fidopt Interactive view
ORION native ROIs, known warp

ORION colorectal cancer, H&E and Hoechst of one section, known warp

A 2048-pixel square (0.67 mm) of one colorectal cancer section: its H&E (fixed, colour) and its Hoechst channel imaged before staining (moving), deformed by a known smooth warp (ORION study, Nature Cancer 2023).

10 test pairscolorectal tissuesame sectionfixed HISTOLOGY, moving FLUORESCENCE2D, distances in µmdelivered alignedresearch only
Look at the images
Read before comparing (4)
  • Images. The provider's pixels, 0.325 µm a pixel, read from the whole slides: nothing is resampled before the warp, and both images carry the square's place in the slide. The square is drawn among those the H&E's overview shows covered by tissue: tissue chosen, not a census of the slide.
  • Start. The H&E was stained after the fluorescence scan, on the same section, and registered onto it by the authors with palom, automatically. That registration is the start, not a truth, and fails where the section changed: prepare.py draws again until the nuclei of the two stains agree, as on CRC21, whose first square fell beside a fold. The 20 squares kept correlate at 0.36 to 0.81.
  • Warp. amplitude_mm and spacing_mm are in this dataset's unit, the µm: B-spline coefficients within 30 µm on knots 100 µm apart move a point by 10 to 12 µm on average and 26 µm at most, twenty times and more the residual of the authors' registration.
  • Border. The warped Hoechst image is filled with its minimum where the warp reads outside the square, a band of at most 92 pixels along the border.
AccuracySmoothnessCost
#MethodTRE µmChangerTRESDlogJFolds %Time sRAM GBGPU GBImages
1ANTs SyNCPUapp10.04 ±1.36−0.640.0107 ±0.00140.1230.00740.8
1NiftyRegCPUapp9.71 ±1.57−0.980.0103 ±0.00171.3766.35190.5
No registration10.68 ±1.090.0114 ±0.00120.0000.00

The order averages each method's ranks on TRE µm, SDlogJ, so the first one is not always the best in every column. ± is the 95% interval of the mean over the 10 pairs. Two neighbours are tied when a paired Wilcoxon test on the shared pairs does not find the upper one better on TRE µm (5%): the places a method could hold are then shown under its own. The best value of each column is in colour. Click a row to see its images, a column title to sort by it. Time and memory measured on Laptop.

TRE: distance between matching landmarks. rTRE: the TRE over the length of the image's diagonal, to compare datasets of other sizes and units. SDlogJ: how irregular the deformation is, 0 is perfectly smooth. Folds: share of the image where the deformation folds over itself.

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
10 test specimens drawn with seed 2026 among the 40 of the public bucket (CRC01 to CRC40; CRC33 is its first piece, CRC33_01), then 10 for development; prepare.py draws one square a specimen. The study gives no patient key for these identifiers, so a specimen is taken as a patient.
Truth
landmarks a known deformation 100 points a case, drawn in the tissue where Hoechst is bright, 150 pixels or more from the border, and carried exactly through the known warp. They measure how well a method undoes that warp, on top of the authors' own registration of the two images, which they do not measure: by the nuclei of the two stains it is off by 0.2 to 0.5 µm (median over tiles of 256 pixels, a case) and 2.3 µm at worst on 19 cases, by 0.8 µm and 4.1 µm at worst on CRC21, whose fluorescence image is partly out of focus.
Licence
None stated for the images, on the bucket, the atlas page or the repository's README. The companion record (Zenodo 7637988, MIT) covers the index and the code, not the images.
10 pairs for tuning, never tested

CRC03, CRC04, CRC22, CRC23, CRC26, CRC30, CRC31, CRC33, CRC34, CRC38