fidoptM-CRIB 2.0 and M-CRIB-WM
Skull-stripped brain T2 MR of two healthy term neonates, with each brain's manual M-CRIB 2.0 parcellation (the ten brains of the M-CRIB atlases).
Read before comparing (8)
- Truth. M-CRIB 2.0 is one person's revision, region by region in ITK-SNAP, of the original M-CRIB, drawn by hand on the T2 by four people from an automatic tissue classification. It follows Mindboggle-101's Desikan-Killiany-Tourville protocol and gives the regions with unclear or arbitrary boundaries to the gyri around them. No agreement between raters is reported. The original 100 regions (osf.io/2yx8u) are not used.
- White matter. M-CRIB-WM was drawn by one person on the T2 and the diffusion tensor's colour map, and checked on every brain by a neurosurgery research fellow and a paediatric radiologist. Some boundaries follow the fibres, which a T2 does not show, and some are arbitrary: shown (dice_wm), never ranked.
- Alignment. The images' geometric centres put together, written in the moving image's and labels' headers, no label used. On the three test pairs that gives a Dice of 0.36 (0.30 to 0.41), equal on every pair to an elastix translation, against 0.19 as delivered (0.00 on P01_P09).
- Images. The authors' preprocessed T2 the labels were drawn on: bias-corrected (N4ITK), skull-stripped (FSL's BET), aligned on the AC-PC line and resampled to 0.63 mm. Values as delivered, 32-bit floats without unit, with 300 to 1,700 negative voxels around each brain, outside its labels. No mask is given.
- Contrast. The white matter, not yet myelinated, is brighter than the cortex on these T2, the reverse of an adult's, and the CSF brighter still: a method trained on adult brains meets another contrast.
- Field of view. It ends close to the top of the head: the labels reach the top slice on four of the ten brains. On P10_P08, after the start, 0.2 % of the fixed brain's CSF lies outside the moving image, for every method alike.
- Subjects. Ten healthy term-born neonates (four girls, six boys), scanned at 40.3 to 43.0 weeks in natural sleep at 3 T at the Royal Children's Hospital, Melbourne, as controls of preterm studies. Three test pairs separate only large differences between methods.
- Outside numbers. None: the pairing is ours. The atlases' templates and probability maps were made from these ten brains, so a method that uses them has seen the test cases.
| Accuracy | Smoothness | Cost | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| # | Method | Dice | Change | HD95 mm | SDlogJ | Folds % | Time s | RAM GB | GPU GB | Images |
| 11–2 | ConvexAdam (MIND)GPUapp | 0.580 ±0.119 | +0.220 | 3.56 ±2.59 | 0.038 | 0.01 | 10 | 2.8 | 3.9 | |
| 21–2 | DemonsCPUapp | 0.473 ±0.149 | +0.114 | 4.36 ±3.11 | 0.046 | 0.00 | 13 | 2.0 | ||
| No registration | 0.359 ±0.146 | 5.12 ±3.34 | 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 3 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
- The three pairs outside development, of the five the ten subjects make, each subject in one pair. The draw alone would take all five as test cases, so, as on UltraCortex, the last two pairs of the seeded order are development and the draw takes the other three.
- Truth
- contours by experts wm by experts the dataset's manual M-CRIB 2.0 parcellation of each brain, 94 regions: the 62 cortical regions of the Desikan-Killiany-Tourville protocol, an 'unknown' label on each side for unassigned voxels in front of the amygdala, and 30 others (cerebral white matter, ventricles, CSF, basal ganglia, thalami, hippocampi, amygdalae, ventral diencephalon, cerebellum, brainstem, corpus callosum)
- Source
- osf.io/4vthr
osf.io/mnwv9
osf.io/2yx8u
doi.org/10.3389/fnins.2019.00034
doi.org/10.1002/hbm.24948
doi.org/10.1016/j.neuroimage.2016.09.068 - Licence
- CC BY 4.0
2 pairs for tuning, never tested
P05_P07, P06_P03