fidoptLiverHccSeg, arterial liver MR
Arterial-phase liver MR of one patient with hepatocellular carcinoma from TCGA-LIHC, the liver and the tumour delineated by hand. The arterial image is used as delivered, without the provider's warps of the other phases.
Read before comparing (4)
- Size. With the Learn2Reg patient left out, only six independent patients remain for development. This small cohort is ranked apart.
- Truth. Rater 1's masks are used for every case, a choice fixed before the draw; rater 2's are an alternative, not combined with them. The tumour masks are unioned without interpolation. A case without a tumour file has no residual tumour in this release; none is invented.
- Labels. The liver label leaves out the tumour's voxels, as KiTSCT does. A whole-liver score joining labels 1 and 2 would be another measurement.
- Overlap. TCGA identifiers are kept. The patients of TotalSegMR's and LiverSegmentsMR's IDC scans are unknown, so the overlap with them cannot be fully checked, and independence from them is not claimed.
| Accuracy | Cost | |||||||
|---|---|---|---|---|---|---|---|---|
| # | Method | Dice | Surface Dice | HD95 mm | Time s | RAM GB | GPU GB | Images |
| 11–2 | KonfAI segmentation app · TotalSegmentator (total_mr), the authors' packageGPUapp1 of 2 structures | 0.898 ±0.041 | 0.374 ±0.042 | 10.43 ±4.49 | 23 | 12.5 | 2.8 | |
| 21–2 | KonfAI segmentation app · TotalSegmentator (total_mr), KonfAI appGPUapp1 of 2 structuresnote | 0.893 ±0.041 | 0.361 ±0.045 | 11.54 ±5.29 | 27 | 2.5 | 4.0 | |
| 3 | KonfAI segmentation app · MRSegmentator, KonfAI appGPUapp1 of 2 structures | 0.001 ±0.001 | 0.000 ±0.000 | 189.59 ±42.55 | 2.5 | 6.1 | ||
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 10 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.
KonfAI segmentation app · TotalSegmentator (total_mr), KonfAI app. The total_mr model has no liver tumour class: on 7 of 10 rows liver_tumour is reported missing. This is the model's label coverage, not a measure of its tumour segmentation.
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
- 10 test patients drawn with seed 2026 among the 16 eligible, one examination each, and the other 6 in development. The archive holds 17 patients: TCGA-DD-A4NJ is left out because L2R_AbdomenMRCT already uses that patient.
- Truth
- contours by experts Rater 1: manual liver and tumour masks under the supervision of two board-certified abdominal radiologists. Label 1 is liver excluding tumour; label 2 combines all available rater-1 tumour masks, with tumour taking precedence. The three patients without residual tumour have liver-only truth.
- Source
- zenodo.org/records/8179129
doi.org/10.5281/zenodo.8179129
www.sciencedirect.com/science/article/pii/S2352340923007473
doi.org/10.7937/K9/TCIA.2016.IMMQW8UQ - Licence
- CC BY 4.0
6 cases for tuning, never tested
TCGA-BC-A3KG, TCGA-BC-A5W4, TCGA-G3-AAV7, TCGA-BC-A10Y, TCGA-G3-AAV1, TCGA-G3-A25T