fidopt Interactive view
Abdomen MR, segmentation

TotalSegmentator liver segments MR

An abdominal MR of one patient, from the University Hospital Basel or the Imaging Data Commons, of many sequences and orientations, the eight Couinaud liver segments drawn by hand (TotalSegmentator's liver_segments_mr training set, version 2.0.0).

10 test casesabdomenimage MR
Look at the images
Read before comparing (9)
  • Source. The archive is the training set of TotalSegmentator's task liver_segments_mr, 115 cases under random ids; the first version held 120, and the record does not say which five went. The paper (Nicoli et al., J Imaging Inform Med 2025) took 120 Basel examinations, a deliberately heterogeneous set, mostly T1-weighted VIBE with Dixon, contrast agents at variable doses.
  • Origin. The archive does not keep to Basel. Of the 20 images built, 12 (6 test cases and 6 development cases) match scans that TotalSegmentator MRI's meta.csv gives as the Imaging Data Commons', on GE and Philips scanners; 7 are Basel's Siemens scans; JJwH2wjxoOMTmp59, not matched, has the size and spacing of a Basel scan.
  • Images. The 10 test cases are 7 axial slabs of 26 to 96 slices 2 to 8 mm apart and 3 coronal ones, with pixels of 0.66 to 1.88 mm; QqIPw7GVi8YiXjfX is tilted by 8 degrees. The values are as the provider wrote them, with no common scale (maxima 548 to 5950). No mask is given.
  • Truth. A junior physician supervised by a radiologist with 3 years of experience drew the segments voxel by voxel by Couinaud's classification, on the NORA platform. No agreement between readers is given for the segments. The record speaks of '9 couinaud liver segments'; the maps hold eight, segment 4 not split.
  • Coverage. A case is scored on the segments its map holds. On the test cases 9 maps hold all eight. VKOefmxCV65H4sJk lacks segments 2 and 3, though its map stays inside the image and the record gives no reason; on QqIPw7GVi8YiXjfX, r50A4gr80MS3vcBx, tqIyK0QqZiVxnASR and u9KfljEa4AEIZCqH the field of view cuts the liver. Development case RkfJ5Qyx0yA2Cx8s holds segments 2 and 3 only.
  • Checks. On the 20 built cases the segments lie where Couinaud's classification puts them, and the liver on the patient's right; the five test livers the field of view does not cut measure 1238 to 2228 mL.
  • Overlap. 21 of the 115 images are scans of TotalSegmentator MRI 3.0.0 that TotalSegMRI selects, found by size, affine and voxels within 0.006; prepare.py lists them. The 20 in TotalSegMRI's pairs stay out; the 21st, 9hkzufdr5cw30T4M (s0253), which no other dataset holds, is a test case. 1oqsYgz8VneMqTBF is TCGA-B8-5158's series AX OP, a patient whose MR and CT L2R_AbdomenMRCT holds.
  • Patients. The paper split its data by study, not by patient, and a small share of its patients had several scans; no patient identifier is given. A test and a development case, or a case and a TotalSegMRI pair, may be one patient. No two of the 115 images share a grid, so no case is another image of the same acquisition.
  • TotalSegmentator. Its task liver_segments_mr was trained on all 120 cases: a method that uses it, or its features, has seen this truth and is not a fair comparison. On test case 9hkzufdr5cw30T4M it gives a Dice of 0.960 to 0.976 a segment (mean 0.968). The paper's nnU-Net reached a mean Dice of 0.847 in five-fold cross-validation on the 120 examinations.

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 TotalSegmentator liver segments MR
The first test caseClick to look at every case.
Test cases
10 test cases drawn with seed 2026 among the 38 that hold segments, are not images of TotalSegMRI and are not of a patient of L2R_AbdomenMRCT, and 10 more of them for development. Of the 59 cases that hold segments, the 20 images of TotalSegMRI's pairs and 1oqsYgz8VneMqTBF are left out.
Truth
contours by experts the eight Couinaud segments of the liver, drawn voxel by voxel by a junior physician under the supervision of a radiologist with 3 years of experience (the paper)
Licence
CC BY 4.0
10 cases for tuning, never tested

OAW7sGirT9OjiRyT, RkfJ5Qyx0yA2Cx8s, rj2wpmGm1b0oHr9Y, 754qsTkWT9ByD3UR, bx1BXmAJZZl6Mxek, If5Yua91GRtsUbU2, kbdGS2laG0N4H0q6, fo2GfcEtcV18VDSC, fGFJmCNTgP3PbPNk, FjaNhQ2Vgr3CT4dP