fidoptMindboggle-101
Skull-stripped brain T1 MR scans of two healthy adults from Mindboggle-101 release 3, with 62 cortical regions labelled by hand on both.
Read before comparing (5)
- Truth. Cortical regions drawn by hand on each brain; they say nothing of the deep structures.
- Alignment. The brains keep their native voxels, each in its own space. The moving image's header carries one rigid transform from elastix mutual information inside each brain, the best of several starts, no annotation used. It turns the brains 3 to 20 degrees; a translation alone left them tilted, the brainstem and corpus callosum visibly apart. Size and shape are left to the methods.
- Baseline. With no registration the 10 test pairs start at a Dice of 0.30 (0.24 to 0.34), against 0.24 after the translation alone and 0.34 after an affine fit on the region centres, and an HD95 of 8.0 mm (6.4 to 10.9).
- MNI files. The dataset also gives each brain in MNI152 space, but there the label volume of 14 of these subjects does not sit on its image, so those files are not used.
- Scanners. The subjects come from three acquisition sites, so a pair may join two scanners.
| Accuracy | Smoothness | Cost | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| # | Method | Dice | Change | HD95 mm | SDlogJ | Folds % | Time s | RAM GB | GPU GB | Images |
| 1 | DemonsCPUapp | 0.407 ±0.019 | +0.109 | 6.74 ±0.64 | 0.026 | 0.00 | 9 | 1.7 | ||
| No registration | 0.298 ±0.025 | 8.04 ±0.99 | 0.000 | 0.00 | ||||||
The order averages each method's ranks on Dice, HD95 mm, 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 Dice (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.
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
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 pairs drawn with seed 2026 among the 26 pairs outside development, and 10 development pairs. 73 of the 101 subjects are paired, each once. The 20 OASIS-TRT-20 subjects stay out, being OASIS-1 subjects the Learn2Reg OASIS task trains on, as do two repeat subjects and Extra-18's template, twin and repeat scans.
- Truth
- contours by experts the dataset's manual cortical labels, 62 regions (DKT31 protocol)
- Licence
- CC BY 4.0, over upstream scans under non-commercial terms (NKI-RS) and the BIRN licence (Kirby21)
10 pairs for tuning, never tested
HLN-12-3_NKI-RS-22-17, NKI-RS-22-12_NKI-TRT-20-1, NKI-TRT-20-20_MMRR-21-11, NKI-TRT-20-10_NKI-TRT-20-7, MMRR-21-14_HLN-12-1, NKI-RS-22-15_HLN-12-11, NKI-TRT-20-19_NKI-RS-22-7, NKI-RS-22-21_MMRR-21-16, NKI-RS-22-18_NKI-TRT-20-17, NKI-RS-22-6_NKI-RS-22-5