The Cost Curve of Electron Microscopy for Connectome Reconstruction

A cubic millimetre of mouse cortex, 523 million synapses, two petabytes of raw image data, and a reconstruction effort that published in 2025. The numbers describe a single experiment — and they explain why connectome-scale scanning is, for the moment, the rate-limiting step in any long-horizon preservation programme.

---

FACT

The MICrONS (Machine Intelligence from Cortical Networks) programme — a five-year, US$100 million IARPA project that ran from 2016 to 2021, with results published in Nature in 2025 — is the most ambitious connectome-scale volumetric electron microscopy (EM) dataset completed in a mammalian brain (Scientific American, 2016; Bae et al., Nature 2025). The team imaged a 1.4 × 0.87 × 0.84 mm block of mouse visual cortex at nanometre resolution, producing roughly 2 petabytes of raw alignment image data that ultimately yielded a reconstruction of about 200,000 cells, 120,000 neurons, and 523 million detected synapses (MICrONS Explorer — Cubic Millimeter; Quanta Magazine, 2016).

The cost of acquisition is one thing; the cost of reconstruction is another. As Jeff Lichtman — Harvard neuroscientist and one of the project’s imaging leads — put it during the programme’s launch year, “[e]ven if the whole world was coloring in for you, it would take a lifetime to get the whole cubic millimeter colored in” (Scientific American, 2016). The automated segmentation and proofreading pipeline that ultimately produced the 2025 dataset is the engineering response to that observation: convolutional neural networks now do the bulk of the volume tracking, with human proofreaders correcting errors at the synapse level.

For comparison, the Drosophila adult brain — roughly 130,000 neurons and ~50 million chemical synapses, an order of magnitude smaller than a cubic millimetre of mouse cortex — was imaged by Janelia’s FlyEM team in 2017 and reconstructed at scale by the FlyWire consortium only in 2023, six years later (Dorkenwald et al., 2024). In 2025, two groups published complete adult Drosophila central nervous system connectomes — the BANC dataset (female, ~160,000 neurons, ~214 million synapses) and the Male CNS dataset (~166,000 neurons, ~312 million synapses) — using high-throughput serial-section transmission EM combined with multi-beam FIB-SEM (Drosophila connectome — Wikipedia).

The endpoints that matter for any long-horizon preservation programme are acquisition and reconstruction throughput, not raw pixel resolution. The Brain Preservation Foundation’s 2018 Large Mammal Prize — awarded to 21st Century Medicine for aldehyde-stabilised cryopreservation (ASC) of a whole pig brain — was verified by independent 3D electron microscopy across the volume to confirm that synaptic connectivity was preserved (Brain Preservation Foundation announcement, March 2018). That verification step is, at present, the only published demonstration that the imaging side of the chain can read a preserved large mammalian brain at the relevant resolution.

IMPLICATION

The shape of the curve is the story. From the 1986 C. elegans connectome (302 neurons, manually reconstructed from printed micrographs) to the 2025 MICrONS dataset (120,000 neurons, ML-reconstructed from petavoxel EM) is roughly six orders of magnitude in neuron count and roughly four orders of magnitude in throughput — across forty years. The slope is consistent with a curve that has been doubling every few years, mostly driven by automation (FIB-SEM, multi-beam SEM, flood-filling networks, convolutional segmentation) rather than by raw microscope speed.

Three practical implications follow for anyone thinking about long-term connectome preservation:

1. Acquisition is no longer the binding constraint at the cubic-millimetre scale. New multi-beam SEM platforms (e.g., the 91-beam systems now in use at Janelia and elsewhere) cut raw acquisition time substantially. The bottleneck has moved to image alignment, segmentation, and proofreading — which is a software and human-effort problem, not a microscope problem.

