX takes you “from a volume to something you can see, segment, and explain.” In between sit the structural-analysis steps — subtomogram averaging, particle picking, segmentation — that raise SNR and localize and interpret structure. And the reconstruction methods this site develops are meant precisely to replace or strengthen how that IMOD step handles the missing wedge.

Previous step: train the denoising / restoration model — clean the volume up, then come back here to see and segment the structure. This is the last stop of the software pipeline.

← Software & Data Processing