Marmoset STPT, between-animal registration

Serial two-photon tomography brains of two marmosets from the Brain/MINDS Marmoset Connectivity Resource, at 0.1 mm, with 36 atlas-derived regional labels.

10 test pairsbraininter-subjectfixed FLUORESCENCE, moving FLUORESCENCEaligned once by the benchmark

Look at the images

Read before comparing (4)

Truth. The labels are atlas regions propagated by the BMCR registration pipeline, automatic rather than drawn on each animal, so the dataset is ranked apart from independently annotated tasks. The release names none of the 36 identifiers: scores use them as they are, with no correspondence to other datasets.
Images. Only org/img.nii.gz and org/label.nii.gz are used; the affine_s and translate_s copies are not other animals. Images keep their intensities and native grids. The labels, stored as floating-point integers, are checked and kept as uint8 with the same values.
Alignment. The moving header carries an elastix translation from the intensities alone, no label used.
Baseline. With no registration the ten test pairs start at a median region Dice of 0.537 (0.222 to 0.721) and an HD95 of 0.784 mm.

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

		Accuracy			Smoothness		Cost
#	Method	Dice	Change	HD95 mm	SDlogJ	Folds %	Time s	RAM GB	GPU GB	Images
	No registration	0.489 ±0.142		1.00 ±0.39	0.000	0.00

Dice: how much the organ contours overlap, 1 is perfect. HD95: distance between the contours' surfaces. SDlogJ: how irregular the deformation is, 0 is perfectly smooth. Folds: share of the image where the deformation folds over itself.

Pair by pair

Measure

Dice	1	2	3	4	5	6	7	8	9	10
No registration	0.609	0.275	0.721	0.279	0.593	0.699	0.314	0.693	0.222	0.481

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

The first test case of Marmoset STPT, between-animal registration

The first test caseClick to look at every case.

Test pairs: The 52 released animals are shuffled with salt 2026:MarmosetSTPT:subjects and paired consecutively into 26 disjoint pairs. Ten test and ten development pairs follow in seed-2026 order, the first ten and the reversed last ten. The template and the six remaining pairs are not used; each animal appears once.
Truth: contours by a program Thirty-six nonzero atlas-derived region identifiers in the released label.nii.gz files. They are preserved by identifier; the derivative archive supplies no numeric-to-anatomical-name legend, so no anatomical names are inferred. Atlas propagation by the BMCR registration pipeline makes this automatic truth, not independent manual delineation on each subject.
Source: zenodo.org/records/19041877
doi.org/10.5281/zenodo.19041876
neurodata.riken.jp/id/20230630-001
doi.org/10.1371/journal.pbio.3002158
Licence: CC BY 4.0

10 pairs for tuning, never tested

R01_0026__R01_0115, R01_0078__R01_0063, R01_0034__R01_0114, R01_0033__R01_0110, R01_0064__R01_0056, R01_0069__R01_0029, R04_0023__R01_0094, R01_0030__R01_0060, R01_0053__R01_0092, R01_0104__R01_0091