fidopt Interactive view
Lesions and tumours, segmentation

Vestibular-Schwannoma-MC-RC2

Contrast-enhanced T1 MR from the routine follow-up of a patient with a vestibular schwannoma referred to King's College Hospital, London, the tumour outlined by a network and accepted or corrected by experts (Vestibular-Schwannoma-MC-RC2, TCIA).

18 test casesbrainimage MR
Look at the images
Read before comparing (9)
  • Cohort. 190 adults with a single sporadic vestibular schwannoma, neurofibromatosis type 2 excluded, first managed at King's College Hospital between 2008 and 2012, some treated before. Their 621 examinations, 1 to 8 a patient, come from more than 15 hospitals; the provider left out those showing another tumour, not covering the brain or with slices over 3.5 mm.
  • Cases. Masks exist only on contrast-enhanced T1s, for 185 of the 190 patients (the article counts 184). For VS_MC_RC2_033, 074 and 075 the case is not the first examination: an earlier one has no masked contrast-enhanced T1. The other sequences and later examinations are the same patients: no later build should take them as other subjects.
  • Images. The provider's NIfTI byte for byte, reoriented and defaced, the face and what lies in front of it set to 0; 64-bit floats in arbitrary units, no mask given. Most test cases are 2D spin-echo slabs of thick axial slices across the internal auditory canals; VS_MC_RC2_026 and VS_MC_RC2_029 are 3D images of 192 slices of 0.9 mm.
  • Truth. An nnU-Net trained on the provider's earlier datasets proposed masks on 427 scans of 139 patients, and 221 were accepted by two neurosurgical fellows and a trained radiologist each. The trained radiologist accepted or corrected the retrained network's proposals for the others, and alone reviewed the 64 patients added later (105 scans). About half the masks are a network's output that three readers accepted.
  • Agreement. None is reported between readers here; on the provider's earlier dataset four annotators agreed at a mean Dice of 0.88 (VestibularSchwannomaMCRC's notes). The last network's median Dice of 0.967 on those 105 scans says how little the reader changed. Any nnU-Net of that family is close to the truth by construction, and the provider's networks have seen these patients.
  • Checks. On the 20 cases built first every mask holds 0 and 1 on its image's grid, in one piece, on an enhancing mass in the internal auditory canal or the cerebellopontine angle; VS_MC_RC2_199 is a cystic tumour outlined with its cysts. The 14 added later pass the grid check.
  • Room. The tumours are small and the slices thick: 0.31 to 3.62 ml on the test cases. Moved by one voxel in the plane, a test mask of the ten drawn first keeps a Dice of 0.86 to 0.97 (mean 0.93); moved by one slice, 0.40 to 0.85 (mean 0.56). The provider sets no surface tolerance: 1 mm.
  • Treatment. Of the ten test patients drawn first, VS_MC_RC2_121 had radiosurgery before its examination, and the examinations of VS_MC_RC2_006 and VS_MC_RC2_074 are those at which radiosurgery was decided.
  • Overlap. The collection page says it does not overlap with the collections of VestibularSchwannomaMCRC and VestibularSchwannoma; no identifier allows a patient check. By image, none of the 487 Vestibular-Schwannoma-MC-RC series on NBIA matches a case built first in scanner model, slice count and matrix, and no test case has the 120 or 160 slices of 512 x 512 of the Gamma Knife T1s.

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 case of Vestibular-Schwannoma-MC-RC2
The first test caseClick to look at every case.
Test cases
18 test cases drawn with seed 2026 by scanner maker among the 185 patients with a tumour mask: 13 Siemens, 3 GE and 2 whose scanner the collection does not give. Then 16 development cases, the last of the same seeded order, those the test draw rests on kept, each maker in proportion: 12 Siemens, 2 GE, 1 Philips and 1 not given. A case is a patient's first examination, by date, with a mask.
Truth
contours by a program, verified by people the dataset's mask of the tumour on the contrast-enhanced T1: proposed by an nnU-Net, accepted by two neurosurgical fellows and a trained radiologist each on their own, or proposed again by the retrained network and accepted or corrected in ITK-SNAP by the trained radiologist, the hardest margins set by a consultant neuroradiologist
Licence
CC BY 4.0
16 cases for tuning, never tested

VS_MC_RC2_071, VS_MC_RC2_157, VS_MC_RC2_092, VS_MC_RC2_027, VS_MC_RC2_148, VS_MC_RC2_199, VS_MC_RC2_068, VS_MC_RC2_136, VS_MC_RC2_018, VS_MC_RC2_077, VS_MC_RC2_210, VS_MC_RC2_007, VS_MC_RC2_115, VS_MC_RC2_196, VS_MC_RC2_064, VS_MC_RC2_141