fidopt Interactive view
Lesions and tumours, segmentation

CancerVerse

A CT of the abdomen, pelvis or chest of a patient with a cancer of one of 13 organs, from hospitals in Switzerland and Turkey, with every tumour of those organs drawn by radiology residents and reviewed by board-certified radiologists (CancerVerse 3.5.1).

30 test casesabdomenimage CTresearch only
Look at the images
Read before comparing (7)
  • Cases. CTs of the abdomen, pelvis or chest as delivered, in Hounsfield units clipped by the provider to -1000..1000; pixels 0.64 to 1.52 mm, slices 0.3 to 4 mm apart, 65 to 1,222 slices. The 30 test CTs hold 39 tumour masks of 9 kinds, 0.25 to 370 mL (median 26 mL): liver 20, kidney 6, and two each or fewer of adrenal, stomach, uterus, pancreas, colon, bladder and oesophagus; 9 cases hold two kinds. A kind absent from a case is not scored there.
  • Truth. Radiology residents drew every tumour in MONAI Label without an AI pre-mask; board-certified radiologists reviewed each mask independently and disagreements were adjudicated (CancerVerse paper). One delineation a case. On a subset of tumours three radiologists drew independently, they agree at a Dice of 0.622, so the outline itself is uncertain.
  • Held out. Three pool scans are held out of the draw: CV_00011242 and CV_00016058, whose masks differ in size in the archive but hold no voxel (no tumour drawn), and CV_00015099, a test case first drawn, which is not an axial abdominal or chest CT but a sagittal spine reformat 78 mm wide (39 x 512 x 1134 voxels of 2.0 x 0.26 x 0.26 mm) whose one tumour is 0.1 mL of liver at the slab's edge.
  • Overlap. Part of CancerVerse comes from University Hospital Basel, as do TotalSegCT, LiverLesionsCT, PericardialEffusionCT and HipImplantCT; the IDs are encrypted, so no identifier links them. Scan level: each of the 60 test and development CTs was compared on content with those sets' 222 CTs (five axial slices at 3 mm, correlated over every shift); its best match scores 0.22 to 0.62, never clear of the next (0.21 to 0.59), while three synthetic copies (corpus CTs regridded) score 0.73 to 1.00 against a next best of 0.36 at most. No case repeats a corpus scan. Patient level it is not settled: CancerVerse is longitudinal (28 of the 60 patients have several scans), and of the four Basel sets only TotalSegCT gives sex and age, without dates; 59 of the 60 patients share sex and age (within a year) with 1 to 12 of TotalSegCT's 113 corpus CTs, so demographics cannot rule out another exam of the same patient.
  • Start. A mask of the whole image scores a Dice of 0.0006 at the median over the test cases and tumour kinds (0.021 at most). TotalSegmentator does not apply: it segments no tumour.
  • No derivatives. CC BY-NC-ND 4.0: the cases are built and scored on the reference machine only and never uploaded; nothing computed from them is shared.
  • Outside numbers. The paper reports a mean tumour Dice of 0.511 on its own test split (2,772 patients), not released, which may include these patients; not comparable.

No method has a result on every case of this dataset yet.

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
30 test patients drawn with seed 2026 among the 4,949 outside development and held out, then 30 development patients, the last of the same seeded order. The pool is one scan a patient: the earliest of the patient's scans with a tumour drawn, 4,982 patients. The paper's official split (2,772 test patients) is not released with the data.
Truth
contours by experts the 13 tumour masks of the scan (<organ>_lesion), drawn by hand in MONAI Label by radiology residents without any AI pre-mask, reviewed independently by board-certified radiologists, disagreements adjudicated (CancerVerse paper, MICCAI 2026)
Licence
CC BY-NC-ND 4.0
30 cases for tuning, never tested

CV_00020024, CV_00021228, CV_00007187, CV_00003033, CV_00021637, CV_00018488, CV_00008294, CV_00006395, CV_00016552, CV_00011165, CV_00017071, CV_00005383, CV_00018119, CV_00000043, CV_00006980, CV_00012303, CV_00016579, CV_00003171, CV_00021651, CV_00015699, CV_00009194, CV_00019895, CV_00007391, CV_00018819, CV_00021514, CV_00011862, CV_00021144, CV_00021923, CV_00010954, CV_00022320