fidopt Interactive view
Lesions and tumours, segmentation

Lausanne TOF-MRA aneurysms

Time-of-flight MR angiography of the head of a patient with unruptured intracranial aneurysms, each traced voxel by voxel by a radiologist and checked by a senior neuroradiologist (Lausanne TOF-MRA Aneurysm Cohort, OpenNeuro ds003949).

10 test casesbrainimage MR
Look at the images
Read before comparing (8)
  • Pool. 284 subjects of Lausanne University Hospital: 157 patients with 198 unruptured aneurysms and 127 controls. The controls and sub-482, whose session holds no mask, are left out, since an empty truth scores nothing. So is sub-450: its TOF holds signal in 18 of its 200 slices only, none where its aneurysm lies.
  • Masks. For 119 patients, sub-013 to sub-441, the masks are spheres sized by hand around each aneurysm: a Dice against a sphere measures the sphere, so they are left out. For the 37 numbered sub-450 to sub-492, consecutive and unselected, each aneurysm is traced voxel by voxel. The 152 spheres match a ball of their own volume at a Dice of 0.93 or more, the 46 tracings at 0.61 to 0.90.
  • Truth. One radiologist with two years of experience in neuroimaging traced each aneurysm axially, slice by slice, in ITK-SNAP; a senior neuroradiologist with more than fifteen years double-checked for false positives and negatives. No agreement between readers is published: an outline is one reader's, and ends flat at its first and last slices. TotalSegmentator makes no aneurysm, so no network check was run.
  • Scanners. The 36 are 10 at 1.5 T, all on the Siemens Aera, and 26 at 3 T, on the Siemens Prisma (11), Skyra (10), TrioTim (3) and Verio (2). The development cases are nine at 3 T and one at 1.5 T. A case is the TOF-MRA slab as delivered, 84 to 133 mm around the circle of Willis, in 32-bit floats without unit; no mask is given.
  • Thrombus. On 18 of the 20 cases built the mask lies on flowing blood. The two large ones, test case sub-472 (7.6 ml) and development case sub-478 (27 ml), are mostly of intermediate signal with a bright lumen at one side: the whole sac is drawn, part of it most likely thrombosed, and a method that segments only flowing blood misses most of it.
  • Size. The ten test cases hold 12 aneurysms, two each on sub-458 and sub-487, of 0.010 to 7.6 ml (median 0.04 ml). On a few hundred voxels the Dice turns on a voxel: a shift of one voxel leaves a Dice of 0.68 to 0.94 (mean 0.83), which the surface Dice at the default 1 mm, two to four voxels in plane, does not see.
  • Outside numbers. None: only detection rates are published on this cohort, no segmentation score. The cases are public with their masks, and the authors' released weights, like any network trained on this cohort, have seen the test cases.
  • Overlap. No other dataset of the corpus holds scans of this cohort; TopCoW and TopBrain come from Zurich.

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 Lausanne TOF-MRA aneurysms
The first test caseClick to look at every case.
Test cases
10 test cases drawn with seed 2026 among the 36 patients whose aneurysms were traced voxel by voxel, the 10 development cases left out, in proportion to field strength: 3 at 1.5 T, 7 at 3 T. A case is a patient's one session. The controls, the patients with spherical masks, sub-482 and sub-450 are left out.
Truth
contours by experts the dataset's manual masks of the aneurysms, traced voxel by voxel, slice by slice in the axial plane in ITK-SNAP, by a radiologist with two years of experience in neuroimaging, and double-checked by a senior neuroradiologist with more than fifteen years of experience to exclude false positives and false negatives; one file an aneurysm, merged into one label
Licence
CC0
10 cases for tuning, never tested

sub-457, sub-489, sub-455, sub-485, sub-478, sub-470, sub-480, sub-453, sub-491, sub-484