fidoptPituitary neuroendocrine tumour MRI
Contrast-enhanced T1 MR of a patient before surgery for a pituitary neuroendocrine tumour (adenoma), the tumour and both internal carotid arteries segmented semi-automatically by neurosurgical residents and checked by experts (Pandit et al. 2025).
13 test casesbrainimage MR
Read before comparing (11)
- Cohort. 136 patients operated on for a pituitary adenoma at a London tertiary centre from July 2021 to December 2023, recurrences, other pathologies and degraded images excluded: 114 macroadenomas and 22 microadenomas, of 0.2 to 57 ml.
- Images. The 3D contrast-enhanced T1 acquired for surgical navigation at 3 T. The provider corrected its bias field with FSL's FAST, resampled it to 0.5 mm isotropic and cropped the face away: the values are bias-corrected 32-bit floats without unit, not normalised.
- T2. A T2 registered to the T1, which the annotators read where it helped, exists for 100 patients but is not given: a score here is a score from the T1 alone.
- Truth. Neurosurgical residents and students grew region-competition snakes in ITK-SNAP from spheres placed by hand, adjusted them by hand on the T1 and the T2, then smoothed and eroded them; an attending neurosurgeon and neuroradiologists checked every one against written criteria until it met them. No agreement between readers is published. It meets the bar of a main leaderboard.
- Labels. 1 is the tumour with its cysts; where the gland is distinct, as in most microadenomas, it holds the tumour alone, otherwise tumour and gland together, so on a microadenoma a method cannot know which. 2 is the cavernous, supraclinoid and ophthalmic segments of both internal carotids, never overlapping the tumour. The labels have no side.
- Left out. 73 and 75 have no carotid mask and 132's is empty, so their carotids would score as false positives. 87_T1.nii.gz is 2_T1.nii.gz byte for byte and 125_T1.nii.gz is 124_T1.nii.gz: 87 and 125 lack their image, and 2 cannot be told from 87.
- Files. prepare.py mends the masks of 8, 9, 14, 16, 21 and 80, cut to their T1 or placed on its grid, no labelled voxel lost. 167 of the 268 other mask files carry a qform 6 to 55 mm from their sform, so every built map carries the T1's header only; development case 80 is the one image not resampled.
- Checks. On every case built the tumour lies in the sella under the chiasm, a carotid on each side. The tumour volumes equal the spreadsheet's on all 136 patients, within 0.002 mm3.
- Room. The test tumours hold 0.66 to 9.53 ml, the carotid labels 0.61 to 2.81 ml. On the first ten test maps a one-voxel (0.5 mm) shift keeps a Dice of 0.92 to 0.96 for the tumour and 0.84 to 0.93 for the carotids, which the surface Dice at 1 mm does not see. A map without the carotids scores 0.5.
- Outside numbers. None: the data paper reports no network trained on the dataset. TotalSegmentator makes neither structure from MR, so no network check was run.
- Overlap. No other dataset of the corpus holds these patients: VestibularSchwannomaMCRC comes from the same hospital, but from patients with a vestibular schwannoma seen from 2012 to 2014.
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
The first test caseClick to look at every case.
- Test cases
- 13 test patients drawn with seed 2026 among the 130 whose files hold their own image and both masks, without strata (one centre, one sequence), then 11 development patients, the last of the same seeded order. Six of the 136 patients are left out, as the notes say.
- Truth
- contours by a program, verified by people the dataset's two masks in one map: 1 the tumour, or the tumour and the gland together where the two cannot be told apart, and 2 the cavernous, supraclinoid and ophthalmic segments of both internal carotid arteries; region-competition snakes grown in ITK-SNAP from spheres placed by hand and adjusted by hand on the T1 and the T2, by neurosurgical residents and students, every segmentation checked against written criteria by an attending neurosurgeon and neuroradiologists and edited until it met them
- Source
- doi.org/10.6084/m9.figshare.27894084.v1
doi.org/10.1038/s41597-024-04218-8
ndownloader.figshare.com/files/50793273
ndownloader.figshare.com/files/50757546 - Licence
- CC BY 4.0
11 cases for tuning, never tested
33, 59, 80, 8, 60, 46, 99, 17, 10, 79, 131