Dataset · 2021
H01 — human cortical fragment
A cubic millimetre of human temporal cortex, imaged by serial-section EM and reconstructed automatically. It is the largest published volume of human brain at synaptic resolution, and the only one most learners will ever meet.
| Species | Human |
|---|---|
| Region | Temporal lobe cerebral cortex |
| Volume | ~1 mm³ |
| Cells | ~57,000 |
| Synapses | ~150 million |
| Data size | 1.4 PB |
| Resolution | 4 nm in-plane, 33 nm sections |
| Modality | Serial-section EM (ssEM) |
| Access | Open, no registration |
| Status | Released |
H01 is the reference dataset for human cortical ultrastructure. One cubic millimetre of temporal lobe, removed during epilepsy surgery, was fixed, stained, sectioned and imaged at 4 nm in-plane resolution, then segmented automatically. The release contains tens of thousands of reconstructed cells, millions of neurite fragments, roughly 150 million annotated synapses, and 104 cells proofread to completion.
What it is good for. Seeing human ultrastructure at all — myelin, spines, glia, and the ratio of excitatory to inhibitory contacts — and for practising the judgement that separates a real synapse from a tangentially cut membrane. The 104 proofread cells give you a small set of objects you can trust while you calibrate.
What it does not support. Circuit-level claims. Only 104 cells are proofread; the automated segmentation elsewhere carries merge and split errors at rates that make connectivity counts unreliable without your own correction pass. There is no functional data: nothing here says what any of these cells did. And it is one sample from one person with epilepsy, so anything you conclude about “the human cortex” is a claim about this tissue until someone replicates it.
Where these numbers come from. Shapson-Coe et al. (2024), Science 10.1126/science.adk4858; H01 release landing page. DOI 10.1126/science.adk4858.