What's a brain?

The hippocampus at 50 micrometres

A human hippocampus scanned at fifty micrometres, the finest diffusion MRI of a human brain reported so far. In September 2026 a team at Zhejiang University put a five millimetre coronal block from the middle of a right hippocampus, from a woman of fifty with no brain disease, into a 14.1 tesla magnet for 78 hours and imaged it at 50 by 50 by 100 micrometres, close enough to the size of cells that the scan started to agree with the stains a pathologist would use. Then they did the same for a man of 91 with Alzheimer's. Below: what that voxel is against the real structure, the layers it could tell apart, and the three pathways of the memory circuit it traced without cutting anything. Drag to turn the hippocampus, scroll to zoom out along the fornix.

Atlas · shape from 1,000+ brains Measured · cables Block position schematic

Loading the hippocampus…

Voxel

Cables Measured · HCP tractography, 1,065 adults

The two hippocampi are the population shape from the Melbourne Subcortex Atlas, so the lengths and voxel counts are real. The cables are the two bundles of a whole-brain tract atlas that end in the hippocampus, drawn in the same millimetres. The block through the middle marks where the specimen came from only roughly: the paper says mid-hippocampal, and no more than that is drawn.

The granule cell layer of the dentate gyrus is about 0.2 mm thick, the thinnest layer the paper segmented, and locally it can be as thin as 60 µm. A clinical voxel is five of those layers stacked. At 100 µm the paper reports the layer markedly attenuated; at 50 µm it is four voxels across, and the scan can see which side of it a fibre is on.

Measured · 0.2 mm layer, this specimenBand position schematic

Where 50 µm sits. Four nanometres is the pixel of the electron microscope that imaged the cubic millimetre; a millimetre is the voxel of a hospital scan. This paper lands close to the middle, at the scale of a pyramidal cell body, which the paper puts at 10 to 50 µm. The scan data have not been released; the paper's figures are open (CC BY 4.0) and appear further down.

Measured

The circuit it traced

Water in tissue moves more easily along fibres than across them, and diffusion MRI measures that. At 50 µm the paper's tractography followed the three legs of the hippocampal trisynaptic circuit in three dimensions: the perforant path, the mossy fibres and the Schaffer collaterals. The drawing here is a textbook section, not their data; press a leg to run a signal along it.

Schematic · textbook section, hand drawn

Inside the tissue

The scan itself has not been released, but the paper is open, so the panels the authors made from it can be walked here. Pick a layer, a pathway, a resolution or a comparison and the viewer flies to the authors' own panel for it, with what they say it shows. Drag to pan, scroll or pinch to zoom, double tap for the whole figure. Everything in this box is specimen data: one control block, one Alzheimer's block.

A figure from Shen et al. 2026, navigated by the menu beside it

Specimen · Shen et al. 2026 · n = 1

What the scan saw

Reported · control specimen

Six laminae. Alveus, pyramidal cell layer, stratum radiatum, stratum lacunosum-moleculare, molecular layer and granule cell layer, segmented by hand on the scan, with the hilus inside. The molecular layer split into inner and outer, a distinction made by which fibres arrive there rather than by any visible edge.

Fibres that match the stains. Diffusion maps and tractography lined up with immunohistology for axons and dendrites, at a scale the paper describes as approaching cellular dimensions.

What 100 µm loses. Downsampling the same scan to 100 µm blurred the boundary between stratum radiatum and lacunosum-moleculare, attenuated the granule cell layer, and preferentially lost streamlines shorter than 10 mm. At 400 µm only grey and white matter contrast was left.

Reported · Alzheimer's specimen

Less contrast between layers. The laminar pattern that is crisp in the control block was reduced in the Alzheimer's block, and the CA1 pyramidal cell layer was thinner.

Local breaks. Discontinuities in stratum radiatum and lacunosum-moleculare, and granule cell dendrites in the molecular layer that had lost their regular radial order.

Matched by pathology. The same regions showed changes on pathological staining, which is what gives a nondestructive scan its claim to be virtual histology.

Not shown, and why. One control block and one Alzheimer's block is a feasibility study, not a cohort, so nothing here draws an Alzheimer's hippocampus against a healthy one. The paper carries no data availability statement and the scan has not been released, so the layers and fibres above are a hand drawn section, badged as such, and the voxel comparisons use the atlas shape and the paper's one layer thickness. When the data are public this page will draw them.

By the numbers

Quantity	Value	Source
Voxel, this paper	50 × 50 × 100 µm	Shen et al. 2026, abstract
Field strength	14.1 T	Shen et al. 2026
Scan time per block	78 hours	Shen et al. 2026, methods: b = 3000 s/mm², 20 directions, 16 averages
Block	16 × 19 × 5 mm field of view	Shen et al. 2026, methods: middle of the right hippocampus
Specimens	2	A woman of 50 without brain disease; a man of 91 with Alzheimer's
Downsampled comparisons	100, 200, 400 µm	Shen et al. 2026, figure 2
Granule cell layer thickness	about 0.2 mm	Shen et al. 2026, results; locally as thin as 60 µm (Amaral et al. 2007)
Hippocampus length, front to back	…	Melbourne Subcortex Atlas mesh, measured on this page
50 µm voxels along that length	…	Computed from the row above
Clinical voxel, for comparison	1 mm	A typical 3 T structural scan

Read the paper

Shen Y, Yu Z, Zhu Q, Zhao Z, Wang P, Yin Y, Jiang P, Wang C, Kong X, Zhang J, Wu D. Diffusion MR Microscopy of the Human Hippocampus at 50 μm Resolution at 14.1 T for 3D Virtual Histology. bioRxiv, posted 9 September 2026, not yet peer reviewed, CC BY 4.0. doi:10.64898/2026.09.04.749425. Zhejiang University, the National Human Brain Bank for Health and Disease, and collaborators in Hangzhou, Shanghai and Ningbo.

Zhao Z, Zhang L, Luo W, et al. Layer-specific microstructural patterns of anterior hippocampus in Alzheimer's disease with ex vivo diffusion MRI at 14.1 T. Human Brain Mapping 2022. doi:10.1002/hbm.26062. The same group's earlier work at 100 µm, the baseline this paper improves on.

Yeh FC. Population-based tract-to-region connectome of the human brain. Nature Communications 2022. doi:10.1038/s41467-022-32595-4. The fornix and parahippocampal cingulum, from the HCP1065 tract atlas, CC BY-SA 4.0, the same data as the white matter page.

Tian Y, Margulies DS, Breakspear M, Zalesky A. Topographic organization of the human subcortex unveiled with functional connectivity gradients. Nature Neuroscience 2020. doi:10.1038/s41593-020-00711-6. The hippocampus meshes, from the Melbourne Subcortex Atlas.

The figures in the tissue viewer are reproduced from the preprint under its CC BY 4.0 licence, unaltered apart from encoding; the boxes and captions are this site's.

Diffusion MRI cannot tell which way a signal travels along a fibre. The direction of the animated signal above follows the textbook, not the scan.