Learn2Reg 2021, AbdomenMRCT, validation

An abdominal MR, fixed, and CT, moving, of one patient, with four organs labelled and ROI masks, from Learn2Reg 2021 Task 1's validation.

3 test pairsabdomensame patientfixed MR, moving CTdelivered alignedresearch only

Look at the images

Read before comparing (2)

Board. The board's HD95 is in voxels of the 2 mm grid, half the value in mm shown here, and its SDlogJ is taken inside the MR's ROI mask.
Overlap. The patients are eight of TCIA's TCGA collections, named on Learn2Reg's datasets page: TCGA-B8-5158 (0002), TCGA-B8-5545 (0004), TCGA-B8-5551 (0006), TCGA-BP-5006 (0008), TCGA-DD-A1EI (0010), TCGA-DD-A4NJ (0012), TCGA-G7-7502 (0014) and TCGA-G7-A8LC (0016). No other dataset of the corpus holds them: LiverSegmentsMR leaves out its one image of TCGA-B8-5158. TotalSegMR v3's training split holds other MR series of TCGA-B8-5158, TCGA-DD-A1EI and TCGA-G7-7502: the same patients, not the same images, so a model trained on it has seen three of these patients.

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	0.310 ±0.387		45.78 ±55.50	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, AbdomenMRCT challenge's own measures, on its validation pairs: mean ± standard deviation.

Method	DSC	HD95	SDlogJ
No registration	0.310 ± 0.156	22.889 ± 11.171	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	0.330	48.65
ConvexAdam	0.750	24.92	0.090
corrField	0.760	23.35	0.100
LapIRN	0.760	22.81	0.120
PIMed	0.780	21.99	0.070
MEVIS	0.710	27.94	0.150
Driver	0.760	27.02	0.130
NiftyReg	0.650	33.09	0.120
LaTIM	0.540	41.17	0.130
Winter	0.550	35.51	0.850
Imperial	0.510	48.60	0.110
Multi-brain	0.540	38.21	0.480

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 VI.

Pair by pair

Measure

Dice	1	2	3
No registration	0.247	0.195	0.487

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 test cases are the challenge's 3 official validation pairs. The 5 development cases are its training subjects 0001 to 0005.
Truth: contours by experts the challenge's validation labels
Source: learn2reg.grand-challenge.org
github.com/MDL-UzL/L2R
learn2reg.grand-challenge.org/Datasets
Licence: CC BY 3.0 (TCIA TCGA CT) and CC BY-NC-SA 4.0 (CHAOS MR)

5 pairs for tuning, never tested

case_0001, case_0002, case_0003, case_0005, case_0004