fidopt Interactive view
Abdomen MR, segmentation

PanSegData, pancreas on T2-weighted MR

A T2-weighted abdominal MR of a participant referred for a pancreatic cyst or a suspected pancreatic cancer at one of five US centres, with the pancreas outlined by a radiologist (PanSegData).

30 test casesabdomenimage MRresearch only
Look at the images
Read before comparing (7)
  • Pool. The deposit's 382 T2-weighted abdominal MRs, one a participant, of patients referred for a pancreatic cyst or a suspected pancreatic adenocarcinoma at five US centres from 2004 to 2022, on 1.5 and 3 T Siemens and GE scanners. The 385 T1-weighted scans are mostly of the same participants (362 of the 382 by the name mappings, the T1 and the T2 of one participant sharing a grid at three centres) and are left out.
  • Images. The deposit's NIfTI files byte for byte, values without unit: 0.74 to 1.68 mm in plane and slices 3.3 to 7.8 mm apart on the test cases, stacks of single-shot slices on many. No mask is given.
  • Truth. One radiologist a centre drew the pancreas by hand in ITK-SNAP on the axial slices after a shared protocol, and a senior radiologist checked every mask. On 50 scans two readers agreed at a Dice of 0.81 on T2 (0.80 on T1), one reader with himself at 0.94 four weeks later. Graded expert.
  • Cases. The test pancreases hold 22 to 140 mL (median 76 mL); the deposit gives no diagnosis a scan, so cysts and tumours inside the gland are labelled pancreas when the reader drew them so.
  • Start. Over the 30 test cases a whole-image mask scores a Dice of 0.006 at the median, an Otsu mask 0.003. TotalSegmentator 2.18.0 (total_mr) on test case NYU_0030 scores 0.29 for the pancreas (12.7 mL against 56.5 mL).
  • Overlap. No other dataset of the corpus names PanSegData, its centres or the deposit. Within it, two pairs of Mayo Clinic Florida scans share the sheet's sex, age and scanner and an image grid (MCF_0044 and MCF_0074, MCF_0067 and MCF_0090), as do five Allegheny scans whose sheet gives no age; the second scan of each pair and four of the five Allegheny scans are held out. A dataset built later from the T1 scans must keep these 60 participants on the same sides.
  • Outside numbers. PanSegNet reports a Dice of 0.863 on T2 in the authors' own split, not these cases.

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 scans drawn with seed 2026 among the 352 outside development, each centre in proportion (13 NYU, 12 Mayo Clinic Florida, 2 Allegheny, 2 Mayo Clinic Arizona, 1 Northwestern), then 30 development scans, the last of the same seeded order. One T2-weighted scan a participant; the provider publishes no split.
Truth
contours by experts the pancreas drawn by hand in ITK-SNAP on the axial T2-weighted scan by one radiologist a centre, after a shared protocol, and double-checked by a senior radiologist (Zhang et al. 2025)
Licence
CC BY-NC 4.0
30 cases for tuning, never tested

NYU_0143, MCF_0118, NYU_0071, MCF_0095, NYU_0132, MCF_0167, AHN_0013, NYU_0039, MCA_0008, MCF_0019, NWU_0019, NYU_0009, MCF_0029, NYU_0083, MCF_0172, NYU_0042, MCF_0154, NYU_0054, MCF_0121, NYU_0024, AHN_0021, MCF_0065, NYU_0063, MCF_0007, NYU_0120, MCA_0015, MCF_0005, NYU_0031, MCF_0063, NWU_0006