2. The cost of full-brain reconstruction is now a projection, not a measurement. No published cost-curve model places a per-cubic-millimetre dollar number with confidence intervals. The MICrONS total programme cost of $100M includes functional imaging, neuroscience research, algorithm development, and cloud storage — not just EM acquisition. Extrapolating that figure to a whole human brain (~1,400,000 mm³) without adjustment is a category error.

3. The “wait for the curve” framing is structurally similar to the way the Human Genome Project was framed in 1990. The genome cost curve famously fell five orders of magnitude in fifteen years. The EM connectome curve is on a similar trajectory but is younger, less capitalised, and has not yet had its equivalent of the capillary-stamping technology breakthroughs. It is plausible — not certain — that the curve will continue to fall steeply enough that, by the time a preserved human brain is ready to be read, the cost of reading it is within reach of a well-funded research programme rather than a national-lab campaign.

HUMILITY

What we cannot do — and what would be a serious error to promise — is shorten the curve ourselves. ASC and the Patient Care Trust are designed to keep the structural asset intact across whatever timeline the science needs. The MICrONS dataset, the FlyWire reconstruction, and the 2025 Drosophila CNS connectomes are remarkable, but they are reconstruction of small volumes and small brains. A whole human brain is roughly 1,400,000 cubic millimetres; the human cortex alone is roughly 10,000 times the volume of MICrONS. We do not have published figures for what it would cost to scan a whole human brain at the resolution needed for synapse-level reconstruction, and we do not have a published engineering path that closes that gap within any specific window.

What we have is a reasonable expectation that the curve will continue to move. The right institutional posture is to preserve the structure with the highest fidelity currently available, fund the science we can fund without overpromising, and let the cost curve do what cost curves tend to do — without pretending to know when it will arrive.

---

This content explores speculative, post-mortem neuro-archival research. It does not constitute medical advice. Mind uploading and whole-brain emulation are theoretical future technologies, not active medical procedures.

Sources

1. Bae, J. A. et al. (2025). “Functional connectomics spanning multiple areas of mouse visual cortex.” Nature 640 (8058): 435–447. — Petavoxel-scale TEM reconstruction of a cubic millimetre of mouse visual cortex, 523M synapses, 75K functionally imaged neurons. https://doi.org/10.1038/s41586-025-08790-w
2. MICrONS Explorer — Cubic Millimeter. — Public dataset summary: 1.4 × 0.87 × 0.84 mm volume, 200K cells, 120K neurons, 523M synapses, multi-terabyte dataset. https://www.microns-explorer.org/cortical-mm3
3. Cepelewicz, J. (2016). “The U.S. Government Launches a $100-Million ‘Apollo Project of the Brain’.” Scientific American. — IARPA MICrONS launch context; Lichtman quote on manual reconstruction scale; 2 PB data estimate. https://www.scientificamerican.com/article/the-u-s-government-launches-a-100-million-apollo-project-of-the-brain/
4. Singer, E. (2016). “Mapping the Brain to Build Better Machines.” Quanta Magazine. — Independent confirmation of the 1–2 petabyte per cubic millimetre estimate; programme scope and timeline. https://www.quantamagazine.org/mapping-the-brain-to-build-better-machines-20160406/
5. Brain Preservation Foundation (2018). Large Mammal Brain Preservation Prize announcement. — ASC pig-brain verification by independent 3D EM; synaptic connectivity preserved across whole brain. https://www.brainpreservation.org/large-mammal-announcement/
6. Wikipedia — Drosophila connectome. — Dorkenwald et al. 2024 (FlyWire, ~130K neurons, ~50M synapses); BANC 2025 (female, ~160K neurons, ~214M synapses); Male CNS 2025 (~166K neurons, ~312M synapses). https://en.wikipedia.org/wiki/Drosophila_connectome
7. Wikipedia — Connectomics. — Background on microscale vs macroscale connectomics methods, recent LICONN expansion-microscopy alternative (2024), and NIH investment in whole-mouse-brain EM. https://en.wikipedia.org/wiki/Connectomics