fidoptWAW-TACE
One CT phase of a patient with hepatocellular carcinoma before a first transarterial chemoembolisation: the phase on which all their liver tumours were segmented (WAW-TACE, Medical University of Warsaw).
Read before comparing (9)
- Pool. Multiphase CTs of 233 treatment-naive patients with unresectable hepatocellular carcinoma, within 90 days before their first TACE. All of a patient's tumour masks lie on one phase: late arterial for 126 patients, portal venous for 74, delayed for 30, before contrast for 3, none of whom was drawn. The other phases and the organ masks, TotalSegmentator's, are left out.
- Truth. Radiologists corrected and validated a TotalSegmentator extension's proposal, smoothing each mask with a 3 mm Gaussian filter. The paper gives no agreement between readers and does not say how much of the proposal remains. Graded expert with that doubt: a method close to TotalSegmentator's liver tumour model may be favoured, and no boundary finer than 3 mm is in the truth.
- TotalSegmentator. Version 2.18's liver_tumor scores a Dice of 0.899 on 18_2 (388.1 mL) and writes no tumour on 510_1 (19.4 mL in two tumours), so the truth is not that model's output.
- Labels. The masks may miss tumours: the clinical table's lesion count, capped at 4, matches the masks for 207 of the 233 patients, is larger for 9 and smaller for 17. Test case 487_2 lists 2 lesions for 3 masks; development cases 375_3, 493_3 and 522_1 one more lesion than masks. A tumour without a mask counts against a method that finds it.
- Cases. The first ten test cases drawn are 6 portal venous and 4 late arterial, with 4.0 to 388.1 mL of tumour (median 39 mL) in 1 to 3 pieces. The other 13 are 7 portal venous, 5 late arterial and 1 delayed, with 3.5 to 327 mL (median 49 mL).
- Images. The deposit's CTs byte for byte, in Hounsfield units. On the first ten test cases the slices lie 0.625 to 5 mm apart, and nine read -3024 outside the reconstructed circle. Development case 493_3 is stored as 32-bit floats. No mask is given.
- Masks. Every map holds 0 and 1 on its CT's grid. Only test case 158_2's deposit mask lies off that grid; prepare.py places it by its position in the world, where it keeps its 83.6 mL on the liver mass.
- Overlap. No other dataset of the corpus holds these CTs, and no descriptor names Warsaw or this deposit; HCCTACE is the same disease and treatment in another cohort, at MD Anderson. A later dataset built from this deposit must leave these 44 patients' other CTs out or keep each patient on the same side.
- Outside numbers. None: the deposit has no challenge, no official split and no surface tolerance.
| Accuracy | Cost | |||||||
|---|---|---|---|---|---|---|---|---|
| # | Method | Dice | Surface Dice | HD95 mm | Time s | RAM GB | GPU GB | Images |
| 1 | KonfAI segmentation app · TotalSegmentator (liver_vessels), the authors' packageGPUapp | 0.572 ±0.158 | 0.325 ±0.136 | 43.15 ±22.46 | 32 | 21.5 | 3.0 | |
The order averages each method's ranks on Dice, Surface Dice, HD95 mm, so the first one is not always the best in every column. ± is the 95% interval of the mean over the 23 cases. Two neighbours are tied when a paired Wilcoxon test on the shared cases 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 Workstation.
Dice: how much the organ contours overlap, 1 is perfect. Surface Dice: the share of the contours' surfaces within a tolerance of each other, 1 is perfect. HD95: distance between the contours' surfaces.
Case by case
Click a score to look at that result: the image, the truth in green and the method's map in red. Underlined scores have their image; for the others you see the case without the result.
About the data
- Test cases
- 23 test cases drawn with seed 2026 among the 212 patients outside development, out of the deposit's 233, and 21 development cases, the last of the same seeded order. Names follow the deposit, <patient>_<phase>: 1 late arterial, 2 portal venous, 3 delayed. The provider publishes no test set.
- Truth
- contours by experts every hepatocellular carcinoma the deposit holds a mask of, under one label: the proposal of a TotalSegmentator extension corrected in 3D Slicer by a radiologist with 6 years of experience, who outlined the tumour on axial sections, filled between them and smoothed the mask with a 3 mm Gaussian filter, then validated and modified by two radiologists with 14 and 11 years of experience (Bartnik et al. 2024)
- Source
- zenodo.org/records/12741586
doi.org/10.1148/ryai.240296
pmc.ncbi.nlm.nih.gov/articles/PMC11605144 - Licence
- CC BY 4.0
21 cases for tuning, never tested
57_3, 46_2, 375_3, 188_2, 493_3, 522_1, 517_1, 14_1, 394_2, 144_1, 33_1, 134_3, 447_1, 381_1, 524_1, 349_1, 179_1, 326_3, 19_3, 48_2, 539_1