fidoptAFIDs
Whole-head brain T1 MR of two subjects of one image set, SNSX at 7 T or LHSCPD at 1.5 T, each with the 32 anatomical fiducials of the AFIDs protocol.
Read before comparing (7)
- Truth. The ranked truth is the 32 fiducials of each brain, the mean placement of several trained raters: points of the deep brain and the ventricles, such as the commissures, the mammillary bodies and the tips of the ventricular horns, about 1 mm apart between raters. They say little of the cortex.
- Alignment. Each subject keeps its native voxels and grid. The moving image's header carries one rigid transform from elastix mutual information inside each head, the best of several starts, no annotation used. It turns the heads by 1 to 12 degrees, where a translation alone left four test pairs tilted by 8 to 12. Size and shape are left to the methods.
- Baseline. With no registration the 10 test pairs start at a fiducial error of 6.7 mm (3.5 to 17.5), against 8.4 mm after a translation alone and 2.9 mm after an affine fit on the fiducials, a label Dice of 0.37 (0.17 to 0.47) and an HD95 of 9.1 mm. On sub-C015_sub-C029 the heads line up but one brain sits lower and further back.
- Labels. SynthSeg (labels.py) segments each whole-head T1 at 1 mm, and its map is brought back onto the image's grid by nearest neighbour. The labels are shown beside the fiducials, which stay the ranked truth. All 32 fiducials of every image built fall inside its labels, and every left region lies on the left.
- SynthMorph. It was trained on label maps made by SynthSeg (Hoffmann et al. 2024, doi.org/10.1162/imag_a_00197), so it is scored against labels of the kind it learned from and may be favoured.
- Images. Whole heads, not skull-stripped, with no mask.
- Left out. The protocol's fiducials on OASIS-1 and on the Human Connectome Project are open too, but their images fall under those projects' terms and are not in the corpus.
| Accuracy | Smoothness | Cost | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| # | Method | Dice | Change | HD95 mm | TRE mm | rTRE | SDlogJ | Folds % | Time s | RAM GB | GPU GB | Images |
| 1 | DemonsCPUapp | 0.477 ±0.092 | +0.107 | 7.87 ±3.19 | 5.17 ±2.24 | 0.0133 ±0.0060 | 0.072 | 0.00 | 15 | 2.4 | ||
| No registration | 0.370 ±0.079 | 9.09 ±3.56 | 6.67 ±3.20 | 0.0172 ±0.0085 | 0.000 | 0.00 | ||||||
The order averages each method's ranks on Dice, HD95 mm, TRE 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. 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
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, 5 from each image set: SNSX, healthy adults at 7 T, and LHSCPD, patients with Parkinson's disease at 1.5 T with contrast agent. 10 development pairs.
- Truth
- contours by a program landmarks by experts SynthSeg's segmentation run by us on each T1 (SynthSeg-robust 2.0 with the cortical parcellation, FreeSurfer 7.4.1): 30 structures (white matter, ventricles, deep grey nuclei, hippocampus, amygdala, cerebellum, brainstem, CSF) and 68 Desikan-Killiany cortical regions 32 anatomical fiducials a brain, the mean placement of several trained raters
- Source
- openneuro.org/datasets/ds004470
openneuro.org/datasets/ds004471
doi.org/10.1038/s41597-023-02330-9
github.com/afids/afids-data - Licence
- CC0
10 pairs for tuning, never tested
sub-116_sub-127, sub-145_sub-087, sub-146_sub-149, sub-133_sub-131, sub-C030_sub-C005, sub-163_sub-156, sub-C031_sub-C028, sub-168_sub-115, sub-C013_sub-C002, sub-C022_sub-C010