fidopt Interactive view
Spatial transcriptomics, adjacent H&E sections (2D)

spatialLIBD DLPFC Visium, adjacent sections

Two adjacent H&E sections, 10 µm apart, of one adult's dorsolateral prefrontal cortex on 10x Visium slides, with the cortical layer people assigned to each spot (spatialLIBD, Maynard et al. 2021).

4 test pairsbrainsame patientfixed BRIGHTFIELD, moving BRIGHTFIELD2D, distances in µmaligned once by the benchmarkresearch only
Look at the images
Read before comparing (6)
  • Sections. Twelve sections of the dorsolateral prefrontal cortex of three neurotypical adults (Br5292, Br5595, Br8100), four each: two pairs of adjacent sections 10 µm apart, the pairs 300 µm apart. Kept although small: no other dataset of the corpus holds spatial transcriptomics.
  • Images. The H&E scans as delivered, 8-bit colour, cut to the box of the spots over the tissue widened by 57.7 µm: the full scans hold the slide's fiducial frame, which elastix aligned instead of the tissue (a start Dice of 0.53 on 151675_151676, 0.72 once cut). Pixels of 0.719 to 0.725 µm from the spots' 100 µm grid; distances are in µm. No mask is given.
  • Truth. The layer the authors gave each spot over the tissue, by hand from the cytoarchitecture and marker genes, spread to the pixels within 57.7 µm of the spot: a map with the spots' 100 µm grain and stepped borders. Br5595's sections, the development pairs, hold no L1 or L2 spots.
  • Alignment. elastix rigid on the cut images: rotations under 2.5 degrees, shifts up to 0.46 mm. From that start the four test pairs have a mean Dice of 0.782 over the layers (0.723 to 0.826) and an HD95 of 283 µm (217 to 355): little room is left above a truth of 100 µm grain.
  • Independence. The four test pairs come from two donors, two pairs each, and are not four independent samples; development comes from the third donor.
  • Overlap. No other dataset of the corpus holds Visium sections or the Lieber Institute's brains.

No method has a result on every pair of this dataset yet.

Pair by pair

Click a score to look at that result: the fixed image, and over it the moving image after registration, with the annotations. Underlined scores have their image; for the others you see the pair without the result. A green cell gained against no registration on that pair, a red one lost.

About the data

Test pairs
Each donor gives two pairs of adjacent sections. The three donors in the order of sha256("2026:DLPFCVisium:<donor>"): the first two give the four test pairs (Br8100, Br5292), the last the two development pairs (Br5595).
Truth
contours by experts the layer (L1 to L6 or white matter) the authors assigned by hand to each Visium spot over the tissue, from the H&E's cytoarchitecture and marker genes, spread to the pixels nearest each spot (10X/barcode_level_layer_map.tsv of the HumanPilot repository)
Licence
none stated for the images (the authors' public S3 bucket); the HumanPilot repository, which holds the spots and their layers, is under the Artistic License 2.0, as is the spatialLIBD package
2 pairs for tuning, never tested

151669_151670, 151671_151672