fidoptCirrMRI600+, cirrhotic liver on T2-weighted MR
A T2-weighted abdominal MR of a patient with liver cirrhosis, from one centre in Istanbul, with the liver outlined (CirrMRI600+).
31 test casesabdomenimage MRresearch only
Read before comparing (6)
- Pool. CirrMRI600+'s 318 T2-weighted abdominal MRs, one a patient with liver cirrhosis, from one centre (ethics approval of the Istanbul Faculty of Medicine), 2015 to 2022, in the provider's split (256 training, 31 validation, 31 test). The 310 T1-weighted scans, 291 of the same patients, and the 55 healthy controls are left out.
- Images. The deposit's NIfTI files byte for byte, mostly single-shot (HASTE) axial slices 4.5 to 9 mm apart, 0.63 to 1.19 mm in plane on the test cases, values without unit; Philips 1.5 and 3 T and Siemens 1.5 T scanners. No mask is given.
- Truth. MRSegmentor's proposal refined by hand by four clinicians over several rounds until they agreed, then verified by them; they agreed at a kappa of 0.87 on T2. How much of MRSegmentor's output remains is not said: graded semi-automatic, and a method close to MRSegmentor may be favoured.
- Cases. The test livers hold 526 to 2,494 mL (median 1,380 mL).
- Start. Over the 31 test cases a whole-image mask scores a Dice of 0.078 at the median, an Otsu mask 0.008. TotalSegmentator 2.18.0 (total_mr) on test case 58 scores 0.891 for the liver (892 mL against 992 mL).
- Overlap. No other dataset of the corpus names CirrMRI600+, its hospital or the deposit; PancreasMR's centres are in the United States. LiverLesionsMR's archive names no sources; a grid comparison found no match but had no known pair to prove it, so overlap is unsettled. That set is mostly contrast-enhanced T1 with lesion masks, and CirrMRI600+ gives no lesion masks. A dataset built later from the T1 scans must keep these 61 patients on the same sides.
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
- The provider's T2 test split taken whole, 31 patients, and 30 development patients drawn with seed 2026 from its T2 training split. One T2-weighted scan a patient; one hospital, so no strata.
- Truth
- contours by a program, verified by people the liver: MRSegmentor's proposal refined by hand by four clinicians over several rounds until they agreed, then verified by them (Jha et al. 2025)
- Licence
- CC BY-NC 4.0
30 cases for tuning, never tested
442, 412, 458, 125, 295, 459, 174, 43, 401, 400, 344, 109, 255, 357, 154, 195, 100, 29, 284, 142, 182, 47, 184, 346, 369, 175, 423, 277, 127, 460