Liver fibrosis, arising from chronic viral or metabolic liver conditions, presents a significant global health challenge. Precise liver segmentation (LiSeg) and fibrosis staging (LiFS) are critical for enabling precise disease management, prognostication, and informed clinical decision-making. This challenge focuses on advancing two interconnected objectives to bridge clinical needs with AI innovation:
Automated Liver Segmentation.
Multi-Phase Fibrosis Staging.
Participants will develop robust AI solutions using multi-center, multi-phase MRI data, addressing real-world variability in imaging protocols and scanner systems.
Tasks
Task 1: Liver Fibrosis Staging (LiFS)
Develop models to stage fibrosis into four stages (S1-S4). leveraging cross-phase complementary information from dynamic MRI sequences. Two clinically critical binary subtasks are evaluated:
Cirrhosis Detection: S1–S3 vs. S4
Substantial Fibrosis Detection: S1 vs. S2–S4
Subtasks:
Non-Contrast Subtask: T1WI, T2WI, and DWI sequences only.
Contrast-Enhanced Subtask: All modalities allowed, including Gd-EOB-DTPA phases (GED1–GED4).
Task 2: Liver Segmentation (LiSeg)
Segment the liver in multi-phase fibrosis, where limited ground truth of Hepatobiliary phase (GED4) MRI is available. Non-constrast data (T2WI, T1, DWI) could be segmented via unsupervised or registration-based approaches to overcome annotation limitations.
Subtasks:
Non-Contrast Subtask: Segment liver anatomy in T2WI, T1, and DWI sequences. No annotations provided.
2) Dataset overview: The track cohort comprises 610 patients (170 newly cases compared to CARE2024) diagnosed with liver fibrosis, all of whom underwent multi-phase MRI scans. The dataset includes multi-phase and multi-center data, with images acquired from clinical centers using three different MRI scanner vendors. The dataset consists of T2-weighted imaging, diffusion-weighted imaging, and Gadolinium ethoxybenzyl diethylenetriamine pentaacetic acid (Gd-EOB-DTPA)-enhanced dynamic MRIs. The Gd-EOB-DTPA-enhanced dynamic MRIs cover the non-contrast phase (T1WI), arterial phase, venous phase, delayed phase, and hepatobiliary phase.
3) Contrast-enhanced dynamic scans: Contrast-enhanced scans were performed based on the injection of the GD-EOB-DTPA agent. The arterial phase is captured 25 seconds after the contrast agent is injected. Subsequently, the portal phase is achieved 1 minute later. After another 3 minutes, the delay phase is obtained, and finally, the hepatobiliary phase is reached 20 minutes thereafter.
4) Data format: The data are all in Nifty format. Each sample may randomly lack phases (except hepatobiliary phase), and the sequences have not applied pre-alignment through spatial registration.
Training set
Vendor
Center
#Cases
#Annotation for seg
A
A
130
10
B
B1
170
10
B
B2
60
10
Validation Set
Vendor
Center
#Cases
#Annotation for seg
A
A
20
20
B
B1
20
20
B
B2
20
20
Test Set
Vendor
Center
#Cases
A
A
40
B
B1
40
B
B2
40
C
C
70
NOTE: The evaluation on the unseen dataset (C) highlights the generalization capabilities of the models and algorithms, which we particularly emphasize.
Rules
Publicly available data (such as LLD-MMRI2023) and pre-trained models are allowed.
Dice Similarity Score (Dice), Hausdorff Distance (HD in mm)
Contrast-Enhanced (GED4)
Dice Similarity Score (Dice), Hausdorff Distance (HD in mm)
NOTE: Participants are welcome to participate in a single subtask, and their performance will still be recorded and displayed on the leaderboard.
LiFS Contrast-Enhanced (ID)
Team
ACC↑(S4 vs. S1–S3)
AUC↑(S4 vs. S1–S3)
ACC↑(S1 vs. S2–S4)
AUC↑(S1 vs. S2–S4)
BioDreamer
68.33
77.19
74.58
78.11
CitySJTU
70.83
75.10
75.00
78.45
NW-Radio
47.50
78.03
65.83
78.20
Sigma
75.83
83.92
77.50
83.42
Team space
71.67
79.39
79.17
80.24
WSQ
72.50
78.31
80.83
84.50
LiFS Contrast-Enhanced (OOD)
Team
ACC↑(S4 vs. S1–S3)
AUC↑(S4 vs. S1–S3)
ACC↑(S1 vs. S2–S4)
AUC↑(S1 vs. S2–S4)
BioDreamer
63.71
67.40
92.86
47.68
CitySJTU
41.43
46.81
64.29
52.62
NW-Radio
32.86
64.11
92.86
52.62
Sigma
58.57
54.12
92.86
75.38
Team space
52.86
60.22
64.29
64.62
WSQ
41.43
66.05
82.86
37.23
LiFS Non-Contrast (ID)
Team
ACC↑(S4 vs. S1–S3)
AUC↑(S4 vs. S1–S3)
ACC↑(S1 vs. S2–S4)
AUC↑(S1 vs. S2–S4)
BioDreamer
70.83
77.89
70.00
77.11
CitySJTU
71.67
78.61
70.00
73.70
NW-Radio
47.50
79.78
65.83
67.27
Sigma
70.00
77.22
65.00
74.28
Team space
74.17
81.33
76.67
83.93
TeamZhang
66.67
71.73
74.17
68.48
potato
72.50
77.22
71.67
78.51
LiFS Non-Contrast (OOD)
Team
ACC↑(S4 vs. S1–S3)
AUC↑(S4 vs. S1–S3)
ACC↑(S1 vs. S2–S4)
AUC↑(S1 vs. S2–S4)
BioDreamer
32.86
56.65
88.29
59.40
CitySJTU
42.86
49.31
70.00
31.08
NW-Radio
32.86
46.25
92.86
40.31
Sigma
71.43
69.47
92.86
86.31
Team space
64.29
52.73
88.57
61.23
TeamZhang
64.29
68.83
91.43
71.38
potato
70.00
71.51
67.14
33.23
LiSeg Contrast-Enhanced (ID)
Team
Dice↑
HD(mm)↓
Team space
0.9641
24.32
BI
0.9644
22.63
MIHL
0.9456
56.59
UON_IMA
0.9684
22.97
Wenting Yue
0.9631
24.12
xcj
0.9647
25.03
Sigma
0.8302
66.78
SuperIdols
0.9610
23.30
Sigma
0.9488
42.12
SUSTVI
0.9612
26.84
BIGS2
0.9619
24.50
Monster
0.9703
21.40
IntelliLiver
0.9630
25.75
BioDreamer
0.9516
29.07
LiSeg Contrast-Enhanced (OOD)
Team
Dice↑
HD(mm)↓
Team space
0.9753
15.72
BI
0.9754
14.37
MIHL
0.9726
21.52
UON_IMA
0.9754
15.11
Wenting Yue
0.9741
15.58
xcj
0.9745
15.84
SuperIdols
0.9770
14.65
Sigma
0.9586
20.65
SUSTVI
0.9652
29.02
BIGS2
0.9688
15.56
IntelliLiver
0.9716
16.41
Monster
0.9765
15.17
BioDreamer
0.9656
19.51
LiSeg Non-Contrast (ID)
Team
T1_Dice↑
T1_HD(mm)↓
T2_Dice↑
T2_HD(mm)↓
DWI_Dice↑
DWI_HD(mm)↓
BIGS2
0.9434
38.06
0.8722
35.24
0.8029
31.04
BioDreamer
0.9196
31.83
0.8201
37.16
0.7210
31.33
CitySJTU
0.9262
27.80
0.8890
34.60
0.8374
26.96
MIHL
0.8642
163.91
0.2024
121.48
0.2612
99.25
Sigma
0.9318
58.99
0.7568
75.44
0.8234
34.19
LiSeg Non-Contrast (OOD)
Team
T1_Dice↑
T1_HD(mm)↓
T2_Dice↑
T2_HD(mm)↓
DWI_Dice↑
DWI_HD(mm)↓
BIGS2
0.9548
22.11
0.9015
35.17
0.8481
22.58
BioDreamer
0.9418
28.24
0.8457
25.84
0.7812
15.96
CitySJTU
0.9444
25.54
0.8894
24.52
0.8857
11.21
MIHL
0.7351
102.79
0.0714
118.06
0.4567
51.76
Sigma
0.9503
29.04
0.6208
72.29
0.9041
19.72
Citations
Please cite these papers when you use the data for publications:
@article{liu2025merit,title={MERIT: Multi-view evidential learning for reliable and interpretable liver fibrosis staging},author={Liu, Yuanye and Gao, Zheyao and Shi, Nannan and Wu, Fuping and Shi, Yuxin and Chen, Qingchao and Zhuang, Xiahai},journal={Medical Image Analysis},volume={102},pages={103507},year={2025}}@inproceedings{gao2023reliable,title={A reliable and interpretable framework of multi-view learning for liver fibrosis staging},author={Gao, Zheyao and Liu, Yuanye and Wu, Fuping and Shi, Nannan and Shi, Yuxin and Zhuang, Xiahai},booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},pages={178--188},year={2023}}@article{wu2022meru,title={Minimizing Estimated Risks on Unlabeled Data: A New Formulation for Semi-Supervised Medical Image Segmentation},author={Wu, Fuping and Zhuang, Xiahai},journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},title={Minimizing Estimated Risks on Unlabeled Data: A New Formulation for Semi-Supervised Medical Image Segmentation},year={2023},volume={45},number={5},pages={6021-6036},}
