One Cubic Millimeter of Human Cortex Reconstructed at Nanoscale Resolution
Alexander Shapson-Coe, Jeff Lichtman and colleagues published A Petavoxel Fragment of Human Cerebral Cortex Reconstructed at Nanoscale Resolution in Science in May 2024 (DOI 10.1126/science.adk4858). The dataset is a single cubic millimeter of human temporal cortex, taken with consent from tissue removed during surgery, imaged with electron microscopy at nanoscale resolution and reconstructed into the most detailed available picture of human brain tissue.
| Measure | Value |
|---|---|
| Volume reconstructed | 1 cubic millimeter of human temporal cortex |
| Cells cataloged | About 57,000 |
| Synapses annotated | About 150 million |
| Imaging data | 1.4 petabytes |
| Reconstruction type | Petavoxel, every voxel annotated |
What the reconstruction found in one cubic millimeter
The sample carried surprises that tissue-scale imaging could not see. The reconstruction revealed rare but powerful connections between neurons that had not been appreciated at this scale in human tissue, including synapses formed directly between axons. It cataloged the full diversity of cortical cell classes in the fragment, from excitatory and inhibitory neurons to glia and vasculature, with their connections intact. Rare structures survive in a dataset of this resolution because every voxel is preserved, and the fragment’s rare connections are strong enough to shape circuit dynamics despite their scarcity.
The numbers also quantify the scale problem of human connectomics. One cubic millimeter of cortex, about one millionth of a human brain by volume, produced 1.4 petabytes of data requiring distributed computational reconstruction. The whole cortex at this resolution sits orders of magnitude beyond that, which is why the paper is simultaneously a biological result and a statement about instrumentation.
What Human Cortex Connectomics Costs
Until this reconstruction, the highest-resolution connectomics datasets came from model organisms. The fly brain datasets, examined in the FlyWire connectome post and the MaleCNS comparison, are complete but small. The worm diagram is smaller still, examined in the mind of a worm post. The petavoxel fragment is the first dataset at this resolution from human cortex itself, which removes extrapolation from rodent or insect anatomy for the structures it contains. Its findings about rare powerful connections have direct analogues in other species, but the human measurements now exist as the reference.
The fragment is also the structural counterpart of the recording bounds problem. The 2013 analysis by Marblestone and colleagues quantified what it would take to record every neuron’s activity, covered in the physical limits of neural recording post, and this reconstruction quantifies the structural side of the same equation, what it takes to see every synapse in a piece of human cortex.
Comparison to The Consciousness AI
This project treats consciousness as an emergent property of physical organization and holds that any theory of that organization must survive contact with the biological substrate at relevant resolution. The petavoxel fragment is the finest structural measurement of the substrate that generates human experience, and its lessons constrain emulation work directly. Rare powerful connections change network dynamics in ways averaged connectivity statistics miss, so any emulation built from coarser human data inherits unknown error. The project’s research code, maintained in the tlcdv/the_consciousness_ai repository, keeps structural fidelity requirements explicit rather than assumed. The dataset also deepens the substrate independence question. Every synapse in the fragment has a physical address and a physical weight, and the fragment says nothing direct about which of those details experience requires. That question is the subject of the structural identity debate examined in Sebastian Seung’s connectome hypothesis.
What One Cubic Millimeter Settles and Leaves Open
The fragment settles that nanoscale reconstruction of human tissue is achievable with current instruments and a petabyte-scale budget. It leaves open the same question every structural dataset leaves open. Dynamics, plasticity state and molecular physiology are not in the image. The reconstruction is the structural half of a two-part measurement problem, and the field’s attempts to close the loop from structure to function in mammalian cortex are examined in the MICrONS mouse cortex analysis. What structural and functional data would together tell us about experience is reviewed in the flagship overview of AI consciousness research, with emulation coverage collected on the brain emulation page.