fidoptBrainWeb, 20 normal anatomical models
Two simulated T1-weighted head MR images of two of BrainWeb's 20 digital brain phantoms, each with the tissue classes it was simulated from.
10 test pairsbraininter-patientfixed MR, moving MRdelivered alignedresearch only
Read before comparing (6)
- Phantoms. Twenty digital brain phantoms, each built from the MR scans of a normal adult: fuzzy and discrete models of twelve classes at 0.5 mm in MNI stereotaxic space. For these, BrainWeb offers one simulation only, a T1 (SFLASH, TR 22 ms, TE 9.2 ms, flip angle 30 degrees, 1 mm), with the simulator's noise, as delivered.
- Truth. Graded synthetic: the classes are those the images were simulated from, exact up to how each 1 mm voxel takes the class that fills most of it, but the images carry no real scanner's artefacts and the anatomy is that of 20 phantoms, not a population.
- Alignment. As delivered: every phantom lies in MNI stereotaxic space, so a pair starts linearly aligned. Without registration the ten pairs have a mean Dice of 0.315 over the eleven classes (0.288 to 0.351; grey matter 0.629, white matter 0.625) and an HD95 of 9.23 mm (6.67 to 12.09).
- Box. The models stop at -90 to 90, -126 to 90 and -72 to 108 mm (RAS), below the cerebellum; the T1 grid is wider and holds only noise outside, labelled 0.
- Overlap. BrainWebSim segments these same 20 images and keeps this split. No other descriptor names BrainWeb, and no other built store has the T1's grid.
- Licence. None stated; the pages ask that their references be cited. The download form's name and e-mail fields are optional and left empty: no account.
No method has a result on every pair of this dataset yet.
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
- The 20 phantoms paired at random into 10 pairs (prepare.py --pool), all of them test pairs: ten independent pairs leave none for development.
- Truth
- contours a known deformation the phantom's discrete anatomical model (0.5 mm, the class that contributes most to each voxel), brought to the T1's 1 mm grid by the class filling most of each voxel; the T1 is simulated from the phantom's fuzzy tissue models, so the classes are exact by construction
- Source
- brainweb.bic.mni.mcgill.ca/brainweb/anatomic_normal_20.html
brainweb.bic.mni.mcgill.ca/brainweb
doi.org/10.1109/TMI.2006.883453 - Licence
- none stated: the BrainWeb pages ask only that their references be cited