BraTS-Reg

Two contrast-enhanced T1 brain MRs of one patient, before surgery and at follow-up, the follow-up scan being the fixed image (BraTS-Reg).

14 test pairsbrainsame patientfixed MR, moving MRdelivered alignedresearch only

Look at the images

Read before comparing (2)

Truth. 6 to 50 landmarks a pair, placed by clinical experts.
Overlap. Whether a BraTS-Reg patient is also in another BraTS-derived dataset of the corpus is not known: the challenge publishes no crosswalk to its source cohorts.

		Accuracy			Smoothness		Cost
#	Method	TRE mm	Change	rTRE	SDlogJ	Folds %	Time s	RAM GB	GPU GB	Images
1	ConvexAdam (MIND)GPUapp	2.23 ±0.76	−4.10	0.0060 ±0.0020	0.030	0.00	11	2.5	2.6
1	DemonsCPUapp	4.79 ±1.85	−1.54	0.0128 ±0.0050	0.026	0.00	7	1.3
	No registration	6.34 ±2.89		0.0170 ±0.0077	0.000	0.00

The order averages each method's ranks on TRE mm, SDlogJ, so the first one is not always the best in every column. ± is the 95% interval of the mean over the 14 pairs. Two neighbours are tied when a paired Wilcoxon test on the shared pairs does not find the upper one better on TRE mm (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

Measure

TRE mm	1	2	3	4	5	6	7	8	9	10	11	12	13	14
ConvexAdam (MIND)	1.46	1.96	1.17	1.90	6.32	1.13	1.40	1.32	2.02	2.95	2.23	2.18	2.60	2.63
Demons	1.97	2.68	1.18	2.16	5.97	3.66	1.76	1.79	5.23	11.79	7.72	7.03	5.24	8.92
No registration	6.72	2.34	1.28	2.03	7.01	2.27	1.62	2.88	5.64	18.24	11.67	7.77	6.57	12.70

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: 14 test pairs drawn with seed 2026 among the 140 training patients, then 12 development pairs among the 126 left, one pair a patient. The challenge's validation landmarks are not public, so only its training pairs are used.
Truth: landmarks by experts the dataset's expert landmarks
Source: zenodo.org/records/14642405
bratsreg.github.io
Licence: CC BY-ND 4.0 (Zenodo); the organisers ask not to redistribute the data

12 pairs for tuning, never tested

BraTSReg_001, BraTSReg_002, BraTSReg_003, BraTSReg_053, BraTSReg_122, BraTSReg_018, BraTSReg_011, BraTSReg_067, BraTSReg_028, BraTSReg_076, BraTSReg_034, BraTSReg_026