Learn2Reg 2021, OASIS, validation

Brain MR of two subjects of OASIS, a validation pair of Learn2Reg 2021 Task 3, with 35 labelled structures.

19 test pairsbraininter-patientfixed MR, moving MRdelivered alignedresearch only

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	0.572 ±0.026		3.83 ±0.36	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.

Scored the way the challenge scores

The same results with the Learn2Reg 2021, OASIS challenge's own measures, on its validation pairs: mean ± standard deviation.

Method	DSC	HD95	SDlogJ
No registration	0.572 ± 0.053	3.831 ± 0.738	0.000 ± 0.000

What the challenge published

On the challenge's test set: other pairs than the ones above, and entries trained or tuned for the task. It shows the level of the challenge; these are not results of this benchmark.

Method	DSC	HD95	SDlogJ
No registration	0.560	3.86
LapIRN	0.820	1.67	0.070
ConvexAdam	0.810	1.63	0.070
lWM	0.790	1.84	0.050
Driver	0.800	1.77	0.080
PIMed	0.780	1.86	0.060
3Idiots	0.800	1.82	0.080
Winter	0.770	2.16	0.080
MEVIS	0.770	2.09	0.070
Multi-brain	0.780	1.92	0.570
corrField	0.740	2.36	0.080
Thorley	0.770	2.21	0.310
NiftyReg	0.730	2.37	0.060
Bailiang	0.670	2.74	0.040
LaTIM	0.740	2.31	0.080
Imperial	0.760	2.43	0.190

Source: Hering et al., Learn2Reg: comprehensive multi-task medical image registration challenge, dataset and evaluation in the era of deep learning, IEEE TMI 2023, Table VII.

Pair by pair

Measure

Dice	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19
No registration	0.591	0.555	0.629	0.496	0.520	0.607	0.452	0.594	0.595	0.589	0.525	0.525	0.551	0.683	0.598	0.631	0.574	0.570	0.579

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: The 19 test pairs are the task's official validation pairs. The 10 development pairs join, two by two in the order of sha256("2026:L2R_OASIS:<subject>"), the training subjects 0001 to 0394 that no validation pair holds.
Truth: contours by a program the challenge's validation labels
Source: learn2reg.grand-challenge.org
github.com/MDL-UzL/L2R
Licence: OASIS Data Use Agreement

10 pairs for tuning, never tested

0346_0000_0288_0000, 0274_0000_0320_0000, 0230_0000_0004_0000, 0162_0000_0283_0000, 0174_0000_0188_0000, 0110_0000_0344_0000, 0051_0000_0196_0000, 0002_0000_0385_0000, 0111_0000_0039_0000, 0367_0000_0357_0000