fidopt Interactive view
Lesions and tumours, segmentation

MCT-LTDiag

The portal venous CT of a patient with a liver tumour of one of five kinds (hepatocellular carcinoma, intrahepatic cholangiocarcinoma, colorectal or breast cancer metastases, haemangioma), every liver tumour segmented on it (MCT-LTDiag, Peking Union Medical College Hospital).

30 test casesliverimage CT
Look at the images
Read before comparing (7)
  • Pool. 517 patients of Peking Union Medical College Hospital with a pathological or clinical diagnosis of one of the five kinds (115 breast cancer metastases, 103 hepatocellular carcinomas, 103 colorectal metastases, 100 cholangiocarcinomas, 96 haemangiomas), and a four-phase CT from 2014 on; poor images, unclear tumour margins and tumours under 1 cm were excluded by the provider.
  • Truth. A U-Net trained on LiTS drafted the tumours on the portal venous phase and one senior radiologist corrected every mask; no second reader and no agreement are reported. The provider's selection leaves tumours under 1 cm out, and the paper does not say whether a smaller lesion beside a kept one is drawn: a method finding one is charged for it. Graded semi-automatic.
  • Cases. The test tumours hold 1.8 to 2,763 mL (median 50 mL). Slices are 5 mm, pixels 0.56 to 0.98 mm; the masks follow the slices, so along the body axis a surface is known to 5 mm.
  • Images. The deposit's NIFTI/pvp.nii.gz as delivered, 16-bit Hounsfield units; the other phases, registered by the provider onto it, are left out (MCTLTDiag_Syn uses the non-contrast one), and so is the network's liver mask.
  • Overlap. No other dataset of the corpus holds these CTs, and no descriptor names this hospital or deposit. MCTLTDiag_Syn draws its patients from the same deposit with these 60 held out, so no patient is an input in one dataset and a target in the other.
  • Start. A mask of the whole image scores a Dice of 0.003 at the median of the test CTs (0.071 at most), an Otsu mask 0.008 (0.0001 to 0.17).
  • Outside numbers. None: the deposit has no challenge and no split; its paper scores classification, not segmentation.

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

Test cases
30 test CTs drawn with seed 2026 by tumour kind among the 487 patients outside development, out of the deposit's 517, then 30 development CTs, the last of the same seeded order, the kinds in proportion. One CT a patient. The deposit has no test set; MCTLTDiag_Syn's 60 patients are other patients of the deposit.
Truth
contours by a program, verified by people every liver tumour of the deposit's mask_pvp.nii.gz, under one label: a U-Net trained on LiTS drew them on the portal venous phase and a senior radiologist (over 10 years of abdominal imaging) reviewed and corrected every mask in ITK-SNAP (Sci Data 2025)
Licence
CC0 1.0
30 cases for tuning, never tested

230408a9, 231109b16, 240722e87, 240122d62, 231025c04, 230218a7, 240504b08, 240722e78, 240112d24, 231025c13, 240504a12, 240504b09, 240722e61, 230906d14, 230218c6, 240620a24, 230816b03, 240722e74, 230906d07, 240229c12, 230312a4, 240504b03, 240722e79, 240122d45, 231025c21, 240620a25, 240504b42, 240722e101, 240112d23, 240620a12