fidoptSynthRAD2023 CBCT/CT, brain
The CT (fixed) and the CBCT (moving) of one patient's brain, rigidly aligned as SynthRAD2023 Task 2 delivers them, with TotalSegmentator labels on both.
Read before comparing (4)
- Truth. TotalSegmentator's task total, through its KonfAI app at a pinned revision, run with one recipe on the CT and on the CBCT. A method that registers on that network's features is scored against its own view of the anatomy.
- Specks. The network also marks structures of the chest, abdomen, pelvis and legs in these head scans, kept as segmented. Most lie on one image only, femurs, hips and gluteal muscles on the CBCT above all, and are never scored, since a structure is scored only where both images hold it. Three test pairs hold some on both images, scored like the rest of the case: lungs, colon, liver, small bowel and bladder in brain_2BA007, the sacrum in brain_2BA009, lungs, colon, kidney cyst, prostate, small bowel, stomach and bladder in brain_2BC011.
- Baseline. The pair is used as the challenge delivers it, rigidly aligned, with the challenge's body mask, the same for both images. With no registration the 18 test pairs score a label Dice of 0.68 (0.22 to 0.89).
- Overlap. SynthRAD2023's centre B holds the same 60 patients in Task 1 and Task 2 under one number: 2BBnnn has 1BBnnn's CT, correlating at 0.99 after a rigid alignment, against at most 0.77 between two patients. No case here is a patient of SynthRAD_BrainMRCT, SynthRAD_BrainMRCT_Syn or SynthRAD_BrainCBCTCT_Syn, under either number.
| Accuracy | Smoothness | Cost | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| # | Method | Dice | Change | HD95 mm | SDlogJ | Folds % | Time s | RAM GB | GPU GB | Images |
| 1 | greedyCPUapp | 0.702 ±0.097 | +0.038 | 22.84 ±14.71 | 0.039 | 0.00 | 28 | 2.4 | ||
| No registration | 0.664 ±0.111 | 23.39 ±14.80 | 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.
2 results 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)GPUapp11/18 pairs | 0.737 ±0.104 | +0.073 | 12.29 ±7.66 | 0.083 | 0.71 | 6 | 2.6 | 3.2 | Laptop | ||
| NiftyRegCPUapp11/18 pairs | 0.740 ±0.116 | +0.076 | 12.51 ±8.56 | 0.036 | 0.05 | 48 | 1.2 | 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), chosen before any result, one case a patient, centres 2BA, 2BB and 2BC alternated; centre-B patients that another dataset holds are skipped, and no patient is in both. The test list takes each centre's consecutive patients in ascending order from 2BA004, 2BB091 and 2BC001, skipping development patients and 2BB102. The development list holds 2BA's first three, seven others and each centre's last two in descending order, for 2BB 2BB198 and 2BB184 (2BB205 and 2BB200 are held elsewhere, 2BB189 is among the seven).
- 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 and on the CBCT
- Licence
- CC BY-NC 4.0
16 pairs for tuning, never tested
brain_2BA001, brain_2BA002, brain_2BA003, brain_2BA045, brain_2BB095, brain_2BB059, brain_2BB189, brain_2BC043, brain_2BC046, brain_2BC018, brain_2BA094, brain_2BB184, brain_2BC090, brain_2BA092, brain_2BB198, brain_2BC089