fidoptKiTS23 CT/CT
Preoperative contrast-enhanced abdominal CTs of two patients operated on or ablated for a suspected kidney cancer in the M Health Fairview system, Minnesota, paired by us from KiTS23's training set, with TotalSegmentator's organs on both and the provider's kidneys.
11 test pairsabdomeninter-patientfixed CT, moving CTaligned once by the benchmarkresearch only
Read before comparing (7)
- Pool. KiTS23's 489 training cases are patients who had a cryoablation or a partial or radical nephrectomy for a suspected renal malignancy from 2010 to 2022, each the most recent contrast-enhanced preoperative CT, corticomedullary or nephrogenic. Cases 00000 to 00209 (C4KC-KiTS) are KiTSCT's, and KiTS23's own 53 test and development cases are KiTS23's: both are left out, so no patient sits in two datasets. 226 patients remain, paired in the seed's order with no regard to sex, age or phase.
- Truth. Between two patients only shared organs correspond: the ranked labels are TotalSegmentator's organs on each CT, made by a network, graded automatic and ranked apart. The provider's kidneys, drawn by trained annotators and reviewed, their tumours and cysts made one with them, are shown as the set kidneys and never ranked: a tumour or a cyst is one patient's, and two kidneys with masses of different sizes do not correspond voxel for voxel.
- Images. The repository's NIfTI files as delivered, in Hounsfield units, slices 0.5 to 5 mm apart; their headers put the slice axis first and the origin at 0, so positions are not the scanner's, but the orientation is anatomical. Some scans cover the kidneys only, others the abdomen and pelvis.
- Alignment. The scans keep their native voxels, each origin at zero. The moving image's header carries one translation from elastix mutual information inside each body, the best of several starts, no annotation used; the size and posture of two patients are left to the methods.
- Baseline. With no registration the 11 test pairs start at a mean Dice of 0.16 at the median (0.08 to 0.31) over the organs both maps hold and an HD95 of 53 mm (24 to 95), the kidney cysts left out; the provider's kidneys at 0.27 (0.12 to 0.66).
- Overlap. KiTSCT, SAROSBodyCT and AorticTreeCTA hold C4KC-KiTS patients, KiTS23 its 53: all left out. No other dataset of the corpus names KiTS21 or KiTS23.
- Outside numbers. None: the pairing is ours.
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
- 11 test pairs drawn with seed 2026 among the 113 pairs, and 10 development pairs. prepare.py puts the 226 patients that neither KiTSCT (C4KC-KiTS, cases 00000 to 00209) nor KiTS23 (its test and development cases) holds in the seed's order and pairs them two by two.
- Truth
- contours by a program kidneys by a program, verified by people TotalSegmentator 2.18 (task total) run by us on each CT, in Hounsfield units, its kidney cysts left out
- Source
- kits-challenge.org/kits23
github.com/neheller/kits23
huggingface.co/datasets/neheller/KiTS-Challenge-Imaging
arxiv.org/abs/2307.01984 - Licence
- CC BY-NC-SA 4.0
10 pairs for tuning, never tested
00280_00216, 00548_00408, 00244_00485, 00546_00225, 00227_00490, 00584_00545, 00583_00446, 00511_00235, 00460_00492, 00431_00299