fidoptPanTrack
The first portal venous CT of a patient's follow-up for pancreatic adenocarcinoma, with the pancreatic tumour a radiologist segmented on it (PanTrack, one German institution).
10 test casesabdomenimage CTresearch only
Read before comparing (8)
- Pool. 45 patients with pancreatic adenocarcinoma followed by portal venous CT at one German institution on identical Siemens protocols, 2 to 11 CTs each; the provider used them only as an out-of-distribution test set and gives a suggested training split. A case is the patient's first CT.
- Truth. One radiologist segmented the pancreatic tumours and the metastases the radiologist judged relevant; no second reader is reported. Graded expert, a single reader's. Pancreatic tumours have fuzzy borders and low contrast, so the outline itself is uncertain.
- Ignore region. Liver metastases and lymph nodes are drawn only in part: those lesions and the liver (the provider's TotalSegmentator label 5) are taken out of the truth and the prediction before scoring, so a lesion found in the liver counts neither way. A lymph node or other lesion found elsewhere still counts against a method.
- Masks. labels/ stores each lesion's identifier modulo 256 (8 bits): organ_annotations.json's 307 is the mask's 51. prepare.py maps them so, checked on all 161 scans.
- Cases. The test tumours hold 9.3 to 28.7 mL (median 19 mL); one test CT also holds a liver or node lesion, inside the ignore region. Slices 0.4 or 0.7 mm, pixels 0.64 to 0.95 mm; the CTs read -8192 or -2025 outside the reconstructed field.
- Overlap. No other dataset of the corpus holds these CTs. PanTrack (registration) draws 20 other patients of the dataset, so no patient is in both.
- Start. A mask of the whole image (ignore region out) scores a Dice of at most 0.001 on the test cases.
- Outside numbers. The provider reports lesion tracking, not segmentation of these first CTs.
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
- 10 test patients drawn with seed 2026 among the 35 outside development, out of the dataset's 45, then 10 development patients, the last of the same seeded order; one CT a patient, the first of the follow-up. The provider gives only a suggested training split; PanTrack (registration) takes 20 other patients.
- Truth
- contours by experts every lesion of the scan's instance mask that organ_annotations.json puts in the pancreas, drawn by a radiologist; liver metastases and lymph nodes, drawn only where the radiologist judged them relevant, are taken out of the scoring with the liver (ignore)
- Source
- huggingface.co/datasets/mrokuss/PanTrack
huggingface.co/datasets/mrokuss/PanTrack/tree/5ce8a560509564148ffe55ad31ebc3da5a89cbfd
arxiv.org/abs/2605.23118
github.com/MIC-DKFZ/LongiSeg - Licence
- CC BY-NC 4.0
10 cases for tuning, never tested
PanTrack_036, PanTrack_027, PanTrack_035, PanTrack_020, PanTrack_026, PanTrack_015, PanTrack_024, PanTrack_001, PanTrack_045, PanTrack_044