fidopt Interactive view
Vessels, segmentation

MSD Hepatic Vessel

Portal venous CT of the liver of a Memorial Sloan Kettering patient with a primary or metastatic liver tumour, with the liver's vessels and tumours segmented (Medical Segmentation Decathlon, Task08 HepaticVessel).

17 test casesabdomenimage CT
Look at the images
Read before comparing (8)
  • Truth. Label 1 is the liver's vessels: a seed drawn by hand, grown by a level set in the Scout application, then adjusted by hand by an expert abdominal radiologist. Label 2 is the tumours, whose making the data paper does not describe. One reader a scan; the challenge paper says the vessel annotations were found not optimal after the release. The liver is not labelled.
  • Vessels. They come in pieces (13 to 92 on the test cases, median 55) and follow the portal vein and the vena cava out of the liver: on hepaticvessel_016, 70% of the vessel label lies inside TotalSegmentator's liver, and all but 0.5% inside its liver, portal and splenic vein and inferior vena cava.
  • Tumours. 6.7 to 1007 ml on the test cases, hepaticvessel_209 a mass filling the right lobe, in one piece on every test case but hepaticvessel_057 (two). On hepaticvessel_016, 83% of the tumour lies inside TotalSegmentator's liver and 99% within about 5 voxels of it.
  • Images. The decathlon's NIfTI files as delivered: Hounsfield units, the values below -1024 set to -1024, the voxels transposed without resampling to the closest RAS orientation.
  • Left out. hepaticvessel_263 to hepaticvessel_460 are patients of TCIA's Colorectal-Liver-Metastases, also Memorial Sloan Kettering's, from which CRLMCT draws its pairs: hepaticvessel_N is CRLM-CT-(N+738). Leaving the range out keeps the two datasets apart by patient.
  • Overlap. 181 of the 190 scans in that range share their CRLM scan's slice count and pixel size, the twin of each of CRLMCT's 40 scans among them, and a slice of hepaticvessel_284 equals one of CRLM-CT-1022. The nine training scans outside the range with such a match hold no slice of it; no case built matches a corpus image by slice count and pixel size.
  • Room. On the challenge's hidden test set the 19 entries' mean Dice ran from 0.14 to 0.63 on the vessels (median 0.55; nnU-Net 0.63) and from 0.11 to 0.69 on the tumours (median 0.38; nnU-Net 0.69). The surface Dice is at 3 mm, the error the challenge's clinicians would accept.
  • Outside numbers. None compare: the challenge ranks its entries on its hidden test set. The test cases here are training cases, public with their labels, so a network trained on the decathlon's task has seen them.

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 MSD Hepatic Vessel
The first test caseClick to look at every case.
Test cases
17 test cases drawn with seed 2026 among the 163 training cases outside development, and 16 development cases, from a pool of 179. The 124 scans of Colorectal-Liver-Metastases are left out (prepare.py), and the labels of the decathlon's 140 test scans were never released.
Truth
contours by a program, verified by people the decathlon's two labels: 1 the vessels of the liver, seeded by hand and grown by a level set in the Scout application, then adjusted by hand by an expert abdominal radiologist; 2 the liver's tumours, whose making the data paper does not describe
Licence
CC BY-SA 4.0
16 cases for tuning, never tested

hepaticvessel_173, hepaticvessel_031, hepaticvessel_181, hepaticvessel_050, hepaticvessel_200, hepaticvessel_042, hepaticvessel_217, hepaticvessel_234, hepaticvessel_197, hepaticvessel_070, hepaticvessel_065, hepaticvessel_255, hepaticvessel_010, hepaticvessel_227, hepaticvessel_208, hepaticvessel_185