fidoptPediatric-CT-SEG
Chest-abdomen-pelvis or abdomen-pelvis CT of two children of one sex and about one age, from three scanners of one hospital, with up to 28 structures contoured (Pediatric-CT-SEG), paired by us.
Read before comparing (9)
- Pairs. In each sex the children are sorted by age, then by the seed, and paired two by two: one age in years, one year apart for 11 of the 179 pairs. One girl is left out. One sex a pair lets the gonads label one organ, and the age bands bring infants, school children and teenagers.
- Truth. Four medical analysts drew 25 structures by hand, the lungs and bones came from thresholds corrected by hand, and a radiation oncologist reviewed them all. The skin, a 2-mm band under the body's outline, is left out.
- Structures. A structure out of the scan range or too noisy on a low-dose scan was not drawn, and the prostate and uterus belong to one sex. A structure is scored on a pair when both children hold it, 17 to 27 on the test pairs.
- Alignment. The scans come in their own frames, up to 0.6 m apart. The moving CT's and labels' headers carry one elastix translation on mutual information inside each body, no structure used. Size differences, and a chest-abdomen-pelvis scan against an abdomen-pelvis one, are left to the methods.
- Baseline. With no registration the 17 test pairs start at a Dice of 0.28 (0.17 to 0.47) and an HD95 of 35.1 mm (11.1 to 78.2). A rigid fit on the structures' centres reached 0.34 on the first ten.
- Scans. Protocols differ in voltage, dose, reconstruction and contrast agent. Some scans show central lines, feeding tubes, wires or diapers, and some organs have an unusual size: the fixed child of 3B5CED31_397 has a large liver and spleen.
- Slices. 297 scans were resampled to 2-mm slices before contouring, which corrupted the first or last slices of some; they are left so. Child 377 misses two slices in its image and contours, each filled from its nearest neighbour.
- Broken pair. Child D1572D48's structure set puts the large intestine around both shoulders, a femoral head in the right shoulder and the small intestine outside the body. Its pair stays in the pool so that the draw is unchanged.
- Outside numbers. None: the pairing is ours.
No method has a result on every pair of this dataset yet.
| Accuracy | Smoothness | Cost | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| # | Method | Dice | Change | HD95 mm | SDlogJ | Folds % | Time s | RAM GB | GPU GB | Images |
| No registration | 0.283 ±0.046 | 35.13 ±8.39 | 0.000 | 0.00 | ||||||
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)GPUapp14/17 pairs | 0.423 ±0.063 | +0.140 | 29.77 ±8.11 | 0.134 | 1.15 | 45 | 5.1 | 13.2 | 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
- 17 test pairs drawn with seed 2026 by age band among the 179, then 16 development pairs among the 162 left. 0D53FEEE_D1572D48, with broken contours, and 0C49C1B7_392, with a horseshoe kidney outside the dataset's table, never become development.
- Truth
- contours by experts up to 28 structures a child, the dataset's contours: 25 organs drawn by hand, the lungs and the bones from a threshold corrected by hand
- Source
- www.cancerimagingarchive.net/collection/pediatric-ct-seg
doi.org/10.7937/TCIA.X0H0-1706
doi.org/10.1002/mp.15485 - Licence
- CC BY 4.0
16 pairs for tuning, never tested
34ECBB32_8877A5A3, 379_385, AA032583_6A2228AD, 7C7F4414_51733317, 4EF8F066_8975C78C, 14403912_CF79E74B, B722C905_B9DA30EF, E69AC901_B0308EC6, 642CA0DF_691FB79F, 381_8538895B, B42CD321_F369B022, 3A4DE292_4934405E, 627003CD_5AEBE68F, 47CA11F8_491B83F3, 211F45E2_29ECC095, C7D4AEFA_792705D7