Tulane UniversityFlyTulane

These are real neurons — and the mitochondria inside them — from a fruit fly’s brain. Every contour of these three-dimensional forms was built from thousands of individual outlines, traced frame by frame through serial sections of the tissue by volunteers like you.

A computer made a first attempt at drawing the outline of the object in each frame. Your job is to check that outline — and correct it where it’s wrong. Approving a good segmentation takes seconds. Correcting a poor one takes a little longer — but that correction is precisely where your contribution matters most. Across thousands of contributors, those seconds and minutes add up to something a computer alone cannot produce: data precise enough to measure, and reconstruct, and publish.

You don’t need a science background. You need a steady hand and a few minutes. A short tutorial will show you everything.

Sign up

AI segmentation of electron microscopy data is not, by itself, a scientific result. It is promising but unverified computational output. There is a hard boundary between EM data you can measure and publish, and EM data you can only look at. Machine learning alone does not get you across that boundary — except under imaging conditions that most experimental protocols cannot achieve. Human proofreading is what gets you across it. TulaneFly is how you get there at scale.

About the project

We image entire neurons at nanometer resolution with serial block-face electron microscopy, then apply deep-learning models to segment subcellular structures. Volunteers correct prediction errors; independent edits from multiple annotators are pooled into consensus labels for research-grade 3D reconstructions and quantitative analysis.

Why this matters

By mapping mitochondrial size, shape, and distribution across entire neurons — cell bodies, dendrites, and axon terminals — we aim to understand how nerve cells regulate the placement of their energy sources to meet local metabolic demands.

Why we need you

AI models segment these massive volumes but still misplace boundaries. Human annotators catch subtle errors that algorithms miss. Each small edit brings us closer to accurate, publication-quality reconstructions.

How do I get started?

Create an account, then click Get Started above. Open the 3D viewer, click a branch segment, and start proofreading masks frame by frame.

How does the paintbrush work?

Select Add to paint or Subtract to erase. Adjust Brush size with the slider, use Opacity to compare against the raw EM image. Your changes auto-save. Keyboard shortcuts: A add, S subtract, ← → navigate.

How do I approve my work?

When a frame looks correct, click Approve. Approved items turn green in the progress bar. Multiple independent annotators review each frame to build consensus.

Tips for efficient annotation

Save often and work in short passes. Use the opacity slider to check edits against the base image. Use arrow keys to quickly move between frames. Assign yourself a segment before editing so your changes are saved.