fidoptSynthRAD2023 MR/CT, pelvis
CT (fixed) and MR (moving) of one patient's pelvis, rigidly aligned by the challenge (SynthRAD2023, Task 1).
Read before comparing (3)
- Truth. Every case's labels are TotalSegmentator's, through its KonfAI app at a pinned revision: a method that registers on that network's features is scored against its own view of the anatomy.
- Alignment. The pair is used as the challenge delivers it, rigidly aligned. The mask is the challenge's body mask, the same for both images.
- Baseline. With no registration the 18 test pairs score a label Dice of 0.69 (0.49 to 0.79).
| Accuracy | Smoothness | Cost | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| # | Method | Dice | Change | HD95 mm | SDlogJ | Folds % | Time s | RAM GB | GPU GB | Images |
| 1 | NiftyRegCPUapp | 0.699 ±0.033 | +0.004 | 13.86 ±2.95 | 0.069 | 0.05 | 63 | 1.6 | ||
| No registration | 0.694 ±0.040 | 14.17 ±3.15 | 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 18 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.
1 result not ranked: an older run, pairs still missing, or a failure on every pair
| Accuracy | Smoothness | Cost | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| # | Method | Dice | Change | HD95 mm | SDlogJ | Folds % | Time s | RAM GB | GPU GB | Machine | Images |
| ConvexAdam (MIND)GPUapp8/18 pairs | 0.749 ±0.027 | +0.054 | 12.08 ±3.38 | 0.154 | 1.97 | 8 | 2.9 | 4.3 | 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
- 18 test patients (10% of the 180 training patients) and 16 development patients (10% of those left), one a patient, chosen before any result, centres 1PA and 1PC alternated. The test list takes each centre's consecutive patients in ascending order, from 1PA144 and from 1PC000. The development list holds 1PA's first three, seven others and each centre's last three in descending order. No patient is in both.
- Truth
- contours by a program TotalSegmentator through its KonfAI app (VBoussot/TotalSegmentator-KonfAI, revision 07185697, konfai-apps of KonfAI 1.8.7), run by us on each image with one recipe (synthrad.py): task total on the CT, task total_mr on the MR, both reduced to the structures the two tasks name, in total_mr's ids
- Licence
- CC BY-NC 4.0
16 pairs for tuning, never tested
pelvis_1PA001, pelvis_1PA004, pelvis_1PA005, pelvis_1PA127, pelvis_1PA171, pelvis_1PC029, pelvis_1PC082, pelvis_1PC039, pelvis_1PC041, pelvis_1PC065, pelvis_1PA188, pelvis_1PC098, pelvis_1PA187, pelvis_1PC097, pelvis_1PA185, pelvis_1PC